From big data to smart energy services: An application for intelligent energy management

Future Generation Computer Systems (

)

–

Contents lists available at ScienceDirect

Future Generation Computer Systems journal homepage: www.elsevier.com/locate/fgcs

From big data to smart energy services: An application for intelligent energy management Vangelis Marinakis a, *, Haris Doukas a , John Tsapelas a , Spyros Mouzakitis a , Álvaro Sicilia b , Leandro Madrazo b , Sgouris Sgouridis c a b c

Decision Support Systems Laboratory, School of Electrical & Computer Engineering, National Technical University of Athens, Athens, Greece ARC Enginyeria I Arquitectura La Salle, Universitat Ramon Llull, Barcelona, Spain Khalifa University, Masdar Institute, United Arab Emirates

highlights • • • • •

High level architecture of a big data platform for smart energy services. Integration of heterogeneous types of data from multiple domains and sources. A ‘‘data-driven’’ DSS for exploiting multidisciplinary data within a smart city context. Support energy managers for managing their building facilities’ energy performance. Pilot application and validation of the developed DSS system in three European cities.

article

info

Article history: Received 21 August 2017 Received in revised form 14 March 2018 Accepted 21 April 2018 Available online xxxx Keywords: Big data Decision support system Energy services Intelligent management Smart Cities

a b s t r a c t Big data is an ascendant technological concepts and includes smart energy services, such as intelligent energy management, energy consumption prediction and exploitation of Internet of Things (IoT) solutions. As a result, big data technologies will have a significant impact in the energy sector. This paper proposes a high level architecture of a big data platform that can support the creation, development, maintenance and exploitation of smart energy services through the utilisation of cross-domain data. The proposed platform enables the simplification of the procedure followed for the information gathering by multiple sources, turning into actionable recommendations and meaningful operational insights for city authorities and local administrations, energy managers and consultants, energy service companies, utilities and energy providers. A web-based Decision Support System (DSS) has been developed according to the proposed architecture, exploiting multi-sourced data within a smart city context towards the creation of energy management action plans. The pilot application of the developed DSS in three European cities is presented and discussed. This ‘‘data-driven’’ DSS can support energy managers and city authorities for managing their building facilities’ energy performance. © 2018 Elsevier B.V. All rights reserved.

1. Introduction The increasing momentum of big data applications constitutes a significant opportunity for the energy sector in the field of energy management, environmental protection, and energy conservation [1]. In recent years, large amounts of energy consumption and production data are being generated and the energy systems are being digitised, with the increasing penetration of emerging Information and Communication Technologies (ICTs). ICTs, Internet of Things (IoT) and data analytics constitute a catalyst towards conceptualising and generating new energy and climate solutions. author. * Corresponding E-mail address: [email protected] (V. Marinakis).

A massive amount of data is generated from a wide spectrum of energy-related sources. Energy big data not only include the massive smart meter reading data, but also the huge amount of related data from other sources, such as weather and climate data, asset management data, Energy Performance Certificates (EP Certificates), building stock auditing, Sustainable Energy and Climate Action Plans (SECAPs), socioeconomic data, energy end-user’s characteristics and comfort levels, etc. Organisations can extract value from this volume and variety of data, by enabling highvelocity capture mechanisms coupled with effective data analytics [2]. Typically, data are not only traditional structured relational data, but also semi-structured data (e.g. weather and Web services data), as well as unstructured data (e.g. customer behaviour data,

https://doi.org/10.1016/j.future.2018.04.062 0167-739X/© 2018 Elsevier B.V. All rights reserved.

Please cite this article in press as: V. Marinakis, et al., From big data to smart energy services: An application for intelligent energy management, Future Generation Computer Systems (2018), https://doi.org/10.1016/j.future.2018.04.062.

2

V. Marinakis et al. / Future Generation Computer Systems (

information contained in documents in formats such as .pdfs or .html, etc.) [3,4]. The lack of a commonly accepted semantics framework makes data integration and exploitation an even harder to tackle problem. Integrating these data in multiple formats to create large volumes of data which can be processed in real time is a first challenge to be addressed. However, the challenge is not constrained to the data acquisition and storage, but it concerns also the analysis and processing of big data. The speed of data analysis and processing should be optimum given the massive volume of data; there are many streaming data and relatively large volumes data changing at a given time. In this respect, data analytics processes that need hours to run are not efficient enough. Combining data from different domains (energy, environment, economics and financing, user behaviour and societal trends, and many more) can open a wide spectrum of innovative ICT solutions. Indeed, dealing with energy big data requires new technologies to efficiently process large quantities of data. As a result, there is the need for a cross-sectorial and cross-disciplinary data integration approach that will foster innovative energy services. Therefore, ICT is the ‘main integrator’ of infrastructure and management solutions and ICT-enabled technologies are rapidly gaining ground to manage energy systems, allowing households, neighbourhoods, towns and cities to share energy production and consumption capacity, generate local energy and become selfsufficient. Through the use of ICT-based management systems, the building’s intelligent systems can communicate with the energy providers to control cooling, heating, lighting and hot water systems, to ensure a more stable energy supply, optimising the management of demand peaks and resources at a local level. In particular, the application of ICT related technology in buildings is closely linked to the implementation of smart meter technologies, providing opportunity for consumer feedback on energy consumption. Energy Management Systems (EMSs) are increasing their diffusion and are anticipated to witness the fastest growth over the next years, also with the integration of big data and data analytics in the existing EMS modules. According to a report published by Grand View Research, the global EMS market is expected to reach $58.6 billion by 2022, registering a Compound Annual Growth Rate (CAGR) of 14.3% over the period 2014–2020. In particular, the European market for EMS is expected to grow at a CAGR of 23.58% to reach $11.89 billion by 2019 from the current estimates of $4.13 billion in 2014 [5]. In particular, the adoption of Building Energy Management Systems (BEMS) is expected to increase rapidly due to its varied application in the Healthcare, Telecom and IT sectors, educational institutions, universities and commercial buildings like shopping malls, offices etc. [6]. A white paper by Navigant Research calculates the BEMS revenues in 2015 in $2.4 billion and estimates the market to grow up to $10.4 billion by 2024 [7]. Europe is also believed to be the most promising market, followed by North America and Asia Pacific. The connectivity and data flow associated with the IoT will also support the broader adoption of BEMS [8]. Rich sources of data enable better ways to manage and measure buildings and portfolios of buildings [9]. Considering that 90 percent of all data generated by devices like smartphones, connected meters and appliances is never analysed or acted upon, the ability to integrate intelligent devices for data analytics will be key to gain competitive advantage and meet the needs of an ever-growing audience of energy management stakeholders. This paper proposes a high level architecture of a big data platform that can support the creation, development, maintenance and exploitation of smart energy services through the utilisation of cross-domain data. This blueprint for a smart energy platform enables the integration of heterogeneous types of data

)

–

(open data, sensor/IoT, historical data, etc.) from multiple domains and sources, including among others data related to energy consumption/production, EP Certificates, SECAPs, European Union (EU) Building Stock Observatory, locational information, weather conditions, environment, financing/prices, end-users’ behaviour and demographics. The proposed platform can simplify the complexity of the information gathered by those sources, turning into actionable recommendations and meaningful operational insights for city authorities and local administrations, energy managers and consultants, utilities and energy providers, Energy Service Companies (ESCOs) and other energy companies, construction companies, building components manufacturers, investors and financiers and other organisations and third parties. A Decision Support System (DSS) has been developed, in line with these directions, aiming effectively utilise multi-sourced data within a smart city context towards the creation of energy management action plans. The pilot application and validation of the developed DSS system in three European cities is presented and discussed: (a) Sant Cugat Town Hall and Theatre, in Spain; (b) Savona Campus and Colombo-Pertini School, in Italy; (c) Zaanstad Town Hall, in the Netherlands. This ‘‘data-driven’’ DSS can, therefore, support energy managers and city authorities for managing their building facilities’ energy performance. The remainder of the paper is structured along six sections. The progress beyond the state-of-the-art of energy data analytic platforms, big data cleaning, processing and analytics, as well as big data exploration and visualisation is provided in Section 2. Section 3 presents the potential for integrating big data in the energy sector. Section 4 is devoted to the presentation of the internal architecture of the big data platform for smart energy services. Section 5 provides a description of the propose DSS. Section 6 presents the real life application of the ‘‘data-driven’’ system in the selected cities. Finally, the last section summarises the key issues that have arisen in this paper. 2. Progress beyond the state-of-the-art 2.1. Energy data analytic platforms In the last couple of years, several energy data repositories and energy data analytics platforms, with various levels of openness, data updates, maintenance, automation and extensibility. Energydata.info [10] is an open data platform providing access to datasets and data analytics that are relevant to the energy sector. Energy DataBus [11], developed by the US Energy Department’s National Renewable Energy Laboratory, is used for tracking and analysing energy consumption data in structures such as smart building and campuses, and achieved scale by building on top of the Cassandra [12] big data storage backend. KNIME [13] is a Swiss start-up providing an open source big data analytics platform with extensive support for energy data as a major use case [14]. Finally, Energywise [15] is an energy analytics tool developed by Agentis Energy to make commercial buildings more energy efficient by profiling energy consumption and giving building managers insights on how to make their buildings more energy efficient. The above-mentioned open source platforms are either experimental, outdated or not actively supported, while proprietary energy/climate data management and analytics software is rather expensive for many stakeholders like smaller utility companies and energy consumers. This paper strives towards the elaboration of a free, easily integrated, architecturally open alternative that will enable high-level energy/climate data management, analytics and monitoring applications. Moreover, the volume and variation of energy/climate data, especially sensor measurements, means that while many solutions might operate ‘well enough’ in isolated cases or for a limited period of time, utilising big data technologies is

Please cite this article in press as: V. Marinakis, et al., From big data to smart energy services: An application for intelligent energy management, Future Generation Computer Systems (2018), https://doi.org/10.1016/j.future.2018.04.062.

