Adaptive building energy management with multiple commodities and flexible evolutionary optimization

Renewable Energy xxx (2015) 1e11

Contents lists available at ScienceDirect

Renewable Energy journal homepage: www.elsevier.com/locate/renene

Adaptive building energy management with multiple commodities and flexible evolutionary optimization* Ingo Mauser a, *, Jan Müller b, **, Florian Allerding b, ***, Hartmut Schmeck a, b a b

FZI Research Center for Information Technology, Haid-und-Neu-Str. 10-14, 76131 Karlsruhe, Germany Karlsruhe Institute of Technology, Institute AIFB, Kaiserstraße 89, 76128 Karlsruhe, Germany

a r t i c l e i n f o

a b s t r a c t

Article history: Received 30 March 2015 Received in revised form 12 August 2015 Accepted 1 September 2015 Available online xxx

To enable the efficient utilization of energy carriers and the successful integration of renewable energies into energy systems, building energy management systems (BEMS) are inevitable. In this article, we present a modular BEMS and its customizable architecture that enable a flexible approach towards the optimization of building operation. The system is capable of handling the energy flows in the building and across all energy carriers as well as the interdependencies between devices, while keeping a unitized approach towards devices and the optimization of their operation. Evaluations in realistic scenarios show the ability of the BEMS to increase energy efficiency, self-consumption, and self-sufficiency as well as to reduce energy consumption and costs by an improved scheduling of the devices that considers all energy carriers in buildings as well as their interdependencies. © 2015 Elsevier Ltd. All rights reserved.

Keywords: Energy management system Building operation optimization Battery storage Trigeneration Multi-commodity Evolutionary algorithm

1. Introduction Nowadays, societies rely on ubiquitous and permanent availability of different energy carriers, such as electricity, hot water, and natural gas. Nevertheless, energy systems all over the world are currently in a phase of transition because of economical, political, and environmental reasons. The intermittent generation by renewable energy sources (RES) and the increasing power feed-in by distributed generation (DG) are already leading to problems in power grids, such as voltage problems and overloads of power lines [59]. Additionally, an efficient utilization of energy carriers is getting increasingly complex because generation does not longer follow consumption [40]. To tackle these problems, smart grids offer solutions: They allow for advanced management and optimization and provide the means for flexibility encompassing the grid as well as individual buildings. For instance, demand side management (DSM) is supposed to enable an economically efficient way of responding to intermittent and decentralized energy feed-in from

*

This document is a collaborative effort. * Corresponding author. ** Corresponding author. *** Corresponding author. E-mail addresses: [email protected] (I. Mauser), [email protected] (J. Müller), fl[email protected] (F. Allerding), [email protected] (H. Schmeck).

RES [22,40]. In order to realize DSM and to ensure an efficient utilization of energy, sophisticated energy management systems (EMS) have to be introduced on all levels of smart grids [3,14], in particular in industrial, commercial, and residential buildings. The different setups of devices in buildings comprising heterogeneous devices, e. g., appliances, DG, storage systems, and electric vehicles, call for a flexible approach towards building energy management systems (BEMS). A major contribution of this article is the comprehensive description and evaluation of a BEMS that is capable of optimizing the operation of devices typically found in commercial and residential buildings. This BEMS can be used in bottom-up simulations and in productive systems. New devices can easily be integrated as simulated devices orein real buildingsdbe connected to the system using drivers and then be optimized using a modular, subproblem based approach to optimization, which has been introduced in Ref. [3]. The approach respects interdependencies of the devices and the actual sub-problems for optimization are adapted to the particular global optimization problem at the runtime of the system, as originally described in Ref. [32]. Another major contribution is the integration of future hybrid household appliances and battery storage systems with novel encodings and mechanisms for optimization that combine scheduling and control logic. The capabilities of the BEMS are demonstrated using two scenarios: a smart residential building and a smart

http://dx.doi.org/10.1016/j.renene.2015.09.003 0960-1481/© 2015 Elsevier Ltd. All rights reserved.

Please cite this article in press as: I. Mauser, et al., Adaptive building energy management with multiple commodities and flexible evolutionary optimization, Renewable Energy (2015), http://dx.doi.org/10.1016/j.renene.2015.09.003

2

I. Mauser et al. / Renewable Energy xxx (2015) 1e11

commercial building. Their operation is optimized with respect to energy costs, which often also increases self-sufficiency, self-consumption, and energy efficiency, using data from real buildings. The following Section 2 provides an overview of architectures for complex systems, in particular EMS, and of optimization in such systems. Section 3 outlines the concrete scenarios and motivates the usage of EMS. Section 4 presents the Organic Smart Home, a modular BEMS with a generic architecture and a flexible approach to optimization. In Section 5, we specify the concrete setups for the simulation of the scenarios, before presenting and discussing the results in Section 6. Finally, the article is concluded in Section 7 with a summary and an outlook to further work. 2. Related work There are several architectures for complex systems and for EMS in smart grids as well as approaches to optimization. This chapter outlines such architectures and approaches, showing that none of these is adequate to fulfill the requirement of resembling the multitude of entities in smart grids and their diverse capabilities, requirements, interdependencies, and complexity, while working in simulations and productive EMS in real buildings.

The Flexible Power Application Infrastructure (FPAI) has been developed to exploit load flexibility in power grids by shifting the operation times of appliances and DG [54]. It comprises a centralized control structure for a BEMS called Flexible Power Runtime, which abstracts the flexibility provided by the devices. Nevertheless, the optimization aspect and its implementation are not precisely specified as it is provided by an external optimization service that is not part of FPAI. Similarly, components enabling interactions with external components, such as a distribution grid or an energy market, have been defined, though not yet realized. In Refs. [38,58], an architecture for building automation and management is proposed that is named Open Gateway Energy Management. It uses a so-called bidirectional energy management interface to connect resources, which are abstracted using a graphlike structure, and applications that provide management functionality. Nevertheless, the architecture lacks an integrated approach, as the optimization of individual resources is provided by separate management applications. 2.3. Optimization in energy management systems

The complexity of technical systems is constantly increasing because of a rise of interconnected devices. Typically, complex systems are prone to breakdowns and fatal errors caused by minor disturbances or emergent effects [36]. Smart grids, which are an example for complex systems, are still in development [6,18]: It is essential to design entities in smart grids with keeping in mind the upcoming complexity caused by many interacting entities, e. g., devices, buildings, and grid operators. Architectures and approaches used in the development of smart grids should facilitate appropriate methods for abstraction, optimization, and selfadaptivity of such systems [6]. The theory of Autonomic Computing [27] focuses on fully autonomic computer systems without any later user interaction after design phase. It provides a central design paradigm: the MAPE cycle, which consists of the four steps monitor, analyze, plan, and execute. In Ref. [20], it has been used in a control architecture for smart micro grids. Hierarchy, stigmergy, and collaboration are used to build an architecture with global and local optimization scopes by using rather simple models of power grids and their entities. In general, the MAPE cycle paradigm is suited to cope with complex distributed systems, although the particular control architecture in Ref. [20] lacks important concepts of device abstraction that are essential for productive systems. Organic Computing addresses basic challenges of complex systems in dynamic environments, such as trustworthiness, flexibility, adaptivity, robustness, and effects of emergence [36]. Based on various scenarios in robotics [36], traffic [42], production [46], and energy [3,25], generic system architectures have been developed and evaluated. A particular example of an architecture for complex systems proposed by Organic Computing is the Observer/Controller Architecture (O/C Architecture), which serves as a generic framework comprising various components that are essential for designing systems showing organic behavior, i. e., an adaptive behavior similar to nature [43].

