Agent based energy management system for CHP engines in Energy Hubs WP 3.6: Development of Energy Management System Dr. Spyros Skarvelis Kazakos, University of Greenwich 1 Summary The aim of this work is to design an energy management system that can regulate the output of CHP generators based on complex decision making in conjunction with other energy sources. Multiple energy carriers such as electricity, heat, gas, will be considered. A distributed decision making methodology will be applied to co ordinate the different sources autonomously, through intelligent software called agents. The concept of energy hubs will be used within the agents, in order to optimize systems with multiple energy carriers. CHP systems will be considered as parts of an energy hub, which converts one energy carrier (e.g. gas) to two others (electricity and heat). 2 Introduction background concepts The latest developments in local energy generation indicate that future distribution networks will be complex and will comprise not only distributed generators but also energy storage devices and controllable loads [1]. Many distributed generation technologies are expected to be capable of Combined Heat and Power (CHP) production [2]. There is also a growing trend in the implementation of distributed energy storage devices in small scale applications, such as domestic [3]. All these resources can be collectively characterized as Distributed Energy Resources (DER). The inclusion of CHP generators, such as microturbines or fuel cells, in local energy systems suggests that the input fuel could be a different energy carrier, such as Natural Gas, hydrogen or biofuels. These carriers are delivered through different networks, which influence the overall efficiency and characteristics of the local energy system. It has been proposed in [4] and [5] that these energy carriers should be included in the analysis of a local energy system. In [6], a tool for integrating economic dispatch and optimal power flows of electricity and gas at the Grid Supply Points of Great Britain has been presented. In [5], an infrastructure planning tool was proposed for the design of energy systems in which heat and electricity carriers are coupled. Multi agent systems (MAS) have been proposed as a promising technology for addressing control and co ordination issues in the power industry [7], [8]. MAS can be defined as systems comprising more than one intelligent agent, i.e. an intelligent controller [8]. MAS are classified as a distributed control architecture, which fits very well with the control of DER [7]. In addition, MAS have been proposed as an approach for controlling clusters of CHP units [9]. ECOTEC 21 (WP3.6) 2014, Dr. Spyros Skarvelis Kazakos 1
This work is addressing the issue of control and coordination of multiple CHP generators in a future scenario where penetration of such resources is sufficient to cover a large part of the load in a local area (e.g. a neighbourhood). Energy management systems for such high CHP generator penetrations must have increased co ordination capabilities, in order to deal with the increased complexity. 2.1 Resource aggregation The approach for coordinating several energy resources is referred to as aggregation. There are two main concepts defined, micro grids and Virtual Power Plants: Micro grids: local groups of domestic customers, who may have micro generators and/or other distributed energy resources installed domestically [2]. Virtual Power Plants: defined as larger control structures used to control distributed energy resources. They are not restricted by locality. An aggregated resource is created, which acts as a single resource [10]. In such control structures, an aggregator is responsible for the overall strategy of the aggregated resource and several local controllers communicate with the aggregator, in order to implement the strategy locally. These controllers can be represented by intelligent agents, as in Figure 1. The agents are intelligent software located in controllers, which are able to take decisions and communicate with each other for co ordinating their activities based on (a) their own individual goals, as well as (b) the overall objective. Local agents are aware of all the local parameters of the resource, such as operational