V. Marinakis et al. / Future Generation Computer Systems (

the only viable way to maintain and operate on a more centralised level, which is required for effective decision and policy making. The proposed architecture is built upon such big data technologies instead of including them to an outdated architecture as an afterthought. It brings together the latest advances in big data storage and analytics technologies with state-of-art energy/climate analytics models and apply data-specific optimisations to achieve performance and scale. 2.2. Big data cleaning, processing and analytics Data storage and management for big data is a flourishing field which has seen large advancements in the latest years. Hadoop [16] has become synonymous to big data, as it has been able to penetrate into large businesses, but truly harnessing its power requires commitment from the developer’s perspective. Cloudera [17] is a cloud-based service to help businesses integrate and manage Hadoop with some extent of data security built in, while MongoDB [18] is a modern alternative to relational databases which can, in certain scenarios, reduce operational overhead up to 95% [19]. Data cleaning can also be an important part of the business workflow, especially in a domain like maritime activities where sensor data can occasionally provide false or defective information. OpenRefine [20] is an open source tool, developed by Google, to explore and clean huge, even partially unstructured, datasets. DataCleaner [21] transforms semi-structured data sets into readable data to improve data visualisation, with support for data warehousing and data management services. Data analytics tools have always been valuable assets to businesses and organisations, and big data analytics are also seeing growing interest and business market shares. BigML [22] is a machine learning service with an easy-to-use interface to get predictions, also offering models for predictive analytics, while private virtual cloud is offered for enterprises. Qubole [23] simplifies big data analytics as it integrates directly data storage services like AWS, Google and Azure. Querying data is also made easy, as support for big data querying tools like Hive and Spark is built in. Both tools are enterprise solutions, but that require little technical background from their users. Lumify [24] is an open source alternative built on top of Hadoop and Accumulo focusing on big data analytics and visualisations, but it has significantly smaller options and analytics algorithms than its commercial counterparts. Finally, Talend [25] is another open source tool suite that offers, between others, the ability to easily run analytics using Hadoop technologies like HDFS, HBase, Hive, Pig, and Sqoop. Although it seems very promising, it is still in beta and it lacks any support for cross-origin data. Currently, energy/climate related data are plagued by the existence of various competing, incompatible formats with unclear semantics, especially when taking under consideration big datasets and streaming data. The proposed big data platform will enrich data with semantic information, thus vastly improving interoperability between different data sources and management of the aggregated data. Existing data cleaning tools can handle big data but are not specialised to the energy/climate domain. The proposed platform will exploit unique characteristics of energy/climate data (such as predictability of sensor intervals, well defined value ranges etc.) in order to improve the data cleaning process. In addition, big data analytic tools and algorithms are becoming extremely powerful, efficient and effective. However, a large barrier-to-entry for non-technical and even technical users, as well as little to no specialisation for different domains, limits their actual impact. We will use powerful analytics tools, alongside with its novelties in energy/climate data semantics to provide a simple interface while minimising overhead in resources and manpower required to utilise big data analytics.

)

–

3

2.3. Big data exploration and visualisation Visualising big data is often considered one of the most prominent difficulties associated with them, as it is seen as fundamentally hard to create a perceivable visual graph of amounts of data inconceivable to human cognitive. Big data visualisations tools have developed at least some insight of the visualised data, in order to propose the most appropriate data metrics that it is meaningful to represent. Zoomdata [26] is a commercial product that provides ‘‘simple, intuitive and collaborative’’ interactive big data visualisations, and it has been integrated with most big data storage solutions. Tableu [27] is another commercial, mature big data visualisation tool offered both as a standalone as well as an online, cloud-based service. Finally, DataWatch [28] provides real time visualisations of streaming data, and it is used by companies, such as IBM and NASDAQ. Most open source products for big data visualisation are clearly inferior to their proprietary counterparts, making it rather cost inefficient for many stakeholders, such as energy consumers, both individuals as well as SMEs, to acquire access to high-quality big energy/climate data visualisations. Developing high quality tools, we will enable stakeholders to consume and create such visualisations. Typically, energy/climate data management tools use standard techniques to visualise original data as well as their results. However, it is becoming clear that novel ideas and approaches achieve significantly better results in visualising data from various energy sub-sectors. The proposed big data platform will explore, develop and test such novel visualisation techniques, aiming to achieve broader impact on stakeholders. 3. Big data platform for smart energy services 3.1. The methodological approach The goal of the proposed big data platform is to unlock the promise of big data to face today’s energy challenges and support sustainable energy transformation towards a low carbon society. New ways to gather and integrate the data and a blend of more effective data analytics are proposed to provide valuable information and new insights to stakeholders for innovative sustainable business services. Through a blueprint for a smart energy platform, we bring together a fragmented community of data, services and stakeholders to give rise to a unified big data repository for energy/climate research and business activities. Gathering, transforming, storing, combining and analysing huge amounts of varied information (TBs of data), several beneficiaries are expected to significantly benefit from the analytics, insights and knowledge offered by the platform.

• Energy Providers and Utility Companies: As the energy sector is becoming more and more competitive, energy providers and utilities need to offer attractive, intelligent energy services. A rapid feedback on the value of innovative pricing programs based upon customer usage patterns can be provided, exploiting smart meter and sensor technologies, as well as weather information, demographic data and occupants’ feedback. User-centred applications that leverage data from smart meters into value for the buildings’ occupants can change consumers’ behavioural energy efficiency and enable intelligent home automation services. Energy providers and utility companies can optimise their data collection processes, reducing costs and thus offering a real value from improved revenue recovery.

Please cite this article in press as: V. Marinakis, et al., From big data to smart energy services: An application for intelligent energy management, Future Generation Computer Systems (2018), https://doi.org/10.1016/j.future.2018.04.062.

4

V. Marinakis et al. / Future Generation Computer Systems (

)

–

• ESCOs, Building Components Manufacturers, Construction

• Volume: The benefit from gathering, processing, and

Companies, etc.: The Covenant of Mayors for Climate and Energy constitutes a first-of-its-kind global initiative of cities and towns committed transparent climate action. A number of SECAPs have been submitted in which the Covenant signatories outline how they intend to reach their longterm CO2 reduction target [29,30]. Moreover, EU countries are obliged to collect EP Certificates, using different energy assessment tools and procedures (Directive 2010/31/EU). However, the available data are stored in disperse databases and in multiple formats. In order to make analyses of SECAP and EP Certificates data cutting across countries and domains, a unique representation of their content is required. Integrating SECAP and EP Certificates with other data sources (socioeconomic indicators, regulations, etc.) advanced energy services can be provided to specific beneficiaries, such as ESCOs and other energy and construction companies, private owners, etc. For instance, manufacturers of insulation materials might be interested to know in which city district and/ or sector they could market their products. • Investors and Financiers: Although several energy efficiency initiatives can be adopted to reduce the energy use in the long term, the scarcity of these types of investments prevents a greater allocation of resources by investors/financiers, such as commercial/green investment banks, institutional investors and their financial advisors, etc. The heterogeneous nature of energy efficiency projects generally make scaling up of investments and underwriting processes a complex exercise. The exploitation of data for investment projects is a key enabler to build the necessary market confidence. The availability of comparable, multilingual, anonymised historical data, pooled from the major buildings and industry market segments can encourage a greater energy efficiency investment flow. • Local Authorities and Other Organisations: More than other stakeholders, the aforementioned have to address climate challenges in a long-term plan, and become a driving force for adoption of relative initiatives towards sustainable energy and climate planning, and promotion of environmentally friendly practices. This multidisciplinary scientific field combines the domains of environment, sociology and economy. Collecting and interrelating asymmetric data related to the organisations’ operation and performance, their carbon footprint can be evaluated. Moreover, an exhaustive list of ‘‘green’’ energy interventions can be provided, supporting local authorities and other organisations to develop their action plans.

analysing large volumes of data streams, existing open data sources and other relative public repositories generates a number of challenges in obtaining valuable knowledge for industrial and research community. • Velocity: Energy-related data to be fed into the platform are both real-time (e.g. smart meters, sensors, weather, energy costs, users’ behaviour) and static data (e.g. EP Certificates, SECAPs, demographics, cadastre, etc.). The challenge for this is to develop different workflows that handle each type of data and perform similar required pre-processing action to both of them. Moreover, the incorporation of streaming data into analytical services along with static require special treatment and also the combination of different streaming endpoints creates the need for more complex data pipelines, since the stream rate may vary from hourly data to data generated every second. In addition, the stream of data acquired via multiple sources needs to be processed at a reasonable speed, which means that low-latency is mandatory and this is challenging in terms both data retrieval and processing, when large volume of information is present. • Veracity: Particular emphasis is laid on the diversity of quality, accuracy and trustworthiness of the data. Initially data that cannot be interconnected (i.e. datasets or existing RDF repositories that use different or non-well-established vocabularies, partially or incorrectly annotated) are brought into the storage infrastructure of the platform, and any sensitive data (e.g. social data) are anonymised to mitigate privacy risks. These data go into a curation process into the platform and a semantic enhancement that extends knowledge on existing resources. Then these data resources are linked back any existing repositories to respect the linkeddata principles. • Value: The available data can be queried through the platform, exposed through useful visualisations and real-time APIs, as well as provide useful analytics interfaces.

