Industrial demand side response modelling in smart grid using stochastic optimisation considering refinery process

Accepted Manuscript Title: Industrial Demand Side Response Modelling in Smart Grid using Stochastic Optimisation considering Refinery Process Author: S.Sofana Reka V. Ramesh PII: DOI: Reference:

S0378-7788(16)30439-X http://dx.doi.org/doi:10.1016/j.enbuild.2016.05.070 ENB 6705

To appear in:

ENB

Received date: Revised date: Accepted date:

6-1-2016 18-5-2016 20-5-2016

Please cite this article as: S.Sofana Reka, V.Ramesh, Industrial Demand Side Response Modelling in Smart Grid using Stochastic Optimisation considering Refinery Process, Energy and Buildings http://dx.doi.org/10.1016/j.enbuild.2016.05.070 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Industrial Demand Side Response Modelling in Smart Grid using Stochastic Optimisation considering Refinery Process

S. Sofana Rekaa, V. Ramesha

a

School of Electrical Engineering, VIT University, Vellore

Tamil Nadu, India

Corresponding Author: Dr Ramesh.V, School of Electrical Engineering, VIT University, Vellore, Tamil Nadu INDIA Email: [email protected], [email protected] Contact number: +91-9944730510

REFEREES 1

Dr. Chandrasekaran K

2

Associate Professor

Professor,

Department of EEE

PSN College of Engineering,

National

Institute

of

Technology,

Pondicherry, India E-Mail: [email protected]

3

Dr. Christopher Columbus C

Dr. Devi, Professor Department of EEE Government College of Engineering , Thrissur E-Mail: [email protected]

India, E-Mail: [email protected]

Highlights The major target of this work to show that 

DR scheme appropriately shifts the load from peak to non-peak hours by RTN scheduling by creating objective function and constraints using stochastic dynamic programming(SDN).The analysis of RTN discrete formulation is evaluated for successively three process units used in refinery process .Realistic data is been used for the performance evaluation of the model.



Another important aspect of the model is done for price based DR programs with different scheduled tasks using scenario based stochastic approach.

Abstract Demand Side Management (DSM) scheme in smart grid technology provides a broader vision of the electricity consumers to participate in power management in the future era. Both residential and industrial sector are active consumers of electric power, in which industry sector are the major consumers of electric power, globally. In this proposed work, demand response modelling scheme for industrial sector is done using Resource task network (RTN) scheduling process and stochastic dynamic programming. The model is designed mathematically and for validating the results practical field data from refinery plant is used. The DR scheme proposed is a new intelligent model for industrial domain with practical approach. The scheme exhibits the refinery processing tasks for scheduling the peak loads by considering the important schedulable tasks in the unit such as distillation unit and other units which involve maximum power. Day ahead pricing scheme is considered for scheduling the loads to shift the demand from peak to non- peak periods. Different pricing schemes are also considered for comparison. The DR problem was mathematically modelled using stochastic programming through minimisation of objective function with set of industrial based constraints. The results are obtained using GAMS solver and Gurobi optimizer in Matlab. By this scheme the shifting of peak hours at different tasks level in the building establishes a reduction of 6.5% cost reduction. The DR scheme proposed is validated with practical results which exhibit to shift the demand from peak to non - peak periods, hence reducing cost.

Keywords: Smart grid; demand side management; resource task network; scheduled tasks; stochastic process.

1. Introduction In the near future, the energy sector at the global level is set to move towards an intelligent modern power grid. In this growing smart grid landscape, Demand Side Management (DSM) is proclaimed as a major dimension of the future power supply. The conventional electric grid exhibits stability with one way power flow between the utility and the consumers [1]. The construction of new power plants can be curtailed in the coming decades, by bringing in smart grid technologies which thereby reduces CO2 emissions. Smart grid in the building division offers an incredible open door for enhancing the power quality issues and unwavering quality of energy sources due to the possibility of decentralization of demand supply matching using DR and decrease of transmission losses [2]. DSM plays a major role by providing consumers with an opportunity for reducing the electricity usage [1- 4] by shifting from peak hours to non-peak hours or rescheduling the loads. Balancing the electricity supply and demand has always been an underlying challenge in smart grid solutions. Demand Response (DR) plays a key role in DSM for balancing the demand and supply [3-5]. Successful smart grid deployment can be brought in globally by efficient design of DR programs [6]. According to the survey done from US energy report in 2014 [7] the share of electricity consumed by residential consumers was 22% where by the industrial consumers targeted up to 32%. To choose a DR model and to implement the scheme, the effectiveness increases by considering the scope of consumers that are involved in it. In accordance to the global report from Industrial energy agency [7, 8] three major sectors like residential, industrial and commercial sectors are the domain where the DR schemes can be applied excluding the transportation sector [6 - 10]. Apparently, designing of efficient DR program in domestic areas is comparatively easier compared to other sectors. Moreover DR schemes applied in this sector should understand that all the consumption patterns are different based on categories of domestic users like short range, long range consumers and real world mixed advancing and postponing users. This set of users infers its consumption attention depending on current, past and future periods whereas short range consumers include the perception about the energy price based on

the current time only. At present, many research works have been carried out on designing DR approaches with different intelligent modelling targeting the residential consumers [1-5]. However industry is the major consumer of electricity. Industrial plants are high energy consumers and handle higher peak loads. Globally [6 -11] industrial sector accounts for a greater proportion of economy. Therefore there is a greater need for developing successful DR programs and to implement DR for industrial facilities. Different energy savings measures have been followed for buildings by developing economic viable policies like renewable energy sources is explained in the work [12]. Further the analysis of importance of energy saving strategies by developing an optimisation model is formulated in [13]. The work proposed in [14] exhibits a series approach of energy efficient improvement method carried out on specific buildings in Serbia by different measures considering various household scenarios. Similar analysis is done in [15] by developing energy saving methods with the installation of automatic temperature control system for Greek buildings. All the approaches emphasise on the importance of energy efficient approaches at different sector at a global vision and so the need of DR measures is required for better energy saving programs at the next level approach. Thereby these programs initiated in industrial sector promote reliability of power systems besides reduction in energy costs. Compared to residential consumers there are many issues in planning DR scheme for industrial sector like scheduling pattern of entire manufacturing process. Many plants require monitoring of loads in a time frame of milliseconds and there are security concerns which require careful handling [16, 17]. Therefore proposals of new intelligent DR systems considering the above facts can improve reliability of the system in industries. There are important tasks and issues for raising DR for industrial domain compared to residential domain. The purpose of developing DR programs in this domain targets on global economy in electricity market. As per the statistics, the industrial domain occupies 32% in consumption [18]. 