limits, availability, etc. Intermediate layers of aggregation can be introduced, to distribute the structure complexity and the intelligence of the agents. In Figure 1, this is shown as the micro grid aggregator. VPP Aggregator A A Micro Grid Aggregator A A A A A G G G A G A A A A G G G G A Agent G Micro Generator Interaction Figure 1 VPP agent structure 2.2 Multiple energy carriers and Energy Hubs Multiple energy carrier systems, or otherwise referred to as multi carrier systems, stem from the need to evolve traditional electricity, gas and other energy systems to more flexible, integrated energy systems [11]. In order to accomplish that, integration of multiple energy systems is necessary, both in the design as well as the analysis phase [5], [12]. ECOTEC 21 (WP3.6) 2014, Dr. Spyros Skarvelis Kazakos 2
The concept of energy hubs, laid out in [4] and [12], presents an integrated approach for optimizing systems with multiple energy carriers, such as electricity, hydrogen, biofuels, or Natural Gas. The energy hub is defined as a unit where multiple energy carriers can be converted, conditioned and stored [4]. Hence, an energy hub may include devices for converting energy between different carriers, such as a CHP unit converting Natural Gas to electricity and heat. It may also include storage such as batteries or thermal storage. The energy hub that will be considered in this work is illustrated in Figure 5, as an example. Power input (port m) Power output (port n) Electricity (P α ) Tx WT Electricity (L α ) Renewable energy (P β ) Natural Gas (P γ ) ν 1 ν 3 ν 5 ν 2 ν 4 PV FC MT B Heat (L β ) Tx Transformer PV Photovoltaic WT Wind Turbine MT Microturbine (CHP) FC Fuel Cell (CHP) B Boiler Figure 5 Energy hub with three inputs in port m and two outputs in port n [4]. Power in an energy hub is input and output through a number of different energy carriers (see Figure 5). Energy conversion between different carriers takes place inside the energy hub through conversion devices such as CHP generators, power electronics converters, or heat exchangers. Storage devices such as batteries or thermal storage tanks can also be part of an energy hub. However, all the conversion and storage elements have certain efficiencies which are accounted for in the hub model. By formalizing the interactions between energy carriers, a whole system analysis and optimization is possible, through regulating the inputs necessary to supply the required power outputs / loads [12], [13]. In Equation (1), the relationship between the inputs (P m ) and outputs (L n ) is defined by the forward coupling matrix (C mn ). In Equation (2), the inverse relationship is defined by the backward coupling matrix (D nm ). Subscripts such as m and n denote different input / output ports in the same energy hub, with different coupling characteristics between them [12]. The elements of matrices C mn and D nm incorporate a dispatch factor ν, which indicates, in percentage, the proportion of the corresponding input that is fed to the particular conversion element of the hub. The efficiencies of the conversion elements are also included in the matrices [12]. ECOTEC 21 (WP3.6) 2014, Dr. Spyros Skarvelis Kazakos 3
L c c P (1) L c c P output L n C mn input P m d d L P (2) P d d L input P m D nm output L n As an example, considering the energy hub in Figure 5, a proportion (ν 3 ) of Natural Gas is supplied to the fuel cell, a proportion (ν 4 ) to the microturbine, whereas the rest (ν 5 ) is supplied to the boiler. The sum of these dispatch factors must equal 1, i.e. ν 3 + ν 4 + ν 5 = 1. A more specific example can be found in [12]. Taking into account Equations (1) and (2), the total energy input required to satisfy the desired energy output can be derived [12]. By modifying the dispatch factors of the individual elements in the energy hub and consequently in the C mn or the D nm matrix, the total input P m can be minimized. An objective function can then be defined as follows: min P L D (3) 3 Objectives The following objectives were defined, in order to develop the agent based energy management system for CHP engines: 1) Define a concept for combining energy hubs with distributed control architectures. 2) Formulate a methodology for implementing multiple energy carrier optimisation within the framework of a distributed control architecture. 