The proposed big data platform addresses the 5 big data challenges, as follows:

• Variety: Data of different formats, types and structures (structured, semi-structured and unstructured data) are processed using efficient methods and tools:

◦ Real-time data generated by IoT technologies (e.g. en-

◦ ◦ ◦ ◦

ergy consumption and energy production using smart meters, sensor-based data). Historical data for model development and pattern recognition (e.g. old weather data, energy costs). Open data from various sources (weather, climate, EP Certificates, SECAPs, costs). Secondary data that are not related directly to energy and climate (e.g. demographics, economics, cadastre). Anonymised social data (e.g. end-users’ behaviour).

3.2. High level architecture The different structural components of the big data platform are the following (Fig. 1):

• A Data Interoperability and Semantification Layer, The layer at the base of the overall system is the single-entry point for feeding all the multiple-sourced data, both open and business, into the platform. This layer should be able to receive and handle data of different types, from rarely changed, static building information, to near real time streaming sensor data or historical data, in various formats. This by itself is a major challenge and the selection of the appropriate tools heavily depends on the specific application needs. Receiving large datasets of historical data requires a batch execution workflow, able to handle structured data coming from relational databases and also semi-structured data in files like CSV, Parquet,1 ORC,2 Avro3 and NetCDF.4 Since the size of these files can make the standard file uploading workflows inefficient, the platform should be able to communicate with either distributed file systems, such as Hadoop Distributed File System5 (HDFS), or cloudbased storage systems, such as Amazon’s S3,6 Microsoft’s 1 https://parquet.apache.org/. 2 https://orc.apache.org/. 3 https://avro.apache.org/. 4 https://www.unidata.ucar.edu/software/netcdf/. 5 https://hadoop.apache.org/docs/r1.2.1/hdfs_design.html. 6 https://aws.amazon.com/s3/.

Please cite this article in press as: V. Marinakis, et al., From big data to smart energy services: An application for intelligent energy management, Future Generation Computer Systems (2018), https://doi.org/10.1016/j.future.2018.04.062.

V. Marinakis et al. / Future Generation Computer Systems (

Azure Blob Storage7 and Google’s Cloud Storage.8 As regards data coming from multiple streaming endpoints, a second workflow for their integration should be created and several technological solutions are available, each one having different characteristics. Apache Spark Streaming9 provides in-memory stream processing in micro-batches of data through an application programming model. For lowlatency, in-memory processing on an already deployed cluster Apache Flink10 is ideal and for complex data pipelines Apache Storm11 and Apache Kafka12 offer great capabilities. Apart from the ability to collect data, there is a major need for data pre-process, cleansing and validation as an initial step before the permanent storage. The received data, especially public open data, is probable to suffer from inconsistencies, missing values, be outdated, or repeated multiple times. Moreover, streaming data can also include inconsistent values when for example a sensor malfunctions. Thus, in order to ensure the quality and therefore the value of the data to be finally stored, an appropriate pre-processing pipeline should be defined, which should able to check the completeness, the consistency and the validity of the incoming data. Except for these inconsistencies, for performance reasons and in order to succeed the desired combination and interlinking of datasets from multiple sources, the data need to have a common ground and that may require a harmonisation, in terms of using common formats, same units of measurements, or creating pre-joined files for fast querying. In addition, depending on the data format, a denormalization process may also be required for structures like the relational databases or NetCDF files, so as to transform them in a flat schema, more suitable for fast queries and analytics. For the above purposes, suitable tools exist, i.e. OpenRefine,13 Trifacta,14 Spark Optimus15 or the Apache Spark transformation functions16 , while custom MapReduce processes on top of Hadoop can also be developed for more specific needs. The final step, is the data semantification, where semantics and metadata are added to the original data, where possible. This can be performed using well-established vocabularies and schemata related to the domains of energy, weather and climate, sensors network, buildings and more, along with input from knowledge experts. Since, the process of semantic annotation would not be efficient when applied in very large datasets, it is proposed that the annotation should be made on the columns of each dataset from the data provider himself, using the API provided by the Linked Data Vocabularies17 (LOV), store the selection in a triple-store and then use the selected annotation automatically when the dataset is extended. Furthermore, metadata can also be used to better describe the datasets, using the Data Catalogue Vocabulary18 (DCAT) so as to enhance interoperability between datasets and assist the selection of the suitable 7 https://docs.microsoft.com/en-us/azure/storage/blobs/storage-blobsintroduction. 8 https://cloud.google.com/storage/. 9 https://spark.apache.org/streaming/. 10 https://flink.apache.org/. 11 http://storm.apache.org/. 12 https://kafka.apache.org/. 13 http://openrefine.org/. 14 https://www.trifacta.com/. 15 https://hioptimus.com/. 16 https://spark.apache.org/docs/latest/ml-features.html#feature-transformers. 17 http://lov.okfn.org/. 18 https://www.w3.org/TR/vocab-dcat/.

)

–

5

datasets to be combined and retrieved for each of the performed query. The dataset’s metadata should also include annotation of the policy and licence type that follows each of them and declare the owner and the access–usage rights for the different users. The semantic enrichment and normalisation under the same vocabularies/terminology is going to enable the integration and cross-examination of data originated from different sources and provide significant added-value on the original data, while the main challenge is to bring together Linked Data and Big Data technologies, in order to perform efficiently more complex queries. • The Data Storage Cluster Utilises industry-proven distributed storage solutions to provide data storage for the incoming data, the data generated from the services and the semantic information and metadata. A NoSQL cluster can be the basic storage of the original data, able to handle both structured and unstructured types data and different formats, such as key–value pairs or text documents. Being suitable for big data and real-time applications, it enables the implementation of big data analytics and operations over large datasets effectively, with the desired speed, scalability and fault-tolerance. The selection of the storage solutions again significantly depends on the application. For high availability, distributed storage Amazon Dynamo19 can be used, or Cassandra,20 especially when similar queries are often performed. When high availability can be sacrificed for consistency, HBase21 can be the case, and for the storage of mainly unstructured data and multimedia MongoDB22 is preferred. For applications that require fast search over text documents or faceted search a full-indexed search engine should be used, like Elasticsearch23 or Apache Solr.24 Nevertheless, the architectural design may not be limited to choosing only one solution, but use two different technologies and dispatch the query execution properly, depending on its type. For example, HBase could be used for executing general purpose queries over files in an HDFS, while some specific, frequently used queries could be dispatched to Cassandra for better performance. A Graph Database cluster is also used, as metadata repository, to store the additional semantics and metadata of the enriched datasets, alongside the original data. This is a vast field of applied research and thus numerous solutions exist to support the tones of triplified semantic data, like Amazon Neptune,25 Virtuoso,26 Apache Jena,27 Neo4j,28 GraphDB29 and more. • The Data Access Policy Control is responsible to isolate data from different providers and grant access to other tools based on data ownership and advanced user permissions. It makes use of the annotations provided from the Data Interoperability and Semantification Layer about the data access policies and licences, in order to quickly resolve conflicts and decide who has rights to view or use the specific data. It is a very important component when multiple datasets 19 https://aws.amazon.com/dynamodb/. 20 http://cassandra.apache.org/. 21 https://hbase.apache.org/. 22 https://www.mongodb.com/. 23 https://www.elastic.co/. 24 http://lucene.apache.org/solr/. 25 https://aws.amazon.com/neptune/. 26 https://virtuoso.openlinksw.com. 27 https://jena.apache.org/. 28 https://neo4j.com/. 29 http://graphdb.ontotext.com/.

Please cite this article in press as: V. Marinakis, et al., From big data to smart energy services: An application for intelligent energy management, Future Generation Computer Systems (2018), https://doi.org/10.1016/j.future.2018.04.062.

6

V. Marinakis et al. / Future Generation Computer Systems (

from different providers are collected and should be combined. Two main methodologies are distinguished, Rolebased Access Control (RBAC) and Attribute-based Access Control (ABAC), which are both widely used. The first is based on assigning permissions to roles and roles to users, so only when a user has a specific role can have access to specific datasets. ABAC is a more fine-grained methodology, dynamic and context-aware, but more complex to implement and grants permissions to a dataset only for the users that have some specific attributes. • Analytics Services help beneficiaries by using data semantics propose meaningful analytics, while exposing ‘‘hidden’’ knowledge or knowledge that general-purpose analytic tools may not easily extract in the energy domain without extensive input from an experienced data analyst. For this, a high performance, cluster computing framework that offers the processing engine for the analysis, along with cluster management and scheduling is required. Apache Spark30 is a suitable choice, which offers management for the execution of distributed application on a cluster. It follows the master/slave architecture along with a cluster manager, like Hadoop YARN31 or Apache Mesos32 that support multiple data processing engines enabling Spark with extended functionalities like enhanced resource management, fair job scheduling and monitor tools over the cluster nodes. For the submission of jobs to the Spark cluster, a set of tools are available, like Apache Livy33 or Spark Job Server34 that enable the remote submission of both batch and interactive jobs through RESTful APIs, from multiple users that can either have shared or separated contexts. A rich set of specialised libraries for machine learning (MLib35 ), graph (GraphX36 ) and stream processing (Spark Streaming37 ) is offered by Spark through high level APIs for Scala, Java, Python and R. By using these libraries and the execution stack that is discussed, the analytical services can be created, incorporating statistical and predictive models on big and semantic data to provide deeper insights for energy usage and more accurate forecasting results. For example, Frequent Pattern Mining using the FP-growth algorithm may identify unobserved patterns of energy consumption in relation to users’ activity or weather parameters, which can be validated from information concerning multiple different buildings. Another example of analytical service is to train classification models like Random Forest or Multilayer Perceptrons with large amounts of multi-sourced data, with many different characteristics, in order to make them identify ‘‘abnormal’’ behaviour of how energy is consumed and then use the trained models to identify such behaviour from streaming data and notify the end-users or suggest actions in real-time. Similar services can also be used for creating a more holistic view of the energy market on the basis of different parameters and assist in evaluating different energy related investments or initiatives. For the selection of data to be fed into the models, a Data Exploration Module would allow the construction of expressive queries so as to preview the data to be used and also use the queries 30 https://spark.apache.org/. 31 https://hadoop.apache.org/docs/current/hadoop-yarn/hadoop-yarnsite/YARN.html. 32 http://mesos.apache.org/.