The industrial domain process considers many raw material resources, intermediate sources like gas and water along with the electricity resources.



Furthermore, industrial process is a real time domain where satisfying the real time needs is a difficult task and building DR programs is important.

The implementation of successful DR schemes is challenging and difficult compared to residential sector. Utmost care is needed in implementing DR schemes in industry for the above reasons dealt with. Many industrial processes are sequential and therefore loads cannot be

rescheduled at random. Further industrial processes are real time and therefore a mismatch between supply and demand can have serious technical and financial impacts. However, designing and developing new intelligent DR methods can increase the reliability and efficiency [19]. Developing new DR methods can benefit the power utilities and exhibit optimal solution of DR to shift the loads from peak to non-peak hours. Very few research studies and reports exhibit the expediency and asset of DR in the industrial domain. In [20-22] the authors had dealt with different smart grid DR approaches for industrial domain and explained the overall analysis of DR scheme. In [21] the paper explains the DR techniques involved to reduce the total energy costs considering few industrial plants as a survey. In [22] the author clearly explains the features and the barriers found in implementing DR scenarios in industry facilities mainly in California region. The work mentioned in [23] elaborates the need of DR model using genetic algorithm for controlling peak demand level with energy efficiency considering a general case study of residential buildings. The case study explains the thermal control architecture of shiftable loads like air conditioning loads. In paper [24] the authors proposes a new model for industrial demand side management considering cement industry by using best behavioural scheme model with tariff schemes. The work mentioned in [25] the author proposes game theory based DR energy management scheme using punishment mechanism for industry facilities. The considered industry for the proposed model in [25] is refrigerated warehouses. In [24, 25] the authors have implemented different mathematical approaches for reduction of the cost for industrial domain considering different industries. In paper [26] the work is done based on wastewater treatment facilities using DR scheme, concerning a particular sector for load shifting and opportunities of DR in California. In the same aspect, paper [27] exhibits new industrial opportunities with DR scheme for optimizing operations. In the work proposed in [27, 28] the author elaborates the feasibility, importance and facilities of DR scheme approach in a general manner in the industrial sector. The studies and reports mentioned in [25- 28] describes different schemes in a general fashion to implement DR scheme in industrial sector. Moreover, the drawback found in the earlier studies does not impose any specific DR based algorithm [17], [24-28] for industrial domain. Conventional studies in the past [29- 31] implementing architecture for industrial domain, do not consider some of the major smart grid components for demand side management like pricing schemes for user interaction. Further in [31,32] the authors have proposed DR architecture for

industrial plants and specifically none of the studies has proposed a particular DR algorithm in general considering the schedulable and non schedulable tasks. In [33,34] the authors have enunciated the application of DR in different sections in industry domain as in cement and refrigerated systems. The work exhibits a general outlook of DR perspectives on the industry and further mathematical modelling and design were not exhibited at a practical situation. The work mentioned in [35] explains the need of demand response and various load patterns impact on industrial and commercial consumers. The analysis is done with critical pricing scheme on Korean market. The system does not depict any generalized scheme for DR exhibiting the smart grid features. In paper [36] complete assessment of DR management is clearly mentioned analysing the methodologies that can be proposed in the future. The work in [36] proclaims the need of DR applicability in various sectors. According to survey done [36] the industrial sector analysis of DR scheme is important at the global electricity market to match the need of economy. The work proposed in [37] portrays the DR assessment model as a case study in Germany. The REmix model is designed for the analysis .The work generalizes the need of DR assessment in developed countries at energy market level. New models or algorithms have not been proposed for industry consumers. In summary [27-39] the studies focus mainly on some specific task or application and failed to incorporate the major smart grid factors like dynamic pricing scheme in future to implement DR algorithm. Thereby in this work a new intelligent DR model is developed using resource task network (RTN). The model is analysed using real time data from the refinery plant unit, an industrial building. The scheme consists of dynamic hourly price schemes. The scheduling problem of the refinery unit is tackled using RTN under energy constraints. RTN process model is used to develop DR model since it exhibits multiple states and can be coupled with many constraints for industrial process. The proposed work in this paper contributes on DR scheme for industrial domain considering the basic units incorporated in a real time refinery process. The model is done based on real time price based DR scheme for industrial facilities. The process is done by scenario based stochastic dynamic programming arriving at an objective function with a set of constraints. The set of constraints have been articulated based on scheduling and non-scheduling tasks [38] of industrial facilities by Resource Task Network (RTN). The DR problem consists of scheduling the tasks such that the objective of reducing the cost is done with the constraints formed for complete process operations. In the proposed mathematical model the scheduling of the tasks in slots for 24 hrs is done using RTN

by fixing multiple location states. The RTN model has been coupled with the constraints by defining the DR problem with stochastic approach. The major objective of this proposed work 

DR scheme appropriately shifts the load from peak to non-peak hours by RTN scheduling by creating objective function and constraints using stochastic dynamic programming (SDN). The analysis of the RTN discrete formulation is evaluated for successively three process units used in refinery process. Realistic data have been used for the performance evaluation of the model.



Another important aspect of the model is done for price based DR programs with different scheduled tasks using scenario based stochastic approach.