There are several approaches to optimization in energy systems and its various problems. Energy systems and their components may be optimized with respect to highly diverse objectives and on different abstraction levels. Usually, publications focus either on optimization of the technical setup of the system [2,26], or energy markets [56], or balancing groups [25]. Often, they do not respect interdependencies or non-linearities [12,21,26,55], or they perform a scheduling that is only exact to the hour [12,44,55], which is not practicable in concrete productive systems, because of averaging effects that hide load peaks, which have to be handled by a BEMS [49,57]. Often, optimization problems in the domain of EMS have been formulated in linear programming (LP) [35,44], mixed integer linear programming (MILP) [1,10,11,15,23,50], or mixed integer nonlinear programming (MINLP) [4,7,21,47]. Modeling problems from the energy domain usually requires several thousand variables and constraints, even if the problem is reduced to an optimization based on time slots of five [11] or 15 min [10,15,23]. Solving such problems may lead to extensive computational requirements that are neither practicable nor reasonable for BEMS that should run on low-power, energy-saving computers with limited system resources. Heuristic optimization has proved to optimize a wide range of optimization problems efficiently [34]. Advantages of heuristics are their low memory and time requirements. Consequently, evolutionary optimization has been used in the optimization of energy systems. Frequently, evolutionary optimization is used to optimize design parameters of devices in energy systems [2,26,55]. Only a fraction uses heuristics to solve scheduling problems [41]: uses particle swarm optimization to schedule devices in a smart residential building on an hourly basis. In Refs. [45], an evolutionary algorithm is used to schedule district heating and cooling plants. The approach of [48] is similar to the one presented in this article. Their encoding uses a string of integers in the evolutionary algorithm. Every integer determines the starting time of the operation cycle of one device exact to the minute. Nevertheless, the approach is limited to electrical loads and does not respect interdependencies of multiple devices and energy storages.

2.2. Architectures for energy management systems

3. Scenarios and material

First of all, EMS require appropriate architectures that provide the flexibility and modularity to adapt the system to the variety of different entities in energy systems.

This article analyzes two scenarios: a smart residential building and a smart commercial building. Both require the optimization of multiple energy carriers, such as electricity and hot water, with

2.1. General architectures and principles for complex systems

Please cite this article in press as: I. Mauser, et al., Adaptive building energy management with multiple commodities and flexible evolutionary optimization, Renewable Energy (2015), http://dx.doi.org/10.1016/j.renene.2015.09.003

I. Mauser et al. / Renewable Energy xxx (2015) 1e11

respect to cost minimization. They use tariffs, statistical data, and input values for the bottom-up simulations, partially collected in the Energy Smart Home Lab (ESHL) at the Karlsruhe Institute of Technology and the House of Living Labs (HoLL) at the FZI Research Center for Information Technology.

3

Table 1 Specification of the households in the smart residential building scenario. Number of persons Electricity consumption Appliances

2 4 3000 kWh/a 4500 kWh/a Dishwasher, washing machine, tumble dryer, hob, oven Normal (Electricity only with h¼1.00), Hybrid (Gas respective hot water with h¼0.77) 2.5 kWpeak 3.8 kWpeak 3000 kWh/a 4500 kWh/a 3 kWh, h¼0.85 750 L 15 kW Electricity: 0.28 EUR/kWh Natural gas: 0.07 EUR/kWh PV feed-in: 0.10 EUR/kWh CHP feed-in: 0.05 EUR/kWh

Appliance variants

3.1. Smart residential building scenario The smart residential building (see Fig. 1 and Table 1) consists of one household that is equipped with a photovoltaic system (PV), a gas-fired condensing boiler, a hot water storage tank, controllable hybrid appliances, and a battery storage. Hot water is produced by the boiler and then stored in the tank, which decouples production and consumption. The dishwasher, the washing machine and the dryer use either the hot water from the tank or heat up the water using electricity. The hob and the oven use either electricity or natural gas. Starting times of the appliances are simulated according to typical statistical values and with real appliance consumption profiles. 3.1.1. Hybrid appliances Usually, appliances use only a single energy carrier, e. g., electricity, hot water, or gas, in their energy intensive processes, such as the heating phases. By contrast, hybrid appliances use multiple energy carriers (see Fig. 2). One example of a hybrid appliance is a washing machine that can use electricity to heat up the water used in a washing cycle or use hot water of the building's domestic hot water supply system. In this article, the power consumption of the appliances is about 1.3 times higher when using gas or hot water instead of electricity because of additional losses in the supply system and additional heat exchangers [53]. When optimizing the operating times and deciding about which alternative profile to use, the energy consumption, i. e., the consumption profile consisting of different energy carriers, directly relates to the states of other devices, such as the hot water storage tank. This causes interdependencies because multiple devices work on the same storage tank affecting its temperature. Nowadays, hybrid appliances are not widely available and may be seen as futuristic. Nevertheless, many washing machines and dishwashers can already be connected to the domestic hot water system and it is possible to connect dryers using heat exchangers or to combine electricity and gas heating in hobs and ovens [33,53]. 3.1.2. Appliances and appliance usage We assume that households are equipped with five appliances: washing machine, dryer, dishwasher, oven, and hob. The average number of yearly appliance operation cycles per household depends on the number of persons and is given in Table 2. These

PV System Hybrid Space Appliances Heating 42

Other Battery Smart Devices Storage Residential Building Electricity

Hot Water

Hot Water Storage Tank

Gasfired Boiler

Battery storage specifications Hot water storage tank capacity Gas-fired condensing boiler power Tariffs

Power

Normal Profile

Electricity

42

Time Power

Hot Water

Hybrid Profile

Time

Heating Phase

Fig. 2. Hybrid washing machine using either electricity or hot water in its heating phase.

values as well as the starting times of the appliances are simulated according to typical usage based on statistical data from Refs. [19,51e53] (see Fig. 3) and real consumption profiles recorded at our labs. The remaining yearly electricity consumption, i. e., consumption not explained by the five appliances, is simulated using the German standard load profile for households. 3.1.3. Photovoltaic system, battery storage and heating Power generation by the PV is simulated using a profile recorded in Karlsruhe, Germany with a resolution of 1 min. The PV has a maximum power of 2.5 kW in the two and 3.8 kW in the four person household. In the smart residential building scenario, the electricity generation is rescaled to match the electricity consumption of the respective household: For a two and a four person household the production is 3000 kWh/a and 4500 kWh/a, respectively. The battery storage is simulated with a capacity of 3 kWh and an efficiency of h¼0.85. It uses a control strategy that charges the battery whenever the electricity feed-in at the house connection is higher than a given threshold of 0.1 kW, and discharges the battery whenever the consumption at the house connection is higher than 0.1 kW. The heating and domestic hot water demands have been determined using TRNSYS [28]da building simulation tooldwith a model of the ESHL and have been scaled to the respective size.