33 https://livy.incubator.apache.org/. 34 https://github.com/spark-jobserver/spark-jobserver.

)

–

to fetch the data from the storage components. The output of the layer’s services is a set of comprehensive and userfriendly reports, visualisations and notifications that are appropriately distributed to the end-users. • The Dashboard serves as a single, integrated entry point for all the stakeholders and end-users that can use the platform. The Dashboard is the presentation interface for all of the provided services. It is responsible for delivering and formatting all the required information to the users in an effective and comprehensive way and also interact with the back-end services, in order to handle all the user’s actions and requests and then display the relevant content. The interface offers tools for searching and exploring different datasets, services, and other related information. These visualisations may vary from simple statistical charts to map views (using leaflet.js38 ) and more, aiming to deliver useful understanding and insights coming from the data to the users. The Dashboard can also offer an environment to create customised services, based on a set of available analytical components implementing different statistical and analytics algorithms, along with the functionality to perform custom queries on the platform’s datasets. Web-based Notebooks can also be used in order to offer to the experienced users like data analysts an environment for writing and interactively executing Spark jobs and well-known choices for that are Jupyter39 and Apache Zeppelin.40 4. A ‘‘data-driven’’ decision support system This Section presents an example from the utilisation of the above-mentioned novel approach for using multi-sourced data within a smart city context towards the creation of energy management action plans. The added value is the ‘‘data-driven’’ DSS that encounters the inherent barriers of the analytical, simulation based tools, which are building/ sector specific, enabling thus the ‘‘plug in’’ of different infrastructures that generate streams of data, their semantic modelling and further processing, through inference rules, to produce short-term actions. It enables the use of a large amount of time-series data obtained from building energy management and monitoring systems, which are integrated with context data, such as weather conditions and energy prices. In contrast with typical decision support systems, which require a simulation model of the building to suggest decisions, the shortterm action plans suggested by the proposed approach are calculated through the forecasting of building performance in the future based on large amounts of historical data [31]. 4.1. Multi-sourced data sets A large amounts of data are collected from five different domains, including buildings’ energy profiles, weather conditions, occupants’ feedback, renewable energy production and energy prices). More specifically, the data captured are the following:

• Buildings’ Energy Profiles: Data regarding energy consumptions and indoor environment, mainly through BEMS.

• Weather Conditions: Access current weather data and forecasts. Moreover, weather data from control units are collected. • Occupants’ feedback: Data from building occupants acquired through the Thermal Comfort Validator (TCV) [32] or social media, regarding comfort aspects of the building’s occupants.

35 https://spark.apache.org/mllib/. 36 https://spark.apache.org/graphx/.

38 http://leafletjs.com/. 39 http://jupyter.org/.

37 https://spark.apache.org/streaming/.

40 https://zeppelin.apache.org/.

Please cite this article in press as: V. Marinakis, et al., From big data to smart energy services: An application for intelligent energy management, Future Generation Computer Systems (2018), https://doi.org/10.1016/j.future.2018.04.062.

V. Marinakis et al. / Future Generation Computer Systems (

)

–

7

Fig. 1. The platform’s high level architecture.

• Energy production: Data regarding the renewable energy production (photovoltaic, etc.). • Energy prices: Data from the energy market, including electricity, oil, gas prices, etc.

4.2. Semantic modelling of data For integrating the data from different domains, a holistic interoperability solution has been implemented, based on Semantic Web technologies. A data integration process is established, which encompasses four steps: ‘‘Data Translation’’, ‘‘Data Communication’’, ‘‘Data Contextualisation’’ and ‘‘Data Storage’’ [33]. The OPTIMUS ontology was developed, based on two already existing ontologies: Urban Energy ontology [34] and Semantic Sensor Network ontology [35]. OPTIMUS ontology was created for all entities that are either included or related to the smart building environment and constitutes the main vocabulary upon which the action plans were based. The OPTIMUS ontology has been coded in OWL language using the ClickOn ontology editor [36]. 4.3. Energy management action plans The semantically modelled data is issued by the prediction models and inference rules to derive the short-term actions to improve the building’s energy performance. Four (4) prediction models have been developed for forecasting renewable energy production, energy consumption, indoor temperature and energy prices. The prediction models connect the semantically integrated data with the inference rules. They take as input both historical and monitored data from the semantic framework in order to forecast the building behaviour [37]. Seven (7) action plans were structured based on inference rules, using them in order to derive suggestions for the energy managers to improve the building’s performance taking into account

energy consumption, energy cost, carbon emissions, renewables production and thermal comfort (Fig. 2). The action plans refer to the energy management in buildings taking into consideration their interaction with energy systems, they can foster efficient management of energy flows at a broader level, integrating energy demand, generation and data/energy infrastructures. 1. Scheduling and management of the occupancy: It aims at the reduction of the building energy consumption by changing the location of building occupants. In this way, a minimum number of thermal zones can be used and the consumption can be reduced by turning off the heating/cooling system in the unoccupied zones [38–40]. The expected reduction of energy consumption is estimated at 5%–9% for cooling and 2%–4% for heating [41,42]. 2. Scheduling the set-point temperature: Based on the application of two inference rules, this action plan is aimed to adjust the indoor temperature set-point by taking into consideration, respectively, thermal comfort as submitted by the building users (using the Thermal Comfort Validator web application), and the adaptive comfort concept [43–47]. The target is to reduce energy use, while maintaining comfort levels in accepted ranges. The expected reduction of energy consumption is estimated at 5%–9% for cooling/heating [48– 50]. 3. Scheduling the ON/OFF of the heating system: The action plan concerns the optimisation of the boost time of the heating system. The expected reduction of energy consumption is estimated at 5%–10% for heating [51]. 4. Management of the air side economiser: It involves the scheduling of the amount of outdoor air to be used for cooling the indoor environment, in order to reduce or eliminate the need for mechanical cooling when favourable outdoor conditions occur, using air-side economiser technology [52– 54]. The expected reduction of energy consumption is estimated at 10%–20% for cooling and 5%–10% for heating [55].

Please cite this article in press as: V. Marinakis, et al., From big data to smart energy services: An application for intelligent energy management, Future Generation Computer Systems (2018), https://doi.org/10.1016/j.future.2018.04.062.

8

V. Marinakis et al. / Future Generation Computer Systems (

5. Scheduling the photovoltaic (PV) maintenance: It aims at the detection of the need for maintenance of the PV system, alerting the user to check if corrective actions are necessary. This facilitates the identification of PV malfunctioning. The expected increase of renewable energy production is estimated at 3%–8% (empirically, based on the available data from the pilot cities). 6. Scheduling the sale/consumption of the electricity produced through the PV system: The DSS suggests for the week ahead a procedure which allows both to improve the exploitation of solar energy maximising the self-consumption of electricity produced by PV on-site and to take advantage from the selling of the energy surplus considering the energy prices. The expected reduction of energy cost is estimated at 5%– 10% [56]. 7. Scheduling the operation of heating and electricity systems towards energy cost Reduction: It reduces the energy cost of the building(s) by scheduling simultaneously the operating schedule of the heating (Combined Heat and Power (CHP) and boilers) and electricity systems (grid, PV plant & batteries) for the upcoming week. Thus, the AP firstly specifies based on the season (winter/summer) the schedule of the heating/cooling systems and then suggests when the energy generated should be used, stored or sold in order to reduce energy cost or even make a surplus. The outcome of the energy demand and RES prediction models, as well as weather and energy prices forecasts, are exploited in order to schedule the energy flows from/to the grid and the batteries and reduce energy cost based on load shifting and peak shaving techniques. The expected reduction of energy cost is estimated at 5%–10% [57].

4.4. Dashboard The DSS integrates two main interfaces, namely the ‘‘DSS management interface’’ and the ‘‘DSS end-user web interface’’ (Fig. 4). At the start of the implementation process, the DSS should be configured in order to meet the specific requirements of the end-user. This is done by means of the DSS interfaces which are available to the user ‘‘administrator’’ (Fig. 3). The DSS user (energy manager, etc.) is in charge of interacting with the DSS on a daily basis, accepting or rejecting the recommendations of the DSS (Fig. 4). The DSS user is in contact with the occupants of the buildings and receives their feedback after the application of the measures recommended by the DSS. 5. Real life application

)

–

◦ Savona Campus (belonging to the University of Genoa): Its structure hosts typical educational buildings (offices, classrooms, and laboratories), research centres, private companies, as well as student residences (somewhat close to residential buildings), it closely resembles any urban district. Furthermore, a number of initiatives in the context of both ‘‘smart city’’ and ‘‘smart grid’’ concepts are being developed in the Campus. In particular, the Smart Polygeneration Microgrid is a mixed thermal and electrical microgrid hosting sources that contribute to satisfy the electrical and thermal demands of the Campus. Thanks to this infrastructure, the site can become an example of a ‘‘smart urban district’’, equipped with distributed generations, local control and supervision infrastructures.