3) Develop a distributed control algorithm for multiple energy carrier optimisation. 4) Write initial code for testing the methodology. 5) Validate the developed methodology and code with a case study. 4 Approach 4.1 Energy Management System Implementation 4.1.1 Programming language/platform Several implementations of agent based distributed control systems in power engineering have been developed in the Java Agent DEvelopment (JADE) platform, which is Java based [14]. Most agent platforms, including JADE, conform with the standards set by the Foundation for Intelligent Physical Agents (FIPA) [15]. This enables the efficient interaction of agents developed in different platforms. The JADE platform will be considered for the purposes of this study, as it has been used in the implementations found in [7] and [16], as well as the previous implementations of the project partners [17]. Intelligent agents can be developed to control the output power of the CHP generators that they are attached to. ECOTEC 21 (WP3.6) 2014, Dr. Spyros Skarvelis Kazakos 4
4.1.2 Control structure description This section defines the interaction of intelligent agents within energy hubs, proposing an agent based energy hubs approach. Each agent is linked to an element in the energy hub. Again, considering the example of Figure 5, one agent would be assigned to each of the following: microturbine, fuel cell, transformer, photovoltaic, wind turbine and one to the boiler. Each of these agents will hold detailed information on the state and characteristics of the device it is linked to (e.g. CHP heat to power ratio, wind speed, solar intensity, etc). A hierarchical aggregation structure is proposed to enable the scalable aggregation of the energy hubs. A commercial aggregation entity would be the highest level of aggregation, which would be able to interact with electricity, gas, ancillary services or emissions markets, functioning as a Virtual Power Plant, as described in [17] and [18]. The agent based control structure includes four different agent types, with the following functions: Hub element agents: These may represent any element in the Energy Hub. They are able to control generation and other components. They are aware of the real time parameters of their associated element for each of the relevant energy carriers, such as conversion efficiencies, availability and responsiveness. Hub agent: Controls the energy hub. It optimizes the inputs and outputs of all the energy carriers in the energy hub, according to the parameters of the hub elements and user preferences and needs. Aggregator agent: Trades energy carriers and services in relevant markets and may request adjustments to specific energy carrier generation (P m ) or demand (L n ) from the Hub agents (e.g. more electricity, less gas), in accordance to trading contracts. A diagram illustrating the architecture of the system is presented in Figure 6. Aggregator agent Α Α Α Α Hub agents Α Α Α Α Α Α Α Α Hub element agents MT PV FC Tx WT Energy Hub Α Agent Interaction Tx Transformer WT Wind Turbine MT Microturbine (CHP) FC Fuel Cell (CHP) PV Photovoltaic Figure 6 Agent based control architecture for energy hubs ECOTEC 21 (WP3.6) 2014, Dr. Spyros Skarvelis Kazakos 5
4.1.3 Energy carrier optimisation in an Energy Hub The Hub agent would be responsible for optimizing the energy hub. Three different objectives can be considered in the proposed optimization process: a) Minimize energy input: Minimizing Equation (3), in order to minimize total energy consumption at the input port for the given loads, improving overall hub efficiency. b) Minimize cost: Modification of Equation (3) to include the cost of each of the energy carriers. Each row of the product of L n D nm is multiplied with a cost factor (e.g. 0.4 per m 3 of Natural Gas). The modified objective function is then minimized against total cost. c) Minimize output emissions: Similar to cost optimization, each row of the product of L n D nm is multiplied with an emission factor (e.g. 1.875 kgco 2 /m 3 of Natural Gas). The modified objective function is then minimized against total CO 2 emissions. The optimization will be constrained by a number of parameters, as described in [13], the most relevant being the operational