Table 2 Appliances: average number of operation cycles per household and year.

EMS Communication

Photovoltaic system

Gas-Grid

Fig. 1. Overview of the smart residential building scenario.

Persons per household

1

2

3

4

5

Washing machine Tumble dryer Dishwasher Oven Hob

100 40 120 100 150

180 70 200 150 250

250 100 280 180 360

320 130 350 210 420

360 150 410 240 480

Please cite this article in press as: I. Mauser, et al., Adaptive building energy management with multiple commodities and flexible evolutionary optimization, Renewable Energy (2015), http://dx.doi.org/10.1016/j.renene.2015.09.003

4

I. Mauser et al. / Renewable Energy xxx (2015) 1e11

0.14 Probability

0.12 0.10 0.08

Table 3 Experiments: specifications of the smart commercial building scenario.

Dishwasher Hob Oven Dryer Washing Machine

Adsorption chiller Micro-CHP

0.06 Hot water storage tank

0.04 0.02 0.00 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

Chilled water storage tank

Hour of Day [h] Tariff Fig. 3. Probability density function describing the appliance usage.

Nominal cooling power: Hot water power: Electric active power: Natural gas power: Capacity: Min. temperature: Max. temperature: Capacity: Min. temperature: Max. temperature: Electricity: Natural gas: CHP feed-in:

9 kW 12.5 kW 5.5 kW 20.5 kW 3250 L 57 C 78 C 3000 L 10 C 15 C 0.28 EUR/kWh 0.06 EUR/kWh 0.08 EUR/kWh

3.2. Smart commercial building scenario The smart commercial building scenario (see Fig. 4) resembles the situation in the HoLL [9] and has initially been presented in Ref. [32]: It consists of a commercial building with a meeting room that has to be air-conditioned. This is done by a system that comprises a small combined heat and power plant (micro-CHP), an adsorption chiller (Ad-A/C), as well as storage tanks for hot and chilled water. Such a system is called trigeneration or combined cooling, heat and power. It combines cogeneration, i. e., a micro-CHP, with an Ad-A/C that produces chilled water. Thus, four energy carriers are involved in this process: electricity, hot water, chilled water, and natural gas. Additionally, multiple devices work on the same storage tanks and their consumption, production, and states directly relate to at least one other device. Therefore, devices of a trigeneration system have to be optimized concurrently. In the present scenario, the chilled water is used to air-condition the meeting room. The Ad-A/C is mainly powered by hot water that is generated by the micro-CHP. Generation of electricity and production as well as consumption of chilled respective hot water are decoupled by the storage tanks. The devices have been modeled according to technical specifications of real devices (see Table 3). The efficiency of the Ad-A/C, which depends on the tank and outdoor temperatures, has been interpolated from the data sheet of a real Ad-A/C [24]. The standing loss of the storage tanks Ploss in kW depends on the current tank temperature Ttank in  C, the ambient temperature Tambient, and the thermal transmittance c of the storage tanks and their insulation:

Ploss ðTtank Þ ¼ c$ðTtank  Tambient Þ: The ambient temperature Tambient has been set to 20 C. Based on

measurements in the HoLL, the thermal transmittance has been set to c¼0.04 kW/ C for the chilled water and c¼0.01 kW/ C for the hot water storage tank. Appointments in the meeting room, which trigger an air-conditioning request, have been extracted from room reservations in the Microsoft Exchange Calendar for a real meeting room in the HoLL. Four weeks in July 2013 have been selected for the simulations and provide recorded real outdoor temperatures. Cooling demand Pdemand in kW as a function of the outdoor temperature Toutdoor in  C is calculated using the following empirical formula that is based on measurements in the real building. An outdoor temperature above 21.88 C leads to a linearly increasing cooling demand:


 kW Pdemand ðToutdoor Þ ¼ max 0 kW; 0:44  $ðToutdoor  21:88+ CÞ : C

4. Energy management system: organic smart home The architectures and systems described in Section 2 show important concepts of abstraction and simulation. Nevertheless, none is capable of optimizing the multiplicity of different devices in heterogeneous scenarios, with respect to individual needs of the users and to grid constraints, anddmost importantlydrealizing this in a system that works in simulations as well as in productive systems in real buildings. Productive systems require an appropriate architecture that provides flexibility and modularity to optimization and adapts to different scenarios, such as industrial, commercial, and residential buildings as well as single devices.

Heat Exchanger Hot Water Meeting Room

Space Heating Demand

Other Devices

Electricity

Chilled Water Chilled Water Storage Tank

AdA/C

Smart Commercial Building

Hot Water Storage Tank

Biogas Factory

EMS micro CHP

Gas-Grid Communication

Fig. 4. Overview of the smart commercial building scenario.

Please cite this article in press as: I. Mauser, et al., Adaptive building energy management with multiple commodities and flexible evolutionary optimization, Renewable Energy (2015), http://dx.doi.org/10.1016/j.renene.2015.09.003

I. Mauser et al. / Renewable Energy xxx (2015) 1e11

5

4.1. Organic smart home: introduction The Organic Smart Home1 (OSH) [3,5] is a BEMS that has been developed for buildings with controllable appliances, electric vehicles, and distributed generation. It has been deployed in real buildings and evaluated in multiple evaluation phases in our labs [39]. Major advantage of the OSH is its usability in both, productive systems in real buildings and simulated buildings with different sets of devices in diverse scenarios [3], enabling the testing of control and optimization functionality in simulations before applying them to productive systems. The OSH has already been used in scenarios optimizing the operation of appliances [3,30], heat-pumps [29], trigeneration [32], and electric vehicles [37]. The broad applicability of BEMS in many different scenarios is inevitable for the efficient integration of RES and DG into energy systems. Therefore, it is important to be able to control and optimize buildings with highly diverse scenarios and devices and to consider the buildings with all their technical systems, energy production and consumption, no matter whether it is electricity or another energy carrier. Consequently, the OSH considers not only electricity in terms of active power, but also electricity in terms of reactive power, natural gas, hot and chilled water consumption, and emissions of greenhouse gases. Usually, systems for building simulation, which focus on thermal energy, use time steps in scale of minutes [13]. The additional integration of electricity leads to the requirement that the optimization with respect to tariffs, payment schemes, and load limitations has to regard time steps as short as possible to take account of short-time consumption and production peaks as well as the actual self-consumption, feed-in, and external supply of electricity [57]. Therefore, the OSH works with time steps on a second to second basis when simulating, while respecting interdependencies and non-linearities of devices [32]. As described in Section 2.3, BEMS should run on low-power computers having limited system resources. In addition, the execution time of the optimization algorithm is crucial because frequent rescheduling is quite likely and a quick reaction on user interaction is desirable [39]. Therefore, we use a meta-heuristicdan evolutionary algorithm (EA)dwith dynamic formulation of the problem instances at runtime of the system. Additionally, solving the optimization problem ex ante is only possible when having complete information about future energy flows. Thus, generating approximate solutions by a heuristic that allows for frequent rescheduling in varying setups promises to be of better use for productive energy management. 4.2. General architecture of the organic smart home The OSH is designed according to the O/C Architecture of Organic Computing (see Section 2.1), which has been applied in a hierarchical way to handle the complexity of such systems and to realize a flexible, modular approach (see Fig. 5). Observation, control, and optimization are done by so-called Observer/Controllerunits (O/C-units), which form regulatory feedback mechanisms with observation, analysis, prediction, learning, and optimization capabilities [36]. The first O/C Layer consists of O/C-units that are dedicated to every device which is included in the EMS and handles their specific management. The second O/C Layer is responsible for the integrated optimization of the complete building. O/C-units use sensors and actuators to observe and control a socalled System under Observation and Control (SuOC). In the present case, the SuOC is a smart residential or commercial building. Every

1

http://www.organicsmarthome.com.