The rest of the paper is organized as follows. Section 2 describes the proposed model for DR management in industrial zone using RTN scheduling tasks and stochastic programming. Section 3 exhibits the mathematical formulation of the proposed scheme with illustrative examples. Section 4 describes the optimization results showing the proposed scheme which reduces the energy cost with the study from refinery process. The paper is concluded in Section 5 with concluding remarks and future work.

2. Proposed Model for Demand Response scheme

The proposed model has been made for industrial demand side management scheme considering the important factors in industry like resources balance, maintaining schedule of the units, production constraints along with the energy balance. As demand reduction in an industrial plant or process requires more sophisticated and complex solutions, an intelligent DR scheme is evaluated considering the example of oil or petrol refinery, to work on the important units in the process to actively participate in shifting of loads in an effective manner without disturbing the work schedule and pattern in an easier way. The important elements of DR industrial model consist of utility data centre interacting with the users by announcing prices depending on the supply and demand. The communications between them are done in a Wide Area Network (WAN).

In an industrial process unit, developing an optimal solution for efficient scheduling solutions depending on the pricing aspect is done by two tasks as non-schedulable and schedulable task. The task model in the proposed work is done by RTN considering a refinery unit .The scheduling model is designed as an abstract layer by different model entities by developing suitable constraints. The sequential scheduling task in a refinery process considering the three major units is processed for demand response modelling. The RTN model develops a unit electricity schedule depending on daily basis tariff. The prices are considered as day ahead pricing. The objective of minimizing the total cost over a day with the minimum set of more processed units in a day is achieved. Fig.1 gives the structure of any major general refinery plant.

They provide good reserve capacity with the decrease of demand, as and when the electricity system runs short on capacity pattern. This plant involves many large scale multistage units with critical separation units for crude oil and many energy related constraints. The crude distillation unit, hydro treatment and the desalination unit is considered for scheduling with energy constraints. The hydro treatment unit removes the contaminants and obtains a clean product, requiring different stages of preheating, cooling and pumping process. This requirement depends on correct processing time and a certain time horizon. So the entire DR model is featured considering the constraints of the three main processing units with RTN and solver process for effective scheduling. 2.1 Resource -Task Network Processing Model

RTN is a schedule based structure [38, 39] mainly incorporated in processing of industry models which have been perceived as a metaphysical layer between the real plant entity and the constraints developed. In this proposed work RTN representation is mainly involved in evaluating the scheduling process done in a unit to obtain an optimal solution. RTN model [3638] consists of tasks and resources with nodes. Each task in the network represents every equipment unit in the processing plant. The material resources in the plant are considered as different states. The process of consumption and production of resources are represented as interlinked nodes between the tasks. Each task or the unit exhibits each processing activity and every state represents the interconnection between them. In general RTN is more preferred

compared to State Task Network (STN) [39 - 42], as the combined effect of the states, tasks and units with the resources leads to a model with a single structure without duplication of constraints [43]. In RTN along with the state and the task nodes [43, 44] processed, equipment is also modelled in representation. In RTN model discrete time based formulation is done for scheduling problems. Fig. 2 shows simple structure of a RTN model. The RTN is done as representation of industry process unit which is used to obtain a structure of scheduling solution. The tasks are represented in the Fig. 2 as rectangles and the resources like equipment and other parameters in a unit are represented as circles. The DR model obtained is integrated with the utility provider which supplies prices according to the need of demand and the supply.

When the price is large the DR model is used to obtain the balance for reduction of cost. To obtain this DR model, the tasks done in the industry have been classified as schedulable tasks and non-schedulable tasks. The non-schedulable tasks are the particular states where the demand required is difficult to schedule for the demand response .The schedulable tasks are the set of tasks in an industry process where the demand has been scheduled at a certain set operating point for DR. When the electricity price is high, the DR model is executed. As the industry sector is complicated and holds different scenarios, the knowledge on schedulable and non-schedulable tasks in each industry like paper, glass, steel refinery should be known a priori for DR scheme. In a refinery unit, the pumping and splitting processes are schedulable tasks compared to the furnace unit which consists of non-schedulable tasks. Setting the operating point among the units for scheduling is important for implementing the RTN model

In Fig.3 an elaborate representation of scheduling and the solver is exhibited between the utility and the user. Further with proper scheduling process depending on the electricity pricing, schedulable tasks’ operating points are set. The scheduling process of major industry units can be shifted from peak periods where the pricing is more to non-peak hours. The complete model works according to the following condition. {P is high  Task sets operating set S sT   Low Power Consumed  Shift load }  Di   Cos t reduction  T  Shift Load  high power consumed {PDi is low  Task set Ss  

When the power demand is high, the consumer reduces the electricity consumption based on shifting the loads with respect to task set operating points. The energy management in smart grid environment for industrial consumers is done, based on RTN by identifying the resources and tasks. The tasks are related to the units for achieving the product at best DR scheme. The RTN is better than STN as the units are explicitly separate and can be done with multiple units. Developing RTN model scheduler by considering the units for shifting loads depending on the information of each task and the pricing conditions from the market provider. 

The scheduling process is interlinked for developing the objective function and different constraints for material balance and process unit modelling by a mathematical solver (GAMS) using stochastic scenario based optimization



The pricing schemes are analyzed for two ways, day ahead pricing scheme and hourly pricing scheme.

2.2 Stochastic Optimisation Approach

The proposed DR problem of operational tasks scheduling is done by creating an objective function and developing constraints according to the stochastic process. The model is designed to obtain a broad DR based algorithm to reduce the energy cost with refinery industry as an example. The scheduled tasks in the process are done by obtaining the optimal schedule using RTN. The entire DR module is done with stochastic programming by a two stage process incorporating day ahead pricing schemes and hourly pricing schemes. The structure is explained in Fig. 4. The process is done in a two stage model by formulating a rolling procedure.