constraints of energy hub devices, i.e. minimum maximum power ratings of generators, power electronic converters, etc. The optimization calculations take place in the Hub agent, using an optimizer. The Hub agent needs to collect the necessary data from all the hub element agents. Then, it constructs the L n and D nm matrices and inputs the objective function to the optimizer, in the form of Equation (3). The parameters that can be varied are the dispatch factors. The optimal dispatch factors are then sent to the hub agents so that they can act upon them. A flowchart describing the energy hub optimization algorithm is presented in Figure 7. Start Wait [time interval] Hub agent records / discovers number of element agents and updates database Hub agent requests from element agents the following: (i) efficiencies and (ii) types of energy inputs / outputs Hub agent constructs forward and backward coupling matrices (C mn and D nm ) Hub agent performs optimization of P m matrix, minimizing Equation Hub agent sends final dispatch factors (ν) to element agents and optimal P m matrix to aggregator agent Figure 7 Energy hub optimization algorithm In order to facilitate the interaction of multiple Hub agents, the optimization process cannot be continuous. A multi period optimization has to be set up, as seen in [6] and [19]. The optimization algorithm would be processed at set time intervals (e.g. every 30 minutes), each time optimizing the objective function for the next time period between intervals. ECOTEC 21 (WP3.6) 2014, Dr. Spyros Skarvelis Kazakos 6
4.2 Case Study 4.2.1 System description and input data The implementation described above was developed in Java. A case study was built, in order to test the functionality of the implementation. Two optimisation targets were considered: (i) cost and (ii) CO 2 emissions. The studied system is a microgrid, based on the system described in [2] and [20]. The microgrid includes a total generating capacity of 63kW, constituted by the distributed generation (DG) units in Table 1. The inputs considered are (i) renewable energy in the form of wind or solar energy, (ii) grid electricity and (iii) natural gas. The outputs that were considered were (i) electricity and (ii) heat. A large boiler was considered as a backup source of heat and the electricity grid as a backup source of electricity. The CHP units and the boiler were considered to be fed by natural gas and the electricity output was considered to be linked to the electricity input through a transformer. Table 1 Energy Hub elements, their efficiency and power characteristics [2, 20, 21]. Hub element Electrical Thermal Electrical Thermal efficiency (%) efficiency (%) output (kw) output (kw) Photovoltaic 15 10 Wind turbine 40 13 Fuel Cell 40.4 56.6 10 14 Microturbine 25.9 67.34 30 78 Large boiler 90 3072 Grid transformer 98 500 The cost was arbitrarily considered to be 0.02 / kwh both for the grid electricity and the natural gas. Emission factors were taken as 430 gco 2 /kwh for grid electricity and 184 gco 2 /kwh for natural gas input [21]. Daily half hour electrical load profiles were taken from [22] and thermal load profiles from [23]. The electrical load profiles were scaled to the maximum electrical load of 116.4 kva at a power factor of 0.85, as in [2]. The thermal load profiles were scaled to 8 kw per household [23], for 22 units (20 households and 2 service areas) [2]. Optimisation of the energy hub was performed for every half hour step using the JOM (Java Optimization Modeler) package, with the IPOPT non linear optimiser option [24]. 4.2.2 Results The results are presented in Figures 8 11. In Figures 8 and 10, a breakdown of the primary inputs is given, on a hub element basis. Figures 9 and 11 show the value of the objective function output, i.e. the total cost (Figure 9) or the total emissions (Figure 11) of the system. Primary power input (kw) 350 300 250 200 150 100 50 Natural Gas (Large Boiler) Natural Gas (Microturbine) Natural Gas (Fuel Cell) Renewable Energy (PV) Renewable Energy (Wind) Grid Electricity 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 Half hour step Figure 8 [cost minimization] Breakdown of the primary power input by resource ECOTEC 21 (WP3.6) 2014, Dr. Spyros Skarvelis Kazakos 7