External Entities

User

energy data

goals

signals

2nd O/C Layer: Integrated Optimization Layer

Observer (O)

Optimization Abstraction Layer

Interdependent Interdependent Problem Part Problem Part

1st O/C Layer: Device Management Layer

Controller (C)

Integrated Building Optimization

O

C

O

C

Interdependent Problem Part

O

C

Device Mgmt.

Device Mgmt.

Device Mgmt.

Device Driver

Device Driver

Device Driver

Hardware Abstraction Layer System under Observation and Control

Fig. 5. Overview of the organic smart home and its O/C architecture.

O/C-unit comprises an Observer and a Controller that structure its features. In order to abstract the SuOC, there are two additional layers: the Hardware Abstraction Layer and the Optimization Abstraction Layer. The Hardware Abstraction Layer realizes the abstraction from distinct devices, protocols, and communication media of the components to generic O/C-units in the first O/C Layer by using device-specific drivers [3]. The Optimization Abstraction Layer utilizes Interdependent Problem Parts (IPP) to integrate all devices into a single global optimization problem. Signals and data from external entities, such as energy tariffs and weather forecasts, and goals of the user are incorporated by specific interfaces and communication drivers. The interface to the human user is called Energy Management Panel and is used to inform the user about the current energy situation in the building and to obtain the user preferences and goals [8]. The energy tariffs and signals from external entities, i. e., utilities and grid operators, mirror the state of the market and the grid, facilitating DSM. 4.3. Dynamic optimization problem and optimization algorithm The modularity and flexibility of the BEMS is enabled by the IPPs, which are an extended version of the Problem Parts presented in Ref. [3] and are more closely described in Section 4.5. The IPPs represent sub-problems of the optimization. They are provided by the O/C-units of the first O/C Layer, and communicated via the Optimization Abstraction Layer to the global O/C-unit at the second O/C Layer (see Fig. 5). Every IPP contains all information that is necessary to optimize the corresponding device with respect to its technical specifications, possible control sequences, and interdependencies with other devices. The IPPs are used in the optimization algorithm to formulate the optimization problem and to create solution candidates (see Fig. 6). The joint evaluation of the solution candidates with the actual interdependencies of the devices' energy flows is done by the Energy Simulation Core [32]. The resulting load profiles are combined to the expected total future load profile and assessed by the fitness function with respect to signals, tariffs, and user preferences. Hence, the optimization problem is not stated in a closed form, but dynamically compiled from IPPs at runtime. This enables the OSH to manage fundamentally differing buildings with different sets of devices by using just another set of device drivers and O/C-

Please cite this article in press as: I. Mauser, et al., Adaptive building energy management with multiple commodities and flexible evolutionary optimization, Renewable Energy (2015), http://dx.doi.org/10.1016/j.renene.2015.09.003

6

I. Mauser et al. / Renewable Energy xxx (2015) 1e11

Fig. 6. Overview of the optimization process with energy simulation core and interdependent problem parts.

units providing the respective IPPs. The IPPs are constructed periodically and communicated at runtime of the OSH, showing the statuses of the devices and abstract representations of their optimization potentials, i. e., degree of freedom (see Section 4.5), and enabling an optimization approach that is similar to “plug-andplay”. The compiled dynamic optimization problem is solved using an EA [3], which optimizes the IPPs. Sub-problems, i. e., the IPPs, are included into the optimization using bit strings that encode the future behavior of the devices. For instance, a bit string may encode the time until a deferrable, i. e., shiftable, appliance will be started (see Section 4.5.1) or the periods when a CHP will be running (see Section 4.5.4). The EA, which is based on a refined version of the generic Genetic Algorithm from the jMetal framework [16,17], operates on the concatenated string of all bit strings. 4.4. Energy simulation core The Energy Simulation Core (ESC), which is depicted in Fig. 7, simulates the local energy flows in the electricity and thermal grid, while respecting the interdependencies of devices and concurrently keeping a modular approach to the optimization. The ESC is used twice in the OSH: Firstly, in the calculation of energy flows in every simulated time step, which are actually measured by metering devices and sensors in productive systems, and, secondly, in the simulation of future energy flows in the optimization process. In the optimization process, the iterative calculation of energy flows over the optimization horizon is necessary to determine the consumption when multiple devices are interconnected and

interdependent [32]. For instance, the operation of thermal devices depends on the current state of the overall system, e. g., indoor, outdoor, and storage tank temperatures. Additionally, characteristics of devices often include non-linear dependencies: For example, the required thermal energy in terms of hot water consumption is non-linear with respect to the generated thermal energy in terms of chilled water. The ESC handles the information exchange between the simulated devices, i. e., consumption and production as well as additional information, such as temperatures or voltages. Thus, devices can observe other devices and react on their status or be simulated with respect to their state. The ESC consists of two main components: the Electrical Simulation and the Thermal Simulation, each handling a different set of energy carriers. These simulate their respective local grid and its distinct energy carriers. The local electricity grid consists of the wiring, i. e., the electrical connections between all devices consuming or producing electricity in a building, whereas the local thermal grid contains all information about the physical interconnections, e. g., pipes between the devices. Energy carriers are divided into commodities, e. g., active and reactive power, heating and domestic hot water. They are further distinguished into virtual commodities, which determine whether they are generated by, e. g., a PV or a micro-CHP, to take different feed-in compensations into account. 4.5. Interdependent problem parts and encodings for devices Every IPP represents a device or a system consisting of joint

Please cite this article in press as: I. Mauser, et al., Adaptive building energy management with multiple commodities and flexible evolutionary optimization, Renewable Energy (2015), http://dx.doi.org/10.1016/j.renene.2015.09.003

I. Mauser et al. / Renewable Energy xxx (2015) 1e11

Determination of Load Profiles

Energy Simulation Core

7

Active Power

Iterate Optimization Horizon

IPP SpaceCooling

Reactive Power

Electrical Simulation Voltage U, Power P

…

Chilled Water Power

AdA/C

Thermal Simulation Local Thermal Grid

Hot Water Power

Temperature T, Water Power P

…

IPP HotWater

Initialization

IPP Ad-A/C

Joint Evaluation

Local Electrical Grid IPP ChilledWater

Ad A/ C

Resulting Load Profiles of all Devices

Fig. 7. Determination of load profiles using interdependent problem parts and the energy simulation core.