• The city of Zaanstad (the Netherlands) has approximately 150,000 inhabitants and is part of the larger Amsterdam metropolitan area. At 2007 the municipality has set a target to become energy neutral at 2020. This ambition resulted in the Integral Climate program of Zaanstad in 2009. The Zaanstad Town Hall was selected as a pilot site. It was built in 2011, has a floor area of 23.000 m2 and is designed as an energy class A+ building. The Town Hall has a public use and is also used as an office building for the employees working at the municipality. It is a new building with a new Building Energy Management System, sensors in every room and flexible working desks. • The city of Sant Cugat del Vallès is located in the region of Catalonia (Spain) approximately 20 km away from Barcelona. The city of Sant Cugat del Vallès is a municipality of around 90.000 inhabitants. The aim of the SEAP is to reduce the energy consumption of the municipality to 248.988 MWh and greenhouse emissions to 108.997 tCO2 . This represents a reduction of 21% in energy consumption and 24% of emissions with respect to the values of 2005. The selected pilot sites are the following:

◦ Town Hall of Sant Cugat: A summary of basic data that provide a quick overview of the building is the following: (a) floors: 3 + 2 basements; (b) total area: 8.593 m2 ; (c) yearly electricity consumption: 1.221.954 kWh. Moreover, there is a PV installation on the roof of the Town Hall (17 kW), which is interconnected directly to the grid. ◦ Theatre Auditori Sant Cugat: A summary of basic data that provide a quick overview of the building is the following: (a) floors: 2 + 1 basement + roof; (b) total area: 4.150 m2 ; (c) yearly electricity consumption: 508.613 kWh; (d) yearly gas consumption: 698.995 kWh.

5.1. Selected pilot sites The effectiveness of the DSS has been verified in the following cities:

• The city of Savona (Italy) extends over an area of 65,5 km2 and has a population of 63.000. The municipality submitted its SEAP to the Covenant of Mayors in July 2014. The selected pilot sites are the following:

◦ ‘‘Colombo-Pertini’’ School: The total floor area inside the building envelope is 6.092 m2 and includes 3 floors and 1 basement. It is a representative of a widespread type of public buildings. There is a PV system on the roof with an installed power of about 20 kW.

5.2. Baseline Scenario The baseline year is 2013. In Savona Campus, Colombo-Pertini School, Sant Cugat Theatre and Zaanstad Town Hall, the natural gas consumption is related to the heating needs, while electricity consumption is related to the cooling, DHW, lighting, etc. In the Sant Cugat Town Hall, electricity is used to cover the needs for heating, cooling, lighting, etc. A microgrid has been installed in Savona Campus, including PV system, energy storage equipment (battery), as well as two CHP units. PV systems have been also installed in Colombo-Pertini School and Sant Cugat Town Hall. Table 1 presents the baseline scenario of the pilot sites for one year period.

Please cite this article in press as: V. Marinakis, et al., From big data to smart energy services: An application for intelligent energy management, Future Generation Computer Systems (2018), https://doi.org/10.1016/j.future.2018.04.062.

V. Marinakis et al. / Future Generation Computer Systems (

)

–

9

Fig. 2. Action plans and inference rules.

Fig. 3. Configuration of the DSS.

Please cite this article in press as: V. Marinakis, et al., From big data to smart energy services: An application for intelligent energy management, Future Generation Computer Systems (2018), https://doi.org/10.1016/j.future.2018.04.062.

10

V. Marinakis et al. / Future Generation Computer Systems (

)

–

Fig. 4. ‘‘Scheduling the operation of heating and electricity systems towards energy cost reduction’’.

Table 1 Baseline scenario, 2013 (total figures). Savona Campus

Colombo-Pertini School

Sant Cugat Town Hall

Sant Cugat Theatre

Zaanstad Town Hall

Total

Total energy consumption (MWh/year) Electricity (MWh/year) Natural gas (MWh/year)

2.261 900 1.361

625 42 582

1.222 1.222 Not Appl .

1.208 509 699

4.435 2.905 1.530

9.751

Total CO2 emissions (tCO2 /year) Electricity (tCO2 /year) Natural gas (tCO2 /year)

723 435 289

138 20 118

538 538 Not Appl .

365 224 141

1.573 1.264 309

3.337

Total energy cost (e/year) Electricity (e/year) Natural gas (e/year)

280.502 162.000 118.502

60.400 10.400 50.000

193.343 193.343 Not Appl .

141.813 99.329 42.484

208.396 163.861 44.535

884.454

Energy Production (MWh/year) PV (MWh/year) Electricity from CHP (MWh/year) Thermal energy from CHP (MWh/year)

668 94 177 397

20 20

22 22

Not Appl.

Not Appl.

710

Not Appl.

Not Appl.

Not Appl.

Not Appl.

5.3. Installation and configuration At the start of the pilot implementation process, the DSS was installed and configured to meet the specific requirements of the pilot cities (Table 2). This is done by means of the DSS interfaces, which are available to the user ‘‘administrator’’. More than 440 occupants in Sant Cugat Town Hall, 300–500 occupants in Sant Cugat Theatre, 200–1200 occupants in Savona Campus, 20–500 occupants in Savona School and 700 occupants in Zaanstad Town Hall interacted with the DSS.

5.4. Pilot implementation and results The use of the DSS has been related to periods different from case to case, based on the availability of the several functionalities in the cities. Thus:

• Sant Cugat Town Hall started using the first version of DSS in February 2016, and started monitoring the building dashboard in May 2016. • Sant Cugat Theatre started using the DSS and monitoring the building dashboard in July 2016.

Please cite this article in press as: V. Marinakis, et al., From big data to smart energy services: An application for intelligent energy management, Future Generation Computer Systems (2018), https://doi.org/10.1016/j.future.2018.04.062.

V. Marinakis et al. / Future Generation Computer Systems (

)

–

11

Table 2 Installation of the DSS to the pilot cities. Sant Cugat

Savona

Zaanstad

Data capturing

For the weather forecasting a service has been contracted with the company TechCom S.R.L. The energy production data capturing module has been set up to get data from SunnyBox provider. For the data capturing on the feedback by the occupants two streams have been configured one for each building. The energy prices data capturing module has been set up to get data from the Spanish market. The de-centralised data capturing module has been set up to get data from each building: Town Hall-BEMS (Energea) and a monitoring system (Desigo); Theatre-Dalkia management system.

For the weather forecasting a service already in use at the Campus for the management of operations of the smart polygeneration microgrid has been offered to DSS without additional charge. The energy production data capturing module has been set up to get data from a FTP server provided by the Campus. For the Social data capturing module two streams have been configured one for each building. The energy prices data capturing module has been set up to get data from the Italian market. The de-centralised data capturing module has been set up to get data from each building.

The data capturing modules have been configured to get data from Zaanstad premises. The weather forecasting a service has been contracted with the company TechCom S.R.L. The energy production data capturing module has not been installed in this pilot. For the Social data capturing module one streams has been configured. The energy prices data capturing module has not been installed in this pilot. The de-centralised data capturing module has been set up to get data from the pilot building through novaPro open 2014 BEMS and FlexWhere system.

Semantic framework

A total number of 147 streams have been set up for the Town Hall and 91 streams for the Theatre.

A total number of 66 streams have been set up for the School building and 24 streams for the Campus.

A total number of 324 streams have been set up for the Town Hall building.

Analytics

The prediction models and action plans encapsulated as Rapidminer processes have been deployed in the RapidAnalytics server. Each of those processes is set up in order to use the data provided by the semantic framework.

DSS front-end environments

10 processes have been configured in RapidAnalytics server.

9 processes have been configured in RapidAnalytics server.

1 prediction model encapsulated as Rapidminer process has been deployed in the RapidAnalytics server in order to use the data provided by the semantic framework.

Town Hall: 15 zones and 5 action plans. Theatre: 8 zones and 2 action plans. 3 virtual sensors have been created.

School: 7 zones and 3 action plans. Campus: 1 zones and 2 action plans. 8 virtual sensors have been created.

Town Hall 34 zones and 3 Actions Plans. 3 virtual sensors have been created.

• Savona Campus started in March 2016 to use the dashboard, whereas the AP5 has been available since April 2016 and the AP7 since May 2016. • Savona Colombo-Pertini School started in May 2016 to use all the functionalities. • Zaanstad Town Hall started to monitor the building dashboard in March 2016, whereas the actual use of the APs is dated 11th July 2016, after a period of adjustments to the configuration of the BEMS started in November 2015. Some insights per action plan during the pilot implementation of the action plans to the three cities are summarised in Table 3. The assessment of the impact in the application of action plans is based on the comparison of the energy consumption before and after their implementation (pre-action vs post-action) [58]. The comparison is built through the real time data collection related both to climate and users and to historical energy consumption. The DSS implementation can achieve significant reduction of the energy consumption, CO2 emissions and energy cost (in some cases even beyond 20%), as well as approximately 10% increase of renewable energy production. The impact of the DSS per pilot site is presented in Table 4. 5.5. Future perspectives: linking DSS with the energy currency concept A future perspective is to link DSS with the energy currency concept. A number of related approaches have been proposed for the reduction of CO2 emissions, such as Ergo [59–62] and DeKo [63], as well as for the in-feed of renewable energy, most notably SolarCoin [64], NRGcoin [65], GENERcoin [66]. While these proposals address environmental issues, they overlook the energy efficiency sector. So, the connection between energy saving behaviour and a digital currency has to be studied and formalised. More specifically, energy end-users are able to earn coins, in terms of digital currencies, by reducing or shifting their energy