The rolling process in the first step includes the first time slot t j with day ahead pricing scheme and the best scheduling is done to reduce the cost. The second stage is done at t j  1 and the scheduling decisions are done by RTN. The total structure is divided into four modules. The first module consists of information including the pricing scheme, stage inputs and task inputs of the industrial process. The second module consists of developing the objective function for the industrial process modelling and information between the utility grid and the industrial consumer. There is a list of decision variables and constraints obtained for the process. The objective is to minimize the cost by developing an objective function by considering different

constraints for the model. The stochastic problem is solved and the mathematical formulation is elaborated in Section 3. 3. Mathematical Formulation The RTN based mathematical stochastic formulation involves the variables consisting of resources Ri , tasks Kt and the time variables ti . The proposed demand response algorithm involves the goal of minimizing the cost in general for industrial domain considering the test cases for the units from the refinery plant. The processes are elaborated to obtain the scheduling of the tasks based on day ahead and hourly pricing schemes. Many industrial scheduling problems can be analysed with RTN models [38] depending upon the different stages in the process. The model provides more stability and resilience to the power system in order to reduce total energy consumption and rescheduling the loads to off peak hours. The mathematical formulation is done with stochastic dynamic programming and the method proposed is capable of handling solutions towards industrial background. The RTN model approach for solving problems with industrial relevance mainly includes different set points and the production rate of every state including the consumption. Further the storage task is also considered for the appropriate scheduling of the process. The model is designed with discrete time formulation as shown in Fig.5 which is interrelated with each time slots ti Tt spanning 24hours. The time slot parameter is given as  which depends on the approximation level of the strategy concerned in scheduling.

The formulation for scheduling problems is established with several constraints of the tasks represented by the processing units. Several constraints are evaluated for the scheduling problems like resource balance constraints, timings, energy balance, operational constraints and utility requirements for industrial process.

3.1 Objective function The proposed model establishes an objective function on the driving factor to minimize cost using DR scheme. The cost function is obtained using scenario based stochastic two stage rolling optimisation processing considering day ahead pricing schemes.

min{ p(a)  f (a)  D[ x(a,  )]} (First stage of the model)

(1)

aA

min{q(b,  ) | H ( )a  I ( )b  h( )} (Second stage)

(2)

b

The Eqs. (1) and (2) exhibits the general two stage stochastic optimisation expressions. minx p(a)  cTt a  D[ X (a,  )]

(3)

a

Subject to Ba  C , a  0

minx q(b,  ) b

(4)

Subject to H ( )a  I ( )b  h( ) . The variable a represents the binary decision variable of the units. A random scheduling vector generated depending on finite number of possible scenarios (DR strategy) is denoted by γ. The first and second stage of the stochastic process is given as p(a) , q(b) . The stages of the problem are optimised to obtain the finalised optimised value of each stage whereas x[a,  ] represents optimal value of the first stage problem. D[ X (a,  )] represents the optimal value of the second stage problem. The assumptions considered for the model is based on the supply electricity and price. C t represents the cost function and f (a) denotes the objective function .The first and second of domain variable is given in Eqs. (3) and (4) for the model to define the cost function. Moving further to the second stage all the variables realized based on the strategy and optimizing of the problem is done. The cost function CTt a of the first stage plus the expected cost value from the second stage at the next period is calculated for discretization. The variable  is considered as a random vector with a different set of scenarios as of DR strategy based on the sum of probabilities x1......xk .The limitations considered for the model is based on 24 hrs schedule with a set point of 24 slots . Further they are been modelled for multiple set points. K

D[ X (a,  )]   xk X (a,  k )

(5)

k 1

The variable  is solved using Gurobi optimizer with MATLAB by GAMS solver

Yy,t*  Yy,t*1   m

k

 ( k , y, Nk ,t   zk , y,k ,t  )   y,t

 0

*

*

*

 y,t*

(6)

for different scenario (unit) construction in the industrial process. So the need of implementing the cost with this process with multiple units exhibits better performance.

The set of resources exhibited are denoted by y and the set of each task implemented in the process is given as j . A fixed amount of equipment resources produced or consumed by the task

j is represented by  k , y, . Coefficient of the material resource at the end of the task j is given as zk , y, and continuous model variable for the resource is denoted as Nk ,t  . External influence on *

the set of resources y is implemented in the model as  y,t* . The following objective function using the two stage scenario process is developed in the equation RN (a) 

1 N

N

D ( a,  i )  i 1

RN (a)  min cT a  t

a A

1 N

(7) N

D(a,  i )  i 1

(8)

where the scenarios are considered as a function, from the process unit depending upon day ahead pricing scheme purchased from the grid. Scenario construction for scheduling is denoted as RN (a) . The probability functions with the cost coefficients are exhibited in Eq. (8). The constraints of the objective function depend on the process unit in the industry sector based on RTN scheduling process. 3.2 Model Constraints Every task in the model is divided into binary and continuous variables. Each single unit in the process is considered for every task. As the process is modelled for a refinery unit, three major units in the process namely, the crude distillation unit, hydro treatment unit and gas separation splitter are considered for scheduling. The constraints are modelled according to the process. In general the tasks [22] will be performed based on two sets of continuous variables. Generally the tasks in RTN model is characterized by two binary variables namely,  k , y, and zk , y, which are the total amount of resource y consumed or produced by the units from the set of task j at time t * . The function of resources Yy,t* depends on the discrete set of time by different interactions with the schedule k . 3.2.1 Resource Constraints The balance between the supply and demand in the process unit is indicated with the resources at a time state and the total amount of resources (material) consumed or produced are given in Eq. (9). The resource balance constraints are evaluated by the equation

Yy,t*  Yy,t*1   m



y ,t *

k

 ( k , y, Nk ,t   zk , y,k ,t  )   y,t *

 0

*

*

 y,t* )

(9)

is the quantity of resources from other external unit starting from the time duration of

0 to  . 3.2.2 Operating constraints To establish the formulation of the process unit for every task, operating constraints for energy balance and storage is needed at every set point for the schedulable task as follows