2.5 Total cost of energy hub ( ) 2 1.5 1 0.5 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 Half hour step Figure 9 [cost minimization] Total cost of energy inputs per half hour Primary power input (kw) 350 300 250 200 150 100 Total emissions of energy hub (kgco2) 50 0 1 3 5 7 9 11131517192123252729313335373941434547 Half hour step Figure 10 [emissions minimization] Breakdown of the primary power input 30 25 20 15 10 5 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 Half hour step Figure 11 [emissions minimization] Total emissions of energy inputs per half hour 5 Analysis of results Natural Gas (Large Boiler) Natural Gas (Microturbine) Natural Gas (Fuel Cell) Renewable Energy (PV) Renewable Energy (Wind) Grid Electricity The following observations can be made from the results presented above: The distribution of the primary power input across the different energy hub elements is almost identical in both the cost and the emissions optimisation. The use of renewables is maximised in both cases, since the cost as well as emissions during their operation is considered as zero, assuming that life cycle costs and emissions are not taken into account. Due to the limitations of the local generators ECOTEC 21 (WP3.6) 2014, Dr. Spyros Skarvelis Kazakos 8
(CHP and renewables), the grid electricity is used as a complementary resource, to fill in the gaps in electricity supply. Likewise, the backup boiler is operated only when the thermal load exceeds the capacity of the CHP units. The output profile of the objective function in the case of emissions optimisation is skewed more towards the electrical load profile, whereas in the cost optimisation towards the electrical load profile. The reason for this is that the cost of electricity and natural gas is considered equal, whereas the emissions factor of the grid electricity is higher than natural gas. The results above validate the operation of the developed system, proving that it is able to optimise the operation of multiple generation units (CHP and renewable) with multiple energy carrier inputs within the context of an energy hub. 6 Conclusions and future work This report presented an approach of control and optimization of energy hubs with distributed energy resources, which utilise multiple energy carriers. A multi agent system was described, which facilitates the coordination of multiple devices converting energy between carriers. The operational procedure of the multi agent system has been described and its fundamental elements have been illustrated. Each individual energy hub would be responsible for optimizing energy carrier input according to associated loads. An implementation was presented, combining agent based control with the energy hubs / multiple energy carriers concept, and was validated against a case study. The main benefit of the developed system is that it combines the flexibility, resilience and extensibility of multiagent systems, with the inclusivity of energy hubs. It is expected that future energy systems will not be dominated by a single energy carrier such as electricity, but will also involve other energy carriers, such as hydrogen or biogas. Optimizing the operation of the whole system will remove inefficiencies and allow the utilization of the optimal combination of energy carriers at any given moment. The next step of this research is the implementation of the control system described in this paper in a suitable platform, utilizing microcontrollers. Subsequent plans include experimental demonstration and testing in a laboratory environment, utilizing a small CHP unit and other distributed energy resources. 7 References [1] T. S. Ustun, C. Ozansoy, A. Zayegh, (2011), Recent developments in microgrids and example cases around the world A review, Renewable and Sustainable Energy Reviews, Vol. 15, No. 8, pp. 4030 4041 [2] S. Papathanassiou, N. Hatziargyriou and K. Strunz, (2005), A benchmark low voltage microgrid network, CIGRE Symposium, Athens, 13 16 April 2005 [3] S. Daniel, S. Skarvelis Kazakos, P. Jain, (2013), Local smart DC networks and distributed storage for reducing and shifting peak load, 22nd International Conference on Electricity Distribution (CIRED), Stockholm, 10 13 June 2013 [4] M. Geidl, G. Koeppel, P. Favre Perrod, B. Klockl, G. Andersson, K. Frohlich, (2007), Energy hubs for the future, IEEE Power and Energy Magazine, Vol. 5, No. 1, pp.24 30, Jan. Feb. 2007 [5] M.T. Rees, J. Wu, B. Awad, J. Ekanayake, N. Jenkins, (2011), A total energy approach to integrated community infrastructure design, 2011 IEEE Power and Energy Society General Meeting, 24 29 July 2011 [6] M. Chaudry, N. Jenkins, G. Strbac, (2008), Multi time period combined gas and electricity network optimization, Electric Power Systems Research, Vol. 78, No. 7, July 2008, pp. 1265 1279 ECOTEC 21 (WP3.6) 2014, Dr. Spyros Skarvelis Kazakos 9