Energy-related Degree of Freedom (EDoF) alternative multivalent / univalent profiles hybrid

devices, e. g., the space heating system. It provides an appropriate physical model regarding the technical specifications and capabilities, the optimization model that encodes the degree of freedom when being optimized, and the control sequences that can be sent to the device. Devices and systems have distinct capabilities in terms of their optimization potential and possible load profiles. The optimization qualifications of appliances with operation cycles can been classified with respect to the flexibilization of the energy consumption in temporal or energy-related terms (see Fig. 8). The former is called Temporal Degree of Freedom (TDoF) and states whether an appliance is deferrable or even interruptible. The latter is called Energyrelated Degree of Freedom (EDoF) and defines whether an appliance has alternative profiles for the same operation cycle or may even change to another commodity [30]. These flexibilities of appliances in the optimization require appropriate encodings for the EA. Devices, such as micro-CHPs or battery storages, have to be optimized in their operating times and modes while respecting their technical constraints and thresholds. For instance, when having an oversupply of electricity by DG, it has to be decided whether to feed the electricity back into the grid or to charge a battery storage. Appropriate bit string encodings for the different devices to be used in the EA are presented in the following sub-sections. They can be combined to a comprehensive bit string that represents the

extended energetically qualified

extended energetically and temporally qualified

fully qualified

energetically qualified

energetically and temporally qualified

energetically and extended temporally qualified

not qualified

temporally qualified

extended temporally qualified

non-deferrable

deferrable

interruptible

Temporal Degree of Freedom (TDoF) Fig. 8. Qualification for optimization with respect to degree of freedom.

energy management problem, and thus enabling the generic, modular optimization. The length of bit strings depends on the devices and the modeling of their search space. It ranges between a few dozen bits in scenarios comprising only appliances and about 1000 bits when additionally optimizing a CHP with a time horizon of 24 h. To take as many different encodings as possible into account while having a homogeneous representation, the representation as a bit string is chosen. This way, some parts of the full bit string represent control sequences, others represent parameter settings, and still others represent time periods devices can be deferred or interrupted in their operation.

4.5.1. Interdependent problem parts of devices with temporal flexibility The IPP of a device with temporal flexibility contains the expected load profile and an appropriate encoding. In Ref. [30], we presented an encoding for devices that can be deferred, e. g., dishwashers, washing machines, and tumble dryers. The maximum delay of the device operation cycle is usually defined by the user, who sets a deadline when the device has to have finished its operation cycle. The period between the release of the device and its deadline, reduced by its operation time, is the TDoF, which is encoded in a bit string. Some of these devices may also be interrupted at certain predefined points in their operation cycle. Thus, the IPPs have to contain the expected load profiles of the device with the possible points of interruption and an adapted encoding that splits the TDoF up between the initial delay and the interruptions as presented in Ref. [30].

4.5.2. Interdependent problem parts of hybrid devices or devices with alternative profiles Devices may also offer alternative load profiles for the same operation cycle or hybrid profiles that allow for a shift of energyconsumption from one commodity to another. The former is realized by, e. g., reducing the power consumption while extending the duration of the load profile. The latter are hybrid devices that are able to utilize different commodities. This offers extensive opportunities to optimize the energy flows in a building by exploiting the EDoF of the devices. A rather simple encoding that enumerates the profiles and allows for their selection in the optimization has been presented in Ref. [30].

Please cite this article in press as: I. Mauser, et al., Adaptive building energy management with multiple commodities and flexible evolutionary optimization, Renewable Energy (2015), http://dx.doi.org/10.1016/j.renene.2015.09.003

8

I. Mauser et al. / Renewable Energy xxx (2015) 1e11

Part n = 1

n=2

n=3

n=4

interpreted by an automaton, i. e., translated into scheduled operating times, and is closely described in Ref. [30].

n=5

Maximum Duration

5. Exemplary experimental setups and simulations

Profile i = 1

Power P

EDoF

i=2

To demonstrate the capabilities of the BEMS, distinct experiments have been performed for both scenarios, which have been introduced in Section 3. The experiments were performed as simulations to ensure comparability and reproducibility. 5.1. Smart residential building scenario

i=3 Earliest Starting Time

TDoF

Latest Starting Time

Time t

Fig. 9. Exemplary load profile alternatives for interruptible hybrid devices using two energy carriers (yellow and red) with minimum and maximum durations of the interruptions and phases of the operation cycle. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.).

4.5.3. Enhanced encoding for interruptible hybrid devices The previous encodings are combined into an integrated encoding that supports the optimization of all devices visualized in Fig. 8 while also respecting minimum and maximum durations of interruptions and phases of the operation cycle (see Fig. 9). To support limited durations of the phases and interruptions in between them, the partitions of the TDoF are calculated with respect to these values, meaning that they are added to the minimum values and limited to the maximum values. Thus, the length lj of the bit string of device j having imax alternative profiles with n parts, i. e., phases and interruptions, of variable duration is:

      lj ¼ n$ log2 TDoFj þ log2 imax;j That way, the differences between phases of the operation cycle and interruptions as introduced in Ref. [30] are removed, which enables an easier implementation of the IPPs.

4.5.4. Interdependent problem parts of the CHP and the adsorption chiller CHPs and Ad-A/Cs can be optimized using signals that switch them on or off. Their IPPs implement oneoff control, ensuring that the storage tank temperatures remain within the thresholds. This thermal management, which is independent of the bit string in the optimization, ensures that every solution candidate is valid with respect to the constraints: The device is forced to be switched on or off, even if the control sequence determined by the bit string in the optimization would originally lead to violations of thresholds. The encoding utilized by the IPPs uses a sequence of bits that is

In this scenario, several combinations of deferrability, operation mode, and availability of battery storage have been chosen (see Table 4). In addition, every experiment has been carried out using a recorded PV generation profile from January and July 2013, respectively, that has been scaled to yearly power generation of 3000 kWh for the 2 person respective 4500 kWh for the 4 person household. All experiments simulated 28 days of the respective smart residential building over 20 runs with different random seeds, which lead to different user behavior. The EA of the OSH applied binary tournament selection, single-point-crossover with two offspring, and bit-flip-mutation using an elitist (m,l)-strategy with a rank based survivor selection. Parameters of the operators have been calibrated manually and set to a crossover probability of 0.7, a mutation probability of 0.01, and 50 generations with 100 individuals. The optimization is triggered whenever an appliance is switched on by the user, having a TDoF of 6 h that determines the optimization horizon. This leads to a variable amount of optimizations per run. 5.2. Smart commercial building scenario The four experiments of the smart commercial building scenario are combinations of the controllable and non-controllable Ad-A/C and micro-CHP, which have been carried out using 10 random seeds for each experiment (see Table 5). All experiments simulated four weeks in July 2013 with real, recorded outdoor temperatures from Karlsruhe, Germany. The EA used the same setup as in the smart residential building scenario, whereas the manually calibrated parameters of the operators (see Fig. 10) have been set to a crossover probability of 0.7, a mutation probability of 0.005, and 200 generations with 100 individuals. The optimization by the EA with an optimization horizon of 18 h has been triggered at least every 4 h or if a temperature threshold of either the hot or the chilled water storage tank has been violated. 6. Results and discussion To asses the performance of the presented BEMS, we use the total costs, the self-consumption as well as the self-sufficiency. The total costs are defined as:

Total costs ¼ Table 4 Smart residential building scenario: experiments with device combinations. Experiment