(Yy,t* min ,Yy,t* max )

 Y min  y,t*    z min N * k ,t  k 

     k ,t*  zk max Nk ,t*k ,t*   

 Yy,t  Yy,t* max  y,t*

(10)

where the operational constraints are added with the objective function to solve an optimal schedule. The operating point of each schedule at the resources tends to work when the price is too high. If the resource representing every unit and task exhibits the time constraints the following conditions apply for the scheduling. For the energy balance constraints the assumptions are done based on task set point for each time interval.  k ,t*  1 for t *  0 {Yy ,t*  1 for every N  1}  k ,t*  0 for any other t * {Yy ,t* remains unchanged and if N  1}

(11)

zk ,t*  0 for any t *

3.2.3 Storage Constraints The entire batch process is done with the objective function considering the energy storage constraints in the formulation

 k min Nin(n1) 

 Yy,t yY

*

 i max Nin(n1)  k  K s

*

 i

t

 kmin

Nin(n1) 

 Yy,t

yYt

(12) max

Nin(n1)  k  K

s

The illustrations of the constraint parameters and the RTN model are given in Fig. 6 and Fig.7. The RTN process with the model constraints for the three tasks, are exhibited in Fig.7. Every time, each task is repeated with two stage stochastic optimisation with the RTN process. Here the tasks are separated into three units. Fig.7 shows the simple schedule made for a particular shift

in the task depending upon the process made and the further representations is done in results section. 4. Results and Discussions The performance of the mathematical formulations made for the three units using RTN by setting an objective function for reducing cost is illustrated based on the processing time. The formulations are raised to stochastic optimisation problems. The problem is implemented in GAMS tool (Generic Algebraic Modelling System) with DECIS solver to solve two stage stochastic problems. GAMS is a mathematical tool solving complex problems. The computational time for the process is much easier when running on Intel (2.07GHz) with 8GB RAM. The real time refinery facilities unit information is gathered. The refinery plant considered has co-generation facilities for the generation of power. In a refinery unit, there are three main units such as the desalination unit, hydro treatment plant and crude oil distillation unit where maximum energy transfer happens. They are the major industrial units, where the process of heating, pumping, and conversion from crude oil is mainly done. Considering the crude oil distillation unit the main process of pumping and cooling have been schedulable tasks. DR energy management in this process is done based on the day ahead pricing from the grid. Table 1 and 2 shows the power consumption made in refinery unit (kW) and the loads variation in the unit.

Generally in production planning of refineries, there are crude pipelines with different units connected to crude oil tanks, many intermediate processing units and intermediate tasks. The three units considered for scheduling operation is the cooling tower, pumping unit and the separation of crude oil. The RTN representation of the three task units are done based on the mathematical optimisation. The major units consisting of two cooling tower units, two heating units (boiler units) five pumping units and one compressor unit are scheduled with DR model in the three main schedulable tasks from the non-peak hours to peak hours. The realistic data of power consumption obtained from each unit in the refinery process is obtained in Table 3. The refinery unit is fed with a cogeneration power plant which produces 30MW power. The main unit considered for the operating schedule in the process is the Naphtha Hydro Treatment unit which separates petrol from naphtha, subsequently crude oil to naphtha and finally petrol. The compressor in the unit takes the maximum power since it is important for

recycling the hydrogen gas produced to make the separation process happen at high pressure. The different units considered above are scheduled using RTN network process. Fig.8 shows the example of the units considered for scheduling

A two stage stochastic optimisation schedule for the proposed three units with DR model is done. As an example the demand of the cooling process during each interval time is set to be dynamic and they are set with three operating points at each task. The stochastic GAMS solver is used to solve for the DR system with different task facilities with one objective function based on Eqs. (1) - (11). The stochastic programming is solved using DECIS solver finding the operating point at each task. At each time interval of processing done for crude oil separation, naphtha treatment, pumping unit is obtained with set of four operating states respectively.

In Fig.9 the day ahead pricing scheme with the required amount of electricity for purchasing or selling back to the grid is found to be 5% lower. Table 4 and Table 5 give the operating set points of RTN tasks, For example the cooling water process in hydro treatment unit is scheduled to operate at the operating point 1 and 2 around (1.00-2.00 pm) where the production is minimum .The electricity demand increases at the time slot when the price is low. In this example the demand increases at 4 and 9 slot where the price is low. Here, excess of pumping is done using the compressor unit which leads to increase of demand. The scheduling process is worked as a two stage rolling process for 24 hrs in the GAMS solver, which provides a mathematical solution based on the probability of shifting the hours according to the best pricing time and without affecting the product process. Table 4 exhibits the different tasks in the unit used for scheduling. There are four units considered for the scheduling process. The main subunits in all the tasks are found to be cooling water tower and the pumping unit which are considered as examples. There are indexes fixed for each task depending on the operating point during one time interval. The cooling water unit measured in (m3) is assumed to be a schedulable task and the response time is short here. More precisely, there are water cooling systems, which are required for the units. The demand for cooling water during each time interval is dynamic and depends on the operating status of the main unit. Like this the other units are considered separately and calculated based on the realistic values. In the cooling unit, the lowest and upper storage bound are fixed up to 600 m3. Further

the maximum electricity demand used for the first task unit is 5000 kWh. Depending upon the demand exhibited the scheduling of the tasks is been done by RTN for the production process. In Table 5 the values are the determined operating set point for the tasks considered during each time interval. For example considering the Naphtha hydro treatment unit it was scheduled at an operating point in a particular time frame to remove the contaminants such as sulphur and nitrogen. The unit at the set point 1 indicates a particular time interval (from 1.00 to 2.00 pm) producing about 580 Nm3 of naphtha consuming 340 m3 of cooling water and requiring of about 2200 kWh of electricity from Table 4. So the maximum set points are found at each unit so that shifting of demand is done. Furthermore for every task a set point is fixed mathematically by the stochastic model and the maximum electricity demand is found out. During the periods where the demand is more, scheduling is done. Further the entire mathematical programming for each task is completed with the day ahead electricity pricing and shown as an example in Fig.10 for scheduled tasks of pumping unit used in all three units. Similarly in Fig. 11, electricity generation of schedulable tasks at each unit of the pumping process is given. The value indicates the pumping level at two stages. In the naphtha treatment unit, where the pumping is made more, the consumptions conveniently reduced to about 8000 kWh. Being an industrial process, the product deployment is carefully done, without obtaining break up in the process unit when shifting the loads.