consumption. The coins earned are related with the actual compared to the predicted energy consumption and the daily amount of energy saved by the end-user (as a result of the implementation of the DSS action plans). The value of the coin varies among the end-users, based on their energy profile, the savings, the hour that these saving are achieved (increase value for the ones achieved in peak hour) and is also determined both from the absolute savings of the individual user and the achievements of the rest of the endusers of the specific energy company that applies this App. For every 1 kWh reduced, the end-user earns the respective coins, which can be used to decrease final energy cost by paying part of the bill or all of it to the energy company. Such an approach enables dynamic billing and allows energy companies to incentivise end users to follow the energy management action plans proposed. Such an interaction can be facilitated by a blockchain based cryptocurrency system. Blockchain is a burgeoning technology for decentralised computation and data storage and is well known for underpinning the cryptocurrency Bitcoin. Blockchain in the energy sector, and in the power systems especially, is a novel approach that combines smart contracts along with the ever-expanding IoT devices ecosystem that allows for a peer-to-peer market where machines can buy and sell energy automatically, according to user-defined criteria [67]. Various companies are developing such applications although these are still in the concept or pilot phase yet [68]. The most reputable of these projects is the so-called Brooklyn Microgrid developed by TransactiveGrid, a joint venture between LO3 Energy and Consensys. Others include RWE, Oneup, Wien Energy, Electron and Vantenfall relevant projects and one may also add SolarCoin, which is a cryptocurrency based project where, however, the SolarCoin is created by the generation of 1 MWh of energy from photovoltaic systems [69]. Besides energy sector, there are several relevant projects from the industry that combine the concept of the blockchain with a wide range of industries such as finance, healthcare, real estate, the government-sector and power systems. The described approach is only an initial phase. The use of blockchain technology to integrate with a community or urban

Please cite this article in press as: V. Marinakis, et al., From big data to smart energy services: An application for intelligent energy management, Future Generation Computer Systems (2018), https://doi.org/10.1016/j.future.2018.04.062.

12

V. Marinakis et al. / Future Generation Computer Systems (

)

–

Table 3 Insights per action plan during the pilot implementation. Action plans

Insights

Scheduling and management of the occupancy

This Action Plan was selected for Zaanstad taking into consideration that the employees do not have fixed working desks but can choose the place where they want to work. The DSS receives data from the building and transform them into real Action Plans, namely which part of the building can be left empty on specific days. During the implementation of the occupancy Action Plan significant reduction of energy consumption was achieved. The DSS user of the Zaanstad Town Hall decided that the scheduling of the set point temperature can be carried out every day. The DSS user informed the technical end user to implement the Action Plans on the following days till further notice. It has never led to complaints from the occupants of the building. The Action Plan concerns the scheduling of the boost time of the heating system. The winter 2016 was unusually hot in Sant Cugat and therefore the model based on a grey box approach that was validated using the heating season 2015, could not be properly applied. However, through simulations it was demonstrated that the management of the start/stop of the heating system according to the model implemented in the DSS can potentially reduce, even if slightly, the energy consumption for space heating. The economiser Action Plan gives the opportunity for the use of the outdoor conditions to precool the building when the outdoor condition are in properly conditions. This kind of suggestions are highly appreciated in the Sant Cugat Town Hall due to its potentiality to save energy for cooling. The Action Plan suggests different schedules in order to apply total or partial free cooling depending on the indoor and outdoor conditions. For instance, in summer time the suggestion was to supply outdoor air without treatment from 5 am till 8 am, in order to precool the building since the outdoor temperature was low enough. The DSS correctly issued warnings for the PV system in the Savona Campus. For instance, during the week 20–24/6 the local control panel of the PV inverter was out of order (kept resetting) and this made the inverter go off-line and back on-line from time to time, resulting in a loss of production. In this respect, the energy manager contacted the inverter maintenance service to solve this malfunction. The DSS suggests for the week ahead a procedure which allows both to improve the exploitation of solar energy maximising the self-consumption of electricity produced by PV on-site and to take advantage from the selling of the energy surplus considering the energy prices. The green strategy implemented in Savona school can contribute to maximise the use the daily amount of energy from renewable sources. The data driven models of energy generation by PV and total electrical energy demand proved to be robust to predict the surplus of energy on the basis of the historical performance data. This rule was highly appreciated in Savona school where electrical loads associated to the computer laboratory were selected as the most suitable shiftable loads for exploiting the potentialities of Action Plan 6. As far as the scheduling of sources is concerned, the power profile computed by the DSS is imposed to the Energy Management System (EMS) of Savona Campus. For instance, the electricity storage system operates following the scheduling suggested by the DSS. In this respect, the technician specifies the ‘‘fixed scheduling’’ operating mode in the input file for the unit commitment module of the Campus EMS, copy the DSS scheduling in the same input file and run the unit commitment algorithm.

Scheduling the set-point temperature

Scheduling the ON/OFF of the heating system

Management of the air side economiser

Scheduling the PV maintenance

Scheduling the sale/ consumption of the electricity produced through the PV system

Scheduling the operation of heating and electricity systems towards energy cost reduction

Table 4 Impact of the DSS per pilot site. Pilot Site Savona Campus Savona School Sant Cugat Town Hall Sant Cugat Theatre Zaanstad Town Hall

Delivered energy [electricity] %

Delivered energy [natural gas] %

−5.3 −29.4 −51.4 −23.5

−39.0

EMS and support offers immense opportunities for time-of-use, dynamic response systems that would allow massive integrate of renewable energy and create a market-based mechanism to ensure a balance between demand management, storage and load shifting and therefore would allow the real-world, real-time application of the Ergo concept discussed earlier. Such an implementation would facilitate the revolutionary inversion of current energy market mechanisms—wholesale supply of RE would be provided through long-term stable price contracts and the retail demand would be moderated through participation in a real-time market bidding mechanism (with some minimum security of supply stipulations) at the consumer level. 6. Conclusions The proposed blueprint for a smart energy platform can provide data-driven innovative pathways in energy efficiency by incorporating novel approaches and technologies. A wide range of data analysis techniques (including among others optimisation, forecasting, classification and clustering) can be applied, supporting the design of innovative business models for several ben-

CO2 emission %

Energy cost %

RE produced %

−4.0 −7.8 −29.4 −47.0 −23.5

−2.8 −7.3 −29.4 −48.0 −23.5

13.3 10.2 7.4

eficiaries (stakeholders), such as city authorities and local administrations, energy managers and consultants, utilities and energy providers, ESCOs and other energy companies, construction companies, building components manufacturers, investors and financiers and other organisations and third parties. The ‘‘data-driven’’ DSS that was developed, exploits multisourced data from five different domains (weather conditions, buildings’ energy profiles, occupants’ feedback, energy prices and energy production), in order to make predictions of the building energy performance and help energy managers to adopt measures (namely short-term action plans) to improve it. The DSS has been designed with the necessary degree of generalisation, so as to be adapted by both public and private sector organisations, with different characteristics, energy infrastructures, needs, priorities and types of energy demand:

• Public sector: The proposed platform as well as the respective DSS can support Signatories to the CoM, which want to monitor and reduce the energy use in their buildings so that they can effectively implement SECAPs. It has been successfully applied in three cities (Sant Cugat, Savona and Zaanstad), to help improving municipal buildings.

Please cite this article in press as: V. Marinakis, et al., From big data to smart energy services: An application for intelligent energy management, Future Generation Computer Systems (2018), https://doi.org/10.1016/j.future.2018.04.062.

V. Marinakis et al. / Future Generation Computer Systems (

• Private sector: The platform is generic enough as to be applied to other privately owned buildings (e.g. private organisations with different types of buildings, who want to improve their energy efficiency and therefore, their energy spending). The added value of the proposed approach consists of correlating various types of real-time data from different sources, hence integrating different systems, in order to optimise and achieve sustainable energy management of the built environment. The DSS implementation can achieve significant reduction of the energy consumption, CO2 emissions and energy cost (in some cases even beyond 20%), as well as approximately 10% increase of renewable energy production. The results strongly depend on the current status of the building, as well as the action plans which will be implemented. More specifically, based on the available data and infrastructure, the energy managers of the city/buildings can decide to plug in single buildings (action plans 1–4), buildings connected to energy production (action plans 5–6) and buildings connected to energy production and other energy systems (action plan 7). The DSS gives the opportunity for the development of new action plans that integrate additional energy systems and domains (priority areas) of the Smart Energy Cities, such as ‘‘Sustainable Urban Mobility’’ (e.g. optimal charging scheduling of the electrical vehicles, etc.). It can be also further upgraded integrating the energy currency concept, in order to nudge end-users into more efficient usage of energy consuming devices, enhance participation and effectiveness of energy conservation actions. Acknowledgements The work presented is based on research conducted within the framework of the projects ‘‘OPTIMising the energy USe in cities with smart decision support system (OPTIMUS)’’, European Union Seventh Framework Programme, Greece, grant agreement №608703 (http://optimus-smartcity.eu/) and ‘‘EU-GCC Clean Energy Technology Network’’, European Commission, Greece FPI service contract number PI/2015/370817 (http://www.eugcccleanergy.net). The authors wish to thank Alfonso Capozzoli, Alice Gorrino and Vincenzo Corrado (Politecnico di Torino — Italy), Victor Martinez (Sant Cugat — Spain), Mansueto Rossi (UNIGE, Italy), Pietro Pera (Savona/IPS — Italy) and Alex Beijers (Zaanstad — the Netherlands), whose contribution, helpful remarks and fruitful observations were invaluable for the development of this work. The content of the paper is the sole responsibility of its authors and does not necessary reflect the views of the EC. References [1] SunGard, Big Data — Challenges and Opportunities for the Energy Industry, 2013. https://www.sungard.com/media/fs/energy/resources/white-papers/Bi g-Data-Challenges-Opportunities-Energy-Industry.ashx. (Accessed 12.05.17). [2] P.D. Diamantoulakis, V.M. Kapinas, G.K. Karagiannidis, Big data analytics for dynamic energy management in smart grids, Big Data Res. 2 (3) (2015) 94–101. [3] K. Zhou, C. Fu, S. Yang, Big data driven smart energy management: From big data to big insights, Renew. Sustain. Energy Rev. 56 (2016) 215–225. [4] X. Jin, B.W. Wah, X. Cheng, Y. Wang, Significance and challenges of big data research, Big Data Research 2 (2) (2015) 59–64. [5] Grand View Research, Energy Management Systems (EMS) Market Analysis By Product (IEMS, BEMS, HEMS), By Component (Sensors, Controllers, Software), By Vertical (Power &Energy, Telecom &IT, Manufacturing, Retail &Offices, Healthcare), By End-Use (Residential, Commercial) And Segment Forecasts To 2024, 2015. http://www.grandviewresearch.com/industry-analysis/ energy-management-systems-market. (Accessed 18.01.17). [6] Mordor Intelligence, Europe Energy Management Systems Market Outlook to 2021 - Market Analysis by Geography, Applications, Type of Solution, HEMS, BEMS, FEMS, Vertical, Competitive Landscape, Key Company Information, 2017. https://www.mordorintelligence.com/industry-reports/Europeenergy-management-systems-market-industry. (Accessed 18.01.17).