Appliance deferrability

Appliance operation mode

Battery storage

SH-1 SH-2 SH-3 SH-4 SH-5 SH-6

non-deferrable deferrable deferrable non-deferrable deferrable deferrable

standard standard hybrid standard standard hybrid

no no no yes yes yes

X i2I

Econ;i $pi 

X

Egen;j $pj :

j2J

Where Econ,i represents the energy consumption in kWh of commodity i of the set I of all commodities relevant in consumption, Egen,j represents the energy generation in kWh of commodity j of the set J of all commodities that are relevant. pi and pj define the feed-in compensations of the commodities i and j in Euro/kWh. Self-consumption and self-sufficiency are defined as:

Please cite this article in press as: I. Mauser, et al., Adaptive building energy management with multiple commodities and flexible evolutionary optimization, Renewable Energy (2015), http://dx.doi.org/10.1016/j.renene.2015.09.003

I. Mauser et al. / Renewable Energy xxx (2015) 1e11

self­consumption ¼

self­sufficiency ¼

total generated energy  total energy feed­in ; total generated energy

total generated energy  total energy feed­in : total consumed energy

6.1. Results: smart residential building scenario In January, the self-consumption is higher than in July, because the low maximum power of the PV in wintertime and the low total generated energy by the PV lead to a low energy feed-in (see Table 6). Subsequently, the higher total generated energy in July leads to a higher self-sufficiency and lower total costs. The total costs even turn negative when the high PV feed-in leads to compensations that are higher than the energy costs. In the assessment of the effects of the optimization, the experiment SH-1 is the non-

Table 5 Smart commercial building scenario: Combinations of Interdependent Problem Parts. Experiment

IPP of Ad-A/C

IPP of micro-CHP

SB-1 SB-2 SB-3 SB-4

non-controllable non-controllable controllable controllable

non-controllable controllable non-controllable controllable

100.00 95.00 90.00 85.00 80.00 75.00 70.00 65.00

MR=0.01 MR=0.005 MR=0.001

1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 11000 12000 13000 14000 15000 16000 17000 18000 19000 20000 21000 22000 23000 24000 25000

Total Costs [EUR]

9

# Evaluations

optimized reference experiment without and SH-4 the one with battery storage. The optimization of deferrable appliances hardly improves the total costs in both cases, with (SH-5) and without battery storage (SH-2). In contrast, the optimization of deferrable hybrid appliances (SH-3 and SH-6) yields improvements in the total costs (see Figs. 11 and 12), self-consumption, and self-sufficiency (see Table 6). The experiments in January show that the battery storage decreases the total costs slightly, although it allows for a self-consumption of almost 100% (see Table 6). In July, the battery storage decreases the total costs significantly, increases the self-consumption, and allows for a self-sufficiency of almost 100% (see Table 6). Comparing the experiments performed with two (see Figs. 11a and 12a) respective four (see Figs. 11b and 12b) persons, the self-consumption and the self-sufficiency are similar (see Table 6), while the total costs are different.

6.2. Results: smart commercial building scenario The experiment SB-1, i. e., without optimization, has been used as reference to quantify the performance of the optimized experiments. Results of the experiments are shown in Table 7 and Fig. 13. These show a decrease of the average total monthly costs by up to 15.6% (see SB-4). The cost reduction is gained through a higher efficiency of the whole system, which is achieved by the optimized operation of both devices. When optimizing only one of the two devices, the optimization of the Ad-A/C (SB-3) leads to better results than that of the micro-CHP (SB-2), although a higher volatility of the achieved total costs is observed in SB-3 (see Fig. 13). The latter can be explained by a surplus of hot and chilled water that remains in the storage tanks at the end of the simulation. The costs for the stored energy are already included in the total costs, whereas the usage did not occur during the experiment. The total costs decrease significantly with increasing flexibility in the energy consumption, which is consistent with the assumptions presented in Section 1.

Fig. 10. Smart commercial building scenario: total energy costs with different numbers of evaluations and mutation rates (MR), based on one run per parameter setting.

6.3. Subsumption of the results Table 6 Smart residential building scenario: mean values of 20 simulation runs per setup. Selfconsumption [%]

Self-sufficiency [%]

4 Pers.

2 Pers.

4 Pers.

2 Pers.

4 Pers.

245.3 245.3 223.9 243.5 243.6 222.4 4.0 3.4 3.5 13.7 14.0 16.3

79.3 79.7 77.8 99.0 99.0 98.9 19.8 20.2 16.7 33.9 34.0 28.5

80.0 80.3 77.1 99.1 99.2 99.0 20.5 21.0 16.1 33.6 33.9 28.9

9.1 9.2 11.7 11.4 11.4 14.8 54.9 56.1 60.7 93.4 93.7 93.7

9.2 9.2 12.1 11.4 11.4 15.4 56.8 58.0 61.0 92.8 93.5 97.4

Month

Experiment

Total costs [EUR] 2 Pers.

January

SH-1 SH-2 SH-3 SH-4 SH-5 SH-6 SH-1 SH-2 SH-3 SH-4 SH-5 SH-6

146.4 146.4 133.5 145.2 145.3 132.5 3.4 3.1 1.7 9.1 9.1 9.5

July

The BEMS presented in this article performs as expected and its optimization capabilities have been proven for scenarios comprising different devices, such as deferrable and hybrid appliances, PVs, battery storages, and trigeneration systems. This proves the adaptability of the system and its optimization to fundamentally different setups, which cannot be achieved by other approaches. Although a comparison to other approaches in terms of performance is difficult, similar cost reductions when optimizing trigeneration have been achieved in Ref. [12]. Currently, the parameters of the optimization and the control strategy of the battery storage are fixed. Nevertheless, the performance of the BEMS with respect to the total costs, the selfconsumption, and the self-sufficiency in the scenarios can be further improved by the use of more refined optimization parameters [31] as well as improved control strategies for the battery storage.

Please cite this article in press as: I. Mauser, et al., Adaptive building energy management with multiple commodities and flexible evolutionary optimization, Renewable Energy (2015), http://dx.doi.org/10.1016/j.renene.2015.09.003

10

I. Mauser et al. / Renewable Energy xxx (2015) 1e11

Fig. 11. Smart residential building scenario: total costs in January.

Fig. 12. Smart residential building scenario: total costs in July.

Table 7 Smart commercial building scenario: statistical values of the total costs and the improvement over the non-optimized experiment SB-1. Experiment

Abs. electricity costs [EUR]

SB-1 SB-2 SB-3 SB-4

Improvement wrt. SB-1 [%]

Avg.

Max.

sn

Min.

Avg.

Max.

sn

82.24 72.66 69.69 68.53

82.24 74.10 73.04 69.39

82.24 75.04 77.34 70.46

0.00 0.54 2.13 0.49

e 8.75 5.94 14.31

e 9.89 11.18 15.63

e 11.64 15.25 16.67

e 0.65 2.59 0.60

7. Summary and outlook To enable a successful integration of renewable energies into energy systems and an efficient utilization of energy carriers, the optimization of buildings and their devices by suitable energy management systems is inevitable. This article presents an adaptive building energy management system and its architecture that enables a flexible approach towards the optimization of building operation in simulations and productive systems. Though having a

Total costs [EUR]

t-test p-value

Min.