The results obtained in Table 6 exhibits that the total energy costs at different cases starting from fixed price consumptions at each time interval extended to dynamic pricing. For instance considering the hourly pricing which is used for reduction in the energy costs by shifting tasks 1 to tasks 4 from peak price to low price and there is considerable decrease in the cost.

The statistics analysis in Table 7 forms the performance analysis for the refinery plant case study where the computation is solved with computational time. At each every tasks in the units there are computational variables and the subsequent energy constraints developed to solve the objective function.

The computational results using GAMS solver for different tasks developed and the energy costs obtained are shown in Table 6. On developing mathematical programming the statistics of the refinery unit case study considering the four tasks are solved in considerably reasonable time. Mainly the day ahead pricing scheme shows a reduction in the energy costs in shifting the task 1task 4 from high price intervals to low price intervals. Fig.12 shows the total demand at each time interval. During the interval 4-5, there is considerable reduction in energy costs compared to the process without the model. Thereby there is reduction in demand at the interval of 7 - 8 where the pumping process unit is shifted two hours before, without losing the comfort of the processing stage.

5. Conclusion

With the advancement of smart grid technology, DR modelling has become the major technology to improve the reliability of the grid with the reduction of the costs at the consumer side. Implementing DR schemes for industrial sector is more complicated compared to residential and commercial sector, since in industry there are many issues concerned like material storage, requirements of quality products and proper reliability of supply management. In this proposed work, a DR scheme is elaborated and designed based on real time industry refinery process with day ahead pricing scheme. The DR scheme is formulated as an optimisation problem to obtain objective function minimisation with different energy based constraints. The mathematical modelling of the unit is designed using RTN stochastic process which is exhibited for four basic task units in the refinery plant. The basic units which take up the maximum power in the unit and require the shifting of loads using schedulable task units are considered. The performance and the modelling are evaluated by optimising the schedulable tasks of the refinery process plant. The results obtained in the scheme are able to shift the demand from peak to non-peak hours, enabling considerable reduction in costs. At the peak level shifting there is a considerable reduction of cost of 6.5 % at the main task level considered for the new DR approach. This is made as an approach at the level of many pricing schemes which

exhibits a reduction of 4.5% change at each tasks. At the main units of the refinery building, tasks level are set up so optimisation level of maximum power is reduced by 6.2% considerably. Different pricing schemes are also analysed for the new proposed DR scheme. As a future perspective, different case studies of real time implementation of DR schemes can be done for testing the performance and efficacy of the scheme. Furthermore many research studies can be carried out to implement the scheme at different sectors like steel and paper industry. The direction of generalising proper DR scheme for industrial sector also needs proper standardization of communication networks with the interface of energy management for practical applications. The model designed in this study elaborated on using schedulable tasks in a processing unit. Additionally, investigating the ambivalence of non schedulable tasks can be done and also climatic conditions can be considered in future.

References

1. Palensky, Peter, and Dietmar Dietrich. "Demand side management: Demand response, intelligent energy systems, and smart loads." Industrial Informatics, IEEE Transactions on 7.3 (2011): 381-388.

2 Kolokotsa, Dionysia."The role of smart grids in the building sector." Energy and Buildings (2015). 3. Güngör, Vehbi C., et al. "Smart grid technologies: communication technologies and standards." Industrial informatics, IEEE transactions on 7.4 (2011): 529-539.

4. Gelazanskas, Linas, and Kelum AA Gamage. Demand side management in smart grid: a review and proposals for future direction. Sustainable Cities and Society 11 (2014): 22-30.

5. Deng, Ruilong, et al. A Survey on Demand Response in Smart Grids: Mathematical Models and Approaches, IEEE transactions on Industrial informatics, 3(11) 2015:570-582.

6. Vardakas, John S., Nizar Zorba, and Christos V. Verikoukis. "A survey on demand response programs in smart grids: pricing methods and optimization algorithms." Communications Surveys & Tutorials, IEEE 17.1 (2015): 152-178.

7. U.S. Energy Inf. Admin., Annual Energy Outlook 2014, Washington, DC, USA, Apr.–Sep. 2014

8. Key World Energy statistics 2014, International Energy Agency, IEA 2014,Paris. www.iea.org/statistics/ 9. M.M. Eissa, “Demand side management program evaluation based on industrial and commercial field data,” Energy Policy, vol. 39, no. 10, pp. 5961–5969, Oct. 2011

10. Key World Energy statistics 2015, International Energy Agency, IEA 2015, Paris. www.iea.org/statistics/ 11. Gils, Hans Christian. "Economic potential for future demand response in Germany – Modelling approach and case study." Applied Energy 162 (2016): 401-415. 12. Verbeeck, G., & Hens, H. (2005). Energy savings in retrofitted dwellings: economically viable?. Energy and buildings, 37(7), 747-754. 13. Malatji, E. M., Zhang, J., & Xia, X. (2013). A multiple objective optimisation model for building energy efficiency investment decision. Energy and Buildings, 61, 81-87. 14. Matic, D., Calzada, J. R., Eric, M., & Babin, M. (2015). Economically feasible energy refurbishment of prefabricated building in Belgrade, Serbia. Energy and Buildings, 98, 74-81. 15. Nikolaidis, Y., Pilavachi, P. A., & Chletsis, A. (2009). Economic evaluation of energy saving measures in a common type of Greek building. Applied Energy, 86(12), 2550-2559. 16. Mohagheghi, Salman, and Neda Raji. "Managing industrial energy intelligently: Demand response scheme." Industry Applications Magazine, IEEE 20.2 (2014): 53-62.