)

–

13

[7] C. Talon, R. Martin, Next-Generation Building Energy Management Systems. Navigant Research, 2015. https://www.navigantresearch.com/research/nextgeneration-building-energy-management-systems. (Accessed 18.01.17). [8] C. Ntanos, C. Botsikas, G. Rovis, P. Kakavas, D. Askounis, A context awareness framework for cross-platform distributed applications, J. Syst. Softw. 88 (1) (2014) 138–146. [9] V. Marinakis, H. Doukas, C. Karakosta, J. Psarras, An integrated system for buildings’ energy-efficient automation: Application in the tertiary sector, Applied Energy 101 (2013) 6–14. [10] Energydata info. https://energydata.info/about(Accessed 11.2.18). [11] Energy DataBus, https://github.com/deanhiller/databus. (Accessed 11.2.18). [12] Cassandra, http://cassandra.apache.org/. (Accessed 11.2.18). [13] KNIME, https://www.knime.org/ 11.2.18). [14] KNIME, https://www.knime.org/files/knime_bigdata_energy_timeseries_whi tepaper.pdf. (Accessed 11.2.18). [15] Energywise, https://github.com/dssg/energywise. (Accessed 11.2.18). [16] Hadoop, http://hadoop.apache.org/. (Accessed 11.2.18). [17] Cloudera, https://www.cloudera.com. (Accessed 11.2.18). [18] MongoDB, https://www.mongodb.com/. (Accessed 11.2.18). [19] MongoDB, https://www.mongodb.com/blog/post/whats-new-mongodb-30part-1-95-reduction-operational-overhead-and-security-enhancements. (Accessed 11.2.18). [20] OpenRefine, http://openrefine.org/. (Accessed 11.2.18). [21] DataCleaner, https://datacleaner.org/. (Accessed 11.2.18). [22] BigML, https://bigml.com/. (Accessed 11.2.18). [23] Qubole, https://www.qubole.com/. (Accessed 11.2.18). [24] Lumify, http://lumify.io/. (Accessed 11.2.18). [25] Talend, https://www.talend.com/. (Accessed 11.2.18). [26] Zoomdata, https://www.zoomdata.com/. (Accessed 11.2.18). [27] Tableu, https://www.tableau.com/. (Accessed 11.2.18). [28] DataWatch, http://www.datawatch.com/. (Accessed 11.2.18). [29] V. Marinakis, H. Doukas, P. Xidonas, C. Zopounidis, Multicriteria decision support in local energy planning: An evaluation of alternative scenarios for the sustainable energy action plan, Omega 69 (2017) 1–16, United Kingdom. [30] V. Marinakis, A.G. Papadopoulou, H. Doukas, J. Psarras, A web tool for sustainable energy communities, Int. J. Inf. Decis. Sci. 7 (2015) 18–31. [31] A. Capozzoli, F. Corno, V. Corrado, A. Gorrino, The overall architecture of a decision support system for public buildings, Energy Procedia 78 (2015) 2196– 2201. [32] V. Marinakis, H. Doukas, E. Spiliotis, I. Papastamatiou, Decision support for intelligent energy management in buildings using the thermal comfort model, Int. J. Comput. Intell. Syst. 10 (1) (2017) 882–893. [33] Á. Sicilia, G. Costa, V. Corrado, A. Gorrino, F. Corno, A Semantic decision support system to optimize the energy use of public buildings, in: 32nd International CIB W078 Conference, Eindhoven, The Netherlands, 2015, pp. 676-685. [34] G. Nemirovski, A. Nolle, A. Sicilia, I. Ballarini, V. Corrado, Data integration driven ontology design, case study smart city, in: 3rd International Conference on Web Intelligence, Mining and Semantics, WIMS ’13, Madrid, 2013. [35] W3C - W3C Semantic Sensor Network Incubator Group. Semantic Sensor Network Ontology. 2011. http://www.w3.org/2005/Incubator/ssn/ssnx/ssn. (Accessed 08.11.16). [36] M. Wolters, G. Nemirovski, A. Nolle, ClickOnA: An editor for DL-liteA based ontology design, in: Informal Proceedings of the 26th International Workshop on Description Logics, vol. 1014, 2013, pp. 377–389, Ulm, Germany: CEURWS.org.. [37] S. Androulaki, H. Doukas, V. Marinakis, L. Madrazo, N.-Z. Legaki, Enabling local authorities to produce short-term energy plans: A multidisciplinary decision support approach, Manag. Environ. Qual. 27 (2) (2015) 146–166. [38] A.A. Adamopoulou, A.M. Tryferidis, D.K. Tzovaras, A context-aware method for building occupancy prediction, Energy Build. 110 (2016) 229–244. [39] S. D’Oca, T. Hong, Occupancy schedules learning process through a data mining framework, Energy Build. 88 (2015) 395–408. [40] Z. Yang, B. Burcin, The coupled effects of personalized occupancy profile based HVAC schedules and room reassignment on building energy use, Energy Build. 78 (2014) 113–122. [41] UNC, A Method for Calculating Chilled Water and Steam Energy Savings Due to Occupancy Scheduling in Large Buildings with Only One Year of Data, 2016. https://save-energy.unc.edu/Portals/2/Calculating%20Occupancy%20Sc hedule%20Savings.pdf?ver=2012-10-26-133759-960. (Accessed 14.12.16). [42] Z. Yang, B. Becerik-Gerber, The coupled effects of personalized occupancy profile based HVAC schedules and room reassignment on building energy use, Energy Build. 78 (2014) 113–122. [43] A. Beizaee, S.K. Firth, A comparison of calculated and subjective thermal comfort sensation in home and office environment, in: Proceedings of Conference: People and Buildings, MC2011, London, 2011. [44] ASHRAE - American Society of Heating, Refrigerating and Air-Conditioning Engineers, Thermal environmental conditions for human occupancy, ASHRAE Standard 55-2010, 2010. [45] ISO - ISO 13790. Thermal performance of buildings - Calculation of energy use for space heating, 2004.

Please cite this article in press as: V. Marinakis, et al., From big data to smart energy services: An application for intelligent energy management, Future Generation Computer Systems (2018), https://doi.org/10.1016/j.future.2018.04.062.