80

75

70 SB-1

SB-2

SB-3

SB-4

Experiment Fig. 13. Smart commercial building scenario: boxplot of the total costs in the experiments system.

e 0.000 0.000 0.000

modular, unitized approach towards devices and the optimization of their operation, it is capable of optimizing energy flows across all energy carriers while respecting interdependencies between devices successfully. Evaluations in realistic smart residential and commercial building scenarios that use data from real buildings show the ability of the building energy management system to reduce energy costs by increasing energy efficiency, flexibility, self-consumption, and self-sufficiency through an optimization of the devices in a holistic approach that incorporates all energy carriers. Future work will evaluate the performance of the chosen evolutionary algorithm in comparison to other optimization algorithms. The optimization of the battery storage will be extended by adding the control logic of the battery storage and its parameters to the operational optimization. This will enable an unprecedented integrated optimization process in building energy management that includes scheduling, parameterization, and control logic. The simulated results will then be verified with data and results from corresponding productive systems in real buildings that are equipped with battery storage systems. Additionally, the interdependent optimization of all energy carriers increases flexibilities and thus the qualification of devices for optimization. This will be analyzed in more detail by combining hybrid appliances with an electric insert heating element in the hot water storage tank.

Please cite this article in press as: I. Mauser, et al., Adaptive building energy management with multiple commodities and flexible evolutionary optimization, Renewable Energy (2015), http://dx.doi.org/10.1016/j.renene.2015.09.003

I. Mauser et al. / Renewable Energy xxx (2015) 1e11

References [1] S. Abras, S. Ploix, S. Pesty, M. Jacomino, A multi-agent home automation system for power management, in: J.A. Cetto, J.-L. Ferrier, J.M. Costa dias Pereira, J. Filipe (Eds.), Informatics in Control Automation and Robotics, Lecture Notes Electrical Engineering, vol. 15, Springer, Berlin Heidelberg, 2008, pp. 59e68. [2] P. Ahmadi, M.A. Rosen, I. Dincer, Multi-objective exergy-based optimization of a polygeneration energy system using an evolutionary algorithm, Energy 46 (1) (2012) 21e31. [3] F. Allerding, I. Mauser, H. Schmeck, Customizable energy management in smart buildings using evolutionary algorithms, in: A.I. Esparcia-Alçzar, A.M. Mora (Eds.), Applications of Evolutionary Computation, Lecture Notes in Computer Science, Springer, Berlin Heidelberg, 2014, pp. 153e164. [4] F. Allerding, M. Premm, P. Shukla, H. Schmeck, Electrical load management in smart homes using evolutionary algorithms, in: J.-K. Hao, M. Middendorf (Eds.), EvoCOP 2012, LNCS, vol. 7245, Springer, Berlin Heidelberg, 2012, pp. 99e110. [5] F. Allerding, H. Schmeck, Organic smart home: architecture for energy management in intelligent buildings, in: Proceedings of the 2011 Workshop on Organic Computing, ACM, 2011, pp. 67e76. [6] S. Amin, B. Wollenberg, Toward a smart grid: power delivery for the 21st century, IEEE Power Energy Mag. 3 (5) (Sept. 2005) 34e41. [7] C. Babu, S. Ashok, Peak load management in electrolytic process industries, IEEE Trans. Power Syst. 23 (2) (2008) 399e405. [8] B. Becker, A. Kellerer, H. Schmeck, User interaction interface for energy management in smart homes, in: Innovative Smart Grid Technologies (ISGT), 2012 IEEE PES, IEEE, 2012, pp. 1e8. [9] B. Becker, F. Kern, M. Loesch, I. Mauser, H. Schmeck, Building energy management in the FZI house of living labs, in: D.A.CH. Energieinformatik 2015, 2015 in press. ~ izares, K. Bhattacharya, [10] M.C. Bozchalui, S.A. Hashmi, H. Hassen, C.A. Can Optimal operation of residential energy hubs in smart grids, IEEE Trans. Smart Grid 3 (4) (2012) 1755e1766. [11] Z. Chen, L. Wu, Y. Fu, Real-time price-based demand response management for residential appliances via stochastic optimization and robust optimization, IEEE Trans. Smart Grid 3 (4) (Dec 2012) 1822e1831. [12] G. Chicco, P. Mancarella, Matrix modelling of small-scale trigeneration systems and application to operational optimization, Energy 34 (3) (2009) 261e273. [13] D.B. Crawley, J.W. Hand, M. Kummert, B.T. Griffith, Contrasting the capabilities of building energy performance simulation programs, Build. Environ. 43 (4) (2008) 661e673. [14] S. Dawson-Haggerty, A. Krioukov, J. Taneja, S. Karandikar, G. Fierro, N. Kitaev, D.E. Culler, Boss: building operating system services, in: NSDI, vol. 13, 2013, pp. 443e458. [15] A. Di Giorgio, L. Pimpinella, An event driven smart home controller enabling consumer economic saving and automated demand side management, Appl. Energy 96 (2012) 92e103. [16] J. Durillo, A. Nebro, E. Alba, The jMetal framework for multi-objective optimization: design and architecture, in: 2010 IEEE Congress on Evolutionary Computation (CEC). Barcelona, Spain, July 2010, pp. 4138e4325. [17] J.J. Durillo, A.J. Nebro, jMetal: a java framework for multi-objective optimization, Adv. Eng. Softw. 42 (10) (2011) 760e771. [18] H. Farhangi, The path of the smart grid, IEEE Power Energy Mag. 8 (1) (January 2010) 18e28. [19] Federal Statistical Office (Destatis), Wirtschaftsrechnungen e Einkommens€hlten und Verbrauchsstichprobe , Ausstattung privater haushalte mit ausgewa gebrauchsgütern, 2013, 2014. Fachserie 15, Heft 1. [20] S. Frey, A. Diaconescu, D. Menga, I. Demeure, A holonic control architecture for a heterogeneous multi-objective smart micro-grid, in: 2013 IEEE 7th International Conference on Self-adaptive and Self-organizing Systems (SASO), Sept 2013, pp. 21e30. [21] M. Geidl, G. Andersson, Optimal power flow of multiple energy carriers, IEEE Trans. Power Syst. 22 (1) (2007) 145e155. [22] S. Gottwalt, W. Ketter, C. Block, J. Collins, C. Weinhardt, Demand side management e a simulation of household behavior under variable prices, Energy Policy 39 (12) (2011) 8163e8174. [23] D.L. Ha, H. Joumaa, S. Ploix, M. Jacomino, An optimal approach for electrical management problem in dwellings, Energy Build. 45 (2012) 1e14. [24] InvenSor, Adsorption Chiller InvenSor LTC 09-Data Sheet, May 2010. [25] A. Kamper, A. Eßer, Strategies for decentralised balancing power, in: A. Lewis, S. Mostaghim, M. Randall (Eds.), Biologically-inspired Optimisation Methods, Studies in Computational Intelligence, vol. 210, Springer, 2009, pp. 261e289. [26] K. Kavvadias, Z. Maroulis, Multi-objective optimization of a trigeneration plant, Energy Policy 38 (2) (2010) 945e954. [27] J.O. Kephart, D.M. Chess, The vision of autonomic computing, Computer 36 (1) (2003) 41e50. [28] S.A. Klein, et al., TRNSYS 17: A Transient System Simulation Program, Solar Energy Laboratory, University of Wisconsin, Madison, USA, 2010. [29] M. Loesch, D. Hufnagel, S. Steuer, T. Fassnacht, H. Schmeck, Demand side management in smart buildings by intelligent scheduling of heat pumps, in: 2014 IEEE International Conference on Intelligent Energy and Power Systems (IEPS), IEEE, 2014, pp. 1e6. [30] I. Mauser, M. Dorscheid, F. Allerding, H. Schmeck, Encodings for evolutionary algorithms in smart buildings with energy management systems, in: 2014 IEEE