17. Mohagheghi, Salman, and Neda Raji. "Intelligent demand response scheme for energy management of industrial systems." Industry Applications Society Annual Meeting (IAS), 2012 IEEE. IEEE, 2012. 18. Lund, Henrik, et al. "From electricity smart grids to smart energy systems–a market operation based approach and understanding." Energy 42.1 (2012): 96-102.

19. Zhang, Xiao, et al. "Industrial demand response by steel plants with spinning reserve provision." North American Power Symposium (NAPS), 2015.IEEE, 2015.

20. McKane, Aimee T. "Opportunities, barriers and actions for industrial demand response in California." Lawrence Berkeley National Laboratory (2009). 21. T. Samad and S. Kiliccote, “Smart grid technologies and applications for the industrial sector,” Comput. Chem. Eng., vol. 47, pp. 76–84, Dec. 2012.

22. Mares, K. C. (2010). Demand response and open automated demand response opportunities for data centers. Lawrence Berkeley National Laboratory.

23.Gao, Diance, and Yongjun Sun. "A GA-based coordinated demand response control for building group level peak demand limiting with benefits to grid power balance." Energy and Buildings 110 (2016): 31-40. 24. Xu, Fang, and Loi Lei Lai. "Novel Active Time Based Demand Response for Industrial Consumers in Smart Grid. 25. Ma, Kai, Guoqiang Hu, and Costas Spanos. "A Cooperative Demand Response Scheme Using Punishment Mechanism and Application to Industrial Refrigerated Warehouses." (2014). 26. Thompson, Lisa. "Opportunities for Open Automated Demand Response in Wastewater Treatment Facilities in California-Phase II Report. San Luis Rey Wastewater Treatment Plant Case Study." Lawrence Berkeley National Laboratory (2010).

27. Agha, Mujtaba H., et al. "Integrated production and utility system approach for optimizing industrial unit operations." Energy 35.2 (2010): 611-627 28.

Merkert,

Lennart,

et

al.

"Scheduling

and

energy–Industrial

challenges

and

opportunities." Computers & Chemical Engineering 72 (2015): 183-198. 29. Olsen, Daniel. "Opportunities for Automated Demand Response in Wastewater Treatment Facilities in California-Southeast Water Pollution Control Plant Case Study." (2014). 30. P. M. Castro, A. P. F. D. Barbosa-Povoa, and H. Matos, “An improved RTN continuous-time formulation for the short-term scheduling of multipurpose batch plants,” Ind. Eng. Chem. Res., vol. 40, no. 9, pp. 2059–2068,Mar. 2001. 31.M. H. Agha, R. Thery, G. Hetreux, A. Hait, and J. M. L. Lann, “Integrated production and utility system approach for optimizing industrial unit operations,” Energy, vol. 35, no. 2, pp. 611–627, Feb. 2010. 32. Y. Wang, P. Zeng, H. Yu, Y. Zhang, and X. Wang, “Energy tree dynamics of smart grid based on industrial internet of things,” Int. J. Distrib. Sens. Netw., vol. 2013, 27pp, 2013, article ID 583846.

33. Grein, Arne, and Martin Pehnt. "Load management for refrigeration systems: Potentials and barriers." Energy Policy 39.9 (2011): 5598-5608. 34. Olsen, Daniel. "Opportunities for energy efficiency and demand response in the California cement industry." (2012). 35. Jang, Dongsik, et al. "Variability of electricity load patterns and its effect on demand response: A critical peak pricing experiment on Korean commercial and industrial customers." Energy Policy 88 (2016): 11-26. 36 . Boßmann, Tobias, and Eike Johannes Eser. "Model-based assessment of demand-response measure - A comprehensive literature review."Renewable and Sustainable Energy Reviews 57 (2016): 1637-1656.

37 .Gils, Hans Christian. "Economic potential for future demand response in Germany– Modelling approach and case study." Applied Energy 162 (2016): 401-415. 38. Song, Guang, Tong Qiu, and Bingzhen Chen. "Improved Resource-Task Network-Based Flare Minimization Model for Ethylene Plant Start-up: Rigorous Treatment of Cracking Furnace and High-Pressure Steam."Industrial & Engineering Chemistry Research 54.24 (2015): 63266333. 39. Paulus, Moritz, and Frieder Borggrefe. "The potential of demand-side management in energy-intensive industries for electricity markets in Germany." Applied Energy 88.2 (2011): 432-441.

40. Wassick, John M., and Jeff Ferrio. "Extending the resource task network for industrial applications." Computers & chemical engineering 35.10 (2011): 2124-2140. 41. Chen, Cheng-Liang, and Chia-Yuan Chang."A resource-task network approach for optimal short-term / periodic scheduling and heat integration in multipurpose batch plants." Applied Thermal Engineering 29.5 (2009): 1195-1208. 42. Castro, Pedro M., Lige Sun, and Iiro Harjunkoski. "Resource–task network formulations for industrial demand side management of a steel plant."Industrial & Engineering Chemistry Research 52.36 (2013): 13046-13058. 43. Shaik, Munawar A., and Ramsagar Vooradi. "Unification of STN and RTN based models for short-term scheduling of batch plants with shared resources." Chemical Engineering Science 98 (2013): 104-124. 44. Wassick, John M., and Jeff Ferrio. "Extending the resource task network for industrial applications." Computers & chemical engineering 35.10 (2011): 2124-2140.