14

V. Marinakis et al. / Future Generation Computer Systems (

[46] ISO - ISO 7730. Ergonomics of the thermal environment - Analytical determination and interpretation of thermal comfort using calculation of the PMV and PPD indices and local thermal comfort criteria, 2005. [47] J.F. Nicol, M.R.B. Kessler, Perception of comfort in relation of weather and indoor adaptive opportunities, ASHRAE Trans. (1998) 1005-1017. [48] S. Attia, S. Carlucci, Impact of different thermal comfort models on zero energy residential buildings in hot climate, Energy Build. 102 (2015) 117–128. [49] N. Fernandez, S. Katipamula, W. Wang, Y. Huang, G. Liu, Energy savings modeling of standard commercial building re-tuning measures: Large office buildings. Prepared for U.S. Department of Energy, 2012 http://www.pnnl.gov/ buildingretuning/documents/pnnl_21569.pdf. (Accessed 14.12.16). [50] M. Sala, C. Gallo, A.A.M. Sayigh, Architecture—Comfort &Energy, Elsevier Technology &Engineering, 1999, p. 234. [51] Curbontrust, How to implement optimum start control, 2016. https://www. carbontrust.com/media/131445/ctl035_how_to_implement_optimum_start_ control.pdf. (Accessed 14.12.16). [52] G. Wang, L. Song, An energy performance study of several factors in air economizers with low-limit space humidity, Energy Build. 64 (2013) 447–455. [53] P. Roach, F. Bruno, M. Belusko, Modelling the cooling energy of night ventilation and economiser strategies on façade selection of commercial buildings, Energy Build. 66 (2013) 562–570. [54] J.E. Seem, J.M. House, Development and evaluation of optimization-based air economizer strategies, Appl. Energy 87 (3) (2010) 910–924. [55] G. Moeseke, I. Bruyère, A. De Herde, Impact of control rules on the efficiency of shading devices and free cooling for office buildings, Build. Environ. 42 (2) (2007) 784–793. [56] D. Steen, A. Tuan, L. Bertling, Price-based demand-side management for reducing peak demand in electrical distribution systems –with examples from Gothenburg, 2016. http://publications.lib.chalmers.se/records/fulltext/ 163330/local_163330.pdf. (Accessed 14.12.16). [57] X. Guan, X. Zhanbo, J. Qing-Shan, Energy-efficient buildings facilitated by microgrid, IEEE Trans. Smart Grid 1 (3) (2010). [58] A. Capozzoli, V. Corrado, H. Doukas, A. Gorrino, M. Piscitelli, M. Rossi, V. Spiliotis, Impact Analysis Report. Deliverable D4.5 of the OPTIMUS Project ‘‘OPTIMising the Energy USe in Cities with Smart Decision Support System’’, 2016. Available online: http://optimus-smartcity.eu/. (Accessed on 2 February 2018). [59] S. Sgouridis, S. Kennedy, Tangible and fungible energy: Hybrid energy market and currency system for total energy management. A Masdar city case study, Energy Policy 38 (2010) 1749–1758. [60] S. Sgouridis, Using energy as currency: Re-establishing the bridge between the financial and the real world, Netw. Ind. Q. 14 (2012) 8–11. [61] S. Sgouridis, D. Csala, U. Bardi, The sower’s way: Quantifying the narrowing net-energy pathways to a global energy transition, Environ. Res. Lett. 11 (9) (2016) 1–8. [62] S. Sgouridis, Defusing the energy trap: The potential of energy-denominated currencies to facilitate a sustainable energy transition, Energy Systems and Policy 2 (2014) 8. [63] N. Gogerty, J. Zitoli, DeKo: An electricity-backed currency proposal, SSRN Electron. J. (2011). [64] SolarCoin, https://solarcoin.org/en/front-page/. (Accessed 16.03.17). [65] M. Mihaylov, S. Jurado, N. Avellana, K. Van Moffaert, I.M. de Abril, A. Nowe, NRGcoin: Virtual Currency for Trading of Renewable Energy in Smart Grids, in: 11th International Conference on the European Energy Market, 2014. [66] GENERcoin—Offering Green Energy at a Reduced Price, http://www.cryptocoi nsnews.com/genercoin/. (Accessed 16.03.17). [67] K. Christidis, M. Devetsikiotis, Blockchains and smart contracts for the internet of things, IEEE Access 4 (2016) 2292–2303. [68] PricewaterhouseCoopers, Blockchain—an opportunity for energy producers and consumers, https://www.pwc.ch/en/2017/pdf/pwc_blockchain_opportun ity_for_energy_producers_and_consumers_en.pdf. (Accessed 16.03.17). [69] J. Luo, Investigate Centralized and Decentralized Information Infrastructure for Future Electricity Market (Master thesis), University of Michigan-Dearborn, 2017.

Dr. Vangelis Marinakis is an Electrical and Computer Engineer of the National Technical University of Athens and holds a PhD in the research domain of decision support systems for sustainable energy planning. His experience and knowledge cover the fields of energy and environmental/climate policy, design and implementation of related decision support systems and information technology systems, promotion of energy efficiency/renewable energy technologies and modern financing mechanisms, intelligent energy management, elaboration of sustainable energy and climate action plans, and energy corporate responsibility of enterprises. His current research is focused on the design and development of innovative systems and ‘IoT’ solutions for Smart Cities, using big data

)

–

technologies and analytics. Dr. Marinakis is working for more than 10 years in the fields of ICT, energy and climate, participating in several European (Horizon 2020, FP7, IEE, etc.) and national projects. He has more than 45 scientific publications in international journals and book chapters, as well as numerous announcements in national and international conferences.

Dr. Haris Doukas is Assistant Professor in the School of Electrical & Computer Engineering, National Technical University of Athens, in the research domain of ‘‘Decision Support Systems for Energy and Environmental Policy’’. Since September 2003 is an Expert on Decision Support for Energy Policy and Planning towards Sustainability in the Energy Policy Unit of National Technical University of Athens. Dr. Doukas key qualifications include among other the decision support systems design, development & administration (expert and multiple criteria decision support methods and systems) for the promotion of renewable energy sources and energy efficiency as well as for sustainable energy planning in a local level. Dr. Doukas has more than 80 scientific publications in international journals and more than 100 announcements in international conferences, chapters in books etc. For his work, Dr. Doukas has received awards by the State Scholarship Foundation (IKY), the NTUA, the AUTH the Technical Chamber of Greece (TCG) and the Hellenic Operational Research Society (HELORS).

Ioannis Tsapelas is an Electrical and Computer Engineer (2015) and a PhD candidate in the Decision Support Systems Laboratory of the School of Electrical and Computer Engineering, NTUA. He also holds a MSc degree in Business Administration (MBA) from AUEB (2017). Since 2015, he participates in EU-funded research projects in the fields of Big Data and Decision Support Systems (DSS), including Big Data Ocean, AEGIS and OPTIMUS. His competencies include collaborative software design and development, database systems management, knowledge representation and DSS implementation. He is fluent in various programming languages, including PHP, Python, Java, HTML, SQL, JavaScript and C, development frameworks including Django, Symfony2, ADF and Spark, while he develops his skills in the fields of Big Data and Business Analytics.

Dr. Spiros Mouzakitis holds a diploma of Electrical and Computer Engineering from the National Technical University of Athens. In 2009, he successfully defended his Ph.D. dissertation, entitled ‘‘A supportive framework for the capability and interoperability assessment of SMEs for the adoption of B2B integration systems’’. He has nine years of industry experience in conception, analysis and implementation of information technology systems. His current research is focused on decision analysis in the field of eBusiness and eGovernment, as well as optimisation systems and algorithms, decision support systems and enterprise interoperability. Spiros Mouzakitis has worked as a project manager and software engineer/research analyst in more than 15 projects related to the open governmental data, market/impact analysis in the context of ICT technologies, web development and enterprise interoperability. Indicative projects include FP7, FP6 and FP5 research projects co-funded by the European Union, eParticipation projects and evaluation reports of National projects.

Álvaro Sicilia, PhD, Computer Science Engineer. Coordinator of the computer area of the group ARC (Architecture Representation Computation). Member of the group since 1999. Degree in Technical Engineering of Computer Science in the School of Architecture and Engineering La Salle in 2003 and in Higher Engineering in 2008. Later, in 2012, he obtained the title of Master in Research in ICT and its Management. He finished his PhD in 2016. He has worked for two years in the professional sector developing traceability applications. He has experience in the representation of domains through Semantic Web technologies: creation of ontologies, semantic data integration processes, and exploitation of semantic data. He has participated as researcher in diverse national and European projects like BARCODE HOUSING SYSTEM, OIKODOMOS, INTUBE, REPENER, SEMANCO, OPTIMUS, ENERSI and OPTEEMAL. The work done in these projects is based on data management and analysis with a focus on Semantic Web technologies.

Please cite this article in press as: V. Marinakis, et al., From big data to smart energy services: An application for intelligent energy management, Future Generation Computer Systems (2018), https://doi.org/10.1016/j.future.2018.04.062.

V. Marinakis et al. / Future Generation Computer Systems ( M.Arch. Ph.D Leandro Madrazo, Architect. Director of research group ARC Enginyeria i Arquitectura La Salle (www.salleurl.edu/sdr). He is a professor at the School of Architecture La Salle, Universitat Ramon Llull, Barcelona, where he has been head of the ARC Architecture Representation Computation research group since its creation in 1999. He graduated in architecture from the Universitat Politècnica de Catalunya in 1984, and studied later as Fulbright scholar in the Master of Architecture programs of Harvard University and at the University of California Los Angeles, where he obtained the Master’s degree in 1988. From 1990 to 1999 he carried out his teaching and research work at the Department of Architecture and CAAD at ETH Zürich, completing his Ph.D. programme in 1995. His research is concerned with the application of ICTs to the AEC sector, in particular to improve energy efficiency in buildings and cities. He has coordinated the FP7 project SEMANCO, and the Lifelong Learning Programme OIKONET (www.oikonet. org). Previously, he has been coordinator of the European programs OIKODOMOS (www.oikodomos.org) and [email protected] (www.housing21eu.net), coordinator of the Spanish research projects REPENER (arc.housing.salle.url.edu/repener) and BARCODE HOUSING SYSTEM (www.barcodehousing.net). He has published widely in international conferences and journals on the application of ICT to architecture.

)

–

15

Dr. Sgouris Sgouridis is an Associate Professor at Masdar Institute. His current research interests focuses on sociotechnical systems modelling including sustainable transportation systems and sustainable energy systems management. Dr. Sgouridis is Principal Investigator researching ‘Commercial Aviation in a Carbon-Constrained Future’ at Masdar Institute and he is co-leading the development of the Sustainable Bioresource Projects. Prior to his role at Masdar Institute, Dr. Sgouridis worked at governmental and private organisations including the US Department of Transportation, the Port Authority of Thessaloniki and the Hellenic Army. He holds a PhD in Engineering Systems from MIT (2007), an MS in Technology and Policy from MIT (2005), and an MS in Transportation from MIT (2005). Dr. Sgouridis received the Martin Fellowship for Sustainability (2004–2005) and an award for Excellence in Academic Performance from the Chambers of Engineers (1998 and 1999).

Please cite this article in press as: V. Marinakis, et al., From big data to smart energy services: An application for intelligent energy management, Future Generation Computer Systems (2018), https://doi.org/10.1016/j.future.2018.04.062.