11

Congress on Evolutionary Computation (CEC), IEEE, 2014a, pp. 2361e2366. [31] I. Mauser, M. Dorscheid, H. Schmeck, Run-time parameter selection and tuning for energy optimization algorithms, in: Parallel Problem Solving from NatureePPSN, XIII, Springer, 2014b, pp. 80e89. [32] I. Mauser, J. Feder, J. Müller, H. Schmeck, Evolutionary optimization of smart buildings with interdependent devices, in: A.M. Mora, G. Squillero (Eds.), Applications of Evolutionary Computation: 18th European Conference, EvoApplications 2015, Copenhagen, Denmark, April 8-10, Lecture Notes in Computer Science, vol. 9028, Springer International Publishing, 2015a, pp. 239e251. [33] I. Mauser, H. Schmeck, U. Schaumann, Optimization of hybrid appliances in future households, in: Proceedings of International ETG Congress 2015: Die Energiewende e Blueprint for the New Energy Age, VDE, 2015 in press. [34] K. Mesghouni, S. Hammadi, P. Borne, Evolutionary algorithms for job-shop scheduling, Int. J. Appl. Math. Comput. Sci. 14 (1) (2004) 91e104. [35] Mohsenian-Rad, A. Leon-Garcia, Optimal residential load control with price prediction in real-time electricity pricing environments, Trans. Smart Grid 1 (2) (2010) 120e133. [36] C. Müller-Schloer, H. Schmeck, T. Ungerer (Eds.), Organic Computing e A Paradigm Shift for Complex Systems, Springer, 2011. [37] M. Mültin, F. Allerding, H. Schmeck, Integration of electric vehicles in smart homes e an ICT-based solution for V2G scenarios, in: Proceedings of the 2012 IEEE PES Innovative Smart Grid Technolgies Conference, IEEE Power & Energy Society, 2012, pp. 1e8. [38] D. Nestle, J. Ringelstein, H. Waldschmidt, Open energy gateway architecture for customers in the distribution grid, it e Inf. Technol. 52 (2) (2010) 83e88. [39] A.-G. Paetz, T. Kaschub, P. Jochem, W. Fichtner, Load-shifting potentials in households including electric mobility e a comparison of user behaviour with modelling results, in: 2013 10th International Conference on the European Energy Market (EEM), May 2013, pp. 1e7. [40] P. Palensky, D. Dietrich, Demand side management: demand response, intelligent energy systems, and smart loads, IEEE Trans. Ind. Inf. 7 (3) (2011) 381e388. [41] M. Pedrasa, T. Spooner, I. MacGill, Coordinated scheduling of residential distributed energy resources to optimize smart home energy services, IEEE Trans. Smart Grid 1 (2) (Sept 2010) 134e143. [42] H. Prothmann, Organic Traffic Control (Ph.D. thesis), Karlsruhe Institute of Technology, 2011. [43] U.M. Richter, Controlled Self-organisation Using Learning Classifier Systems (Ph.D. thesis), Karlsruhe Institute of Technology, 2009. [44] A. Rong, R. Lahdelma, An efficient linear programming model and optimization algorithm for trigeneration, Appl. Energy 82 (1) (2005) 40e63. [45] M. Sakawa, K. Kato, S. Ushiro, Operational planning of district heating and cooling plants through genetic algorithms for mixed 0e1 linear programming, Eur. J. Op. Res. 137 (3) (2002) 677e687. [46] H. Seebach, F. Ortmeier, W. Reif, Design and construction of organic computing systems, in: 2007 IEEE Congress on Evolutionary Computation (CEC 2007), IEEE, 2007, pp. 4215e4221. [47] E. Shirazi, S. Jadid, Optimal residential appliance scheduling under dynamic pricing scheme via hemdas, Energy Build. 93 (2015) 40e49.  Gomes, A multi-objective genetic [48] A. Soares, C.H. Antunes, C. Oliveira, A. approach to domestic load scheduling in an energy management system, Energy 77 (2014) 144e152.  Gomes, C.H. Antunes, H. Cardoso, Domestic load scheduling using [49] A. Soares, A. genetic algorithms, in: 16th European Conference, EvoApplications 2013, Vienna, Austria, April 3-5, 2013. Proceedings, Lecture Notes in Computer Science, vol. 7835, Springer, Berlin Heidelberg, 2013, pp. 142e151. [50] K.C. Sou, J. Weimer, H. Sandberg, K.H. Johansson, Scheduling smart home appliances using mixed integer linear programming, in: 2011 50th IEEE Conference on Decision and Control and European Control Conference (CDCECC), IEEE, 2011, pp. 5144e5149. [51] R. Stamminger, V. Anstett, The effect of variable electricity tariffs in the household on usage of household appliances, in: Smart Grid and Renewable Energy, vol. 4, Scientific Research Publishing, 2013a, pp. 353e365. [52] R. Stamminger, V. Anstett, Effectiveness of demand side management by variable energy tariffs in the households e results of an experimental design with a fictive tariff model, in: Proceedings of the ECEEE Summer Study, 3e8 June, Presqu'île de Giens, France, 2013, pp. 2159e2166. [53] R. Stamminger, G. Broil, C. Pakula, H. Jungbecker, M. Braun, I. Rüdenauer, C. Wendker, Synergy Potential of Smart Domestic Appliances in Renewable Energy Systems, Tech. rep., University of Bonn, 2008. [54] Stichting Flexiblepower Alliance Network, Flexible Power Application Infrastructure, Tech. rep., Stichting Flexiblepower Alliance Network (FAN), 2013. [55] J.-J. Wang, Y.-Y. Jing, C.-F. Zhang, Optimization of capacity and operation for CCHP system by genetic algorithm, Appl. Energy 87 (4) (2010) 1325e1335. [56] A. Weidlich, D. Veit, A critical survey of agent-based wholesale electricity market models, Energy Econ. 30 (4) (2008) 1728e1759. [57] A. Wright, S. Firth, The nature of domestic electricity-loads and effects of time averaging on statistics and on-site generation calculations, Appl. Energy 84 (4) (2007) 389e403. [58] M. Zillgith, D. Nestle, M. Wagner, Security architecture of the OGEMA 2.0 home energy management system, in: Security in Critical Infrastructures Today, Proceedings of International ETG-Congress 2013; Symposium 1, VDE, 2013, pp. 1e6. €st, Impacts of volatile and uncertain renewable energy sources [59] M. Zipf, D. Mo on the German electricity system, in: 2013 10th International Conference on the European Energy Market (EEM), IEEE, 2013, pp. 1e8.

Please cite this article in press as: I. Mauser, et al., Adaptive building energy management with multiple commodities and flexible evolutionary optimization, Renewable Energy (2015), http://dx.doi.org/10.1016/j.renene.2015.09.003