Gasoline Spillter

Crude Main Pump

Diesel Hydro Desulpuriza tion

Central Operation Unit Crude Distillation unit

Refinery

Naptha Hydro Treatment Unit

Desalination Unit

Crude Distillation Unit

Fig.1. Structure of General Refinery Plant

R6

R1

Task 1

R3

Task 3 R5

R2

Task 2

R4

R7

Fig. 2. RTN general structure

Task 4

Resource Task Network framework Scheduling Process

Operation of each task

Mathematical Modeling STOCHASTIC PROGRAMMING – objective function and constraints (Process Model) Industry Constraints – Schedulable and Non schedulable Task (Energy cost reduction)

Fig.3. Representation of the Entire Set up with Solution

Time First stage

Time next slot

Fig.4. Scheduling process with scenario based stochastic process

1

2

3

4

5

7

0

Fig.5. Time Grid horizon for the model

T-4

T-3

T-2

T-1

Time (24hrs schedule)

T

Temp T

Task 1 Distillation-1

Heat

P2

Task 2 Crude Distillation

Start Stage 1

Stage 2

Task 4 Hydro treatment Unit

P1

Task 3 Processing

Fig.6. RTN process model featuring the unit with processing tasks and transfer with time

Distillation 60 60

60

60

60

60

60

60

Hydro Unit

0.5

Seperation

0.5

0.5

0.5

0.5

0.5

0.5

0.5

0.5

0.5

0.5

0.5

50 14.8

Time (24 hrs)

Fig.7. Optimal schedule for example done with RTN process

Crude oil

naptha

CT1

HEATING UNIT

COMPRESSOR UNIT CT2

CT3

Atmospheric Residue

Diesel PUMP UNIT

SEPERATION UNIT

Cracked Naptha

Fig.8. RTN representation of the Schedulable unit considered for DR

35

Price (Rs/Kwh)

30 25 20 15 10 5 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

Time (24 hr Schedule)

Fig.9. Pricing (day ahead) for obtaining electricity

12000 10000

Demand

8000 6000 4000 2000 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

Time (hr)

Fig.10. Electricity demand of pumping unit (as an example) during each interval

10000

Demand

8000 6000 4000 2000 0 1 2 3 4 5 6 7 8 9 101112131415161718192021222324

Time (hr)

Fig.11. Schedule electricity generation made for pumping process at each interval time

25000 With model

Demand

20000

With out model

15000

10000

5000

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

Time (hr)

Fig.12. Total electricity demand considering the scheduling of tasks with DR and without DR model

Table 1 Power Consumption in a refinery plant For 24

hrs. sending and For 24

ending @ 7.00AM Day1

ending @ 7.00AM Day 2

W

W

MW POWER

GENERATION 406001

hrs. sending and

MW 16.92

578098

24.09

3.04

225566

9.4

13.88

352532

14.69

FROM COGEN TOTAL

EXPORT

OF 72896

POWER REFINERY CONSUMPTION.

ENERGY 333105

Table 2 Different loads in the Refinery Unit.

INSTANT LOADS

MW

Unit 1 Task 1 (L1)

0.25

Unit 2 Task 2 (L2)

0.19

Unit 3 Task 4

0.155

R1 (Separation Unit)

0.27

Gas INCOMER # 1

6.33

Gas INCOMER # 2

6.14

TOTAL

12.462

Other Units

8.56

Other Units

19.44

TOTAL GENERATION

28.01

TOTAL REFINERY LOAD

15.545

TOTAL PDO LOAD

13.31

Table 3 Realistic Practical data from refinery unit S.No

Main Unit

Units

Power (kW)

1

Gasoline splitter

Net Gas Compressor

800

Refrigeration Compressor

300

Stabilizer

Re-boiler 185

Circulation Pump

2

Crude Distillation unit

Splitter Overhead Pump

200

PF Tower bottom pump

350

Long Residue pump

315

PF Tower bottom pump

350

Long Residue pump

220

Gas oil pump

150

Crude Charge Pump

795

Pre-Flash Tower Re-boiler 290 circulation pump

3

4

Naphtha Hydro Treatment unit

Desalination Unit

Crude oil pump

185

Gas Oil Pump

200

Fuel Oil Bunker Pump

250

Stabilizer Re-boiler Pump

210

Charge Pump

290

Recycle Gas Compressor

1000

Re-injection Pump

420

FW Pump

250

FW pump

360

Cooling Water Pump

475

Primary

Cooling

Water 475

Pump 5

Crude Pipe(Transport)

main Main Pump Booster Pump

2100 280

consumption

6

7

Diesel

Hydro Regal Oil R&O32

500

Desulphurization

Makeup Compressor

450

PENEX

Makeup Compressor

180

It’s a unit to treat the light Charge Pump

210

Naphtha

132

Charge Pump

Table 4 Operating points at time interval Operating Point

Cooling water Pumping

and Naphtha

Electricity

demand(1-2)

separation(2-4)

Treatment (5-8) Demand

(m3)

(Nm3)

Heating (m3)

kWh

Task 1

400

350

300

5000

Task 2

55

259

290

1369

Task 3

600

320

560

1699

Task 4

340

321

580

2200

Table 5 Set Operating points for day ahead pricing at each interval (at each time interval)

Time interval

Task 1

(24 hrs )

(Set point at

Task 2

Task 3

Task 4

each hour) 0-1

1

4

5

7

1-2

7

8

1

5

2-3

1

4

5

8

3-4

8

1

4

6

4-5

6

1

2

4

5-6

3

2

1

4

6-7

2

4

1

5

7-8

1

6

3

5

8-9

5

2

1

1

9-10

2

4

5

1

10-11

1

2

3

7

Table 6 Energy costs obtained

Cost in Dollars ($) Tasks

Fixed Price$

Hourly Pricing ($)

Day ahead pricing$

Task 1

11264

10594.6

7890

Task 2

10867

10672

9900

Task 3

12900

11600

8909

Task 4

12999

12000

7998

Table 7 Computational performance of the tasks Tasks

No of variables

No of Constraints

CPU time(sec)

Task 1

1356

789

14

Task 2

378

104

18

Task 3

1257

469

16

Task 4

213

103

12

processing