[7] A.L. Dimeas and N.D. Hatziargyriou, (2005), Operation of a Multiagent System for Microgrid Control IEEE Transactions on Power Systems, Vol. 20, No. 3, pp.1447 1455 [8] S.D.J. McArthur, E.M. Davidson, V.M. Catterson, A.L. Dimeas, N.D. Hatziargyriou, F. Ponci and T. Funabashi, (2007), Multi Agent Systems for Power Engineering Applications Part I: Concepts, Approaches, and Technical Challenges, IEEE Transactions on Power Systems, Vol. 22, No. 4, pp.1743 1752 [9] K.H. van Dam, M. Houwing, Z. Lukszo, I. Bouwmans, (2008), Agent based control of distributed electricity generation with micro combined heat and power Crosssectoral learning for process and infrastructure engineers, Computers & Chemical Engineering, Vol. 32, No. 1 2, January 2008, pp. 205 217 [10] D. Pudjianto, C. Ramsay and G. Strbac, (2007), Virtual power plant and system integration of distributed energy resources, IET Renewable Power Generation, Vol. 1, No. 1, pp. 10 16 [11] P. Favre Perrod, M. Geidl, B. Klockl, G. Koeppel, (2005), A vision of future energy networks, 2005 IEEE Power Engineering Society Inaugural Conference and Exposition in Africa, 11 15 July 2005 [12] M. Geidl, G. Andersson, (2005), A modeling and optimization approach for multiple energy carrier power flow, 2005 IEEE Russia Power Tech, 27 30 June 2005 [13] M. Geidl, G. Andersson, (2007), Optimal Power Flow of Multiple Energy Carriers, IEEE Transactions on Power Systems, Vol. 22, No.1, pp.145 155 [14] F.L. Bellifemine, G. Caire and D. Greenwood, Developing multi agent systems with JADE, John Wiley and Sons, 2007, ISBN 9780470057476, England. [15] IEEE Foundation for Intelligent Physical Agents (FIPA), Agent Communication Language Specifications, found in http://www.fipa.org/repository/aclspecs.html [16] S.D.J. McArthur, E.M. Davidson, V.M. Catterson, A.L. Dimeas, N.D. Hatziargyriou, F. Ponci and T. Funabashi, (2007), Multi Agent Systems for Power Engineering Applications Part II: Technologies, Standards, and Tools for Building Multi agent Systems, Power Systems, IEEE Transactions on, Vol.22, No.4, pp.1753 1759 [17] S. Skarvelis Kazakos, E. Rikos, E. Kolentini, L.M. Cipcigan, N. Jenkins, (2013), Implementing agent based emissions trading for controlling Virtual Power Plant emissions, Electric Power Systems Research, Vol. 7, pp. 1 7, September 2013 [18] A.L. Dimeas and N.D. Hatziargyriou, (2007), Agent based control of Virtual Power Plants, International Conference on Intelligent Systems Applications to Power Systems (ISAP) 2007, 5 8 November 2007 [19] S. Skarvelis Kazakos, P. Papadopoulos, I. Grau, A. Gerber, L.M. Cipcigan, N. Jenkins and L. Carradore, (2010), Carbon Optimized Virtual Power Plant with Electric Vehicles, 45th Universities Power Engineering Conference (UPEC), Cardiff, 31 Aug 3 Sept 2010 [20] S. Skarvelis Kazakos, B.A. Giwa, D. Hall, (2014) Microgrid power balancing with redox flow batteries, 5th IEEE PES International Conference and Exhibition on Innovative Smart Grid Technologies (ISGT Europe 2014), Istanbul, 12 15 October 2014 [21] S. Skarvelis Kazakos, L.M. Cipcigan and N. Jenkins, (2009), Micro Generation For 2050: Emissions Performances Of Micro Generation Sources During Operation, Pollack Periodica, Vol.4, No.2, pp.89 99 [22] DTI Centre for Distributed Generation and Sustainable Electrical Energy, United Kingdom Generic Distribution System (UKGDS), http://www.sedg.ac.uk/ukgds.htm [23] S. Abu Sharkh, R.J. Arnold, J. Kohler, R. Li, T. Markvart, J.N. Ross, K. Steemers, P. Wilson and R. Yao, (2006) Can microgrids make a major contribution to UK energy supply?, Renewable and Sustainable Energy Reviews, Vol. 10, No. 2, pp. 78 127 [24] P. Pavon Mariño, JOM (Java Optimization Modeler), http://www.net2plan.com/jom/index.php ECOTEC 21 (WP3.6) 2014, Dr. Spyros Skarvelis Kazakos 10