Abstract

The proposed multi-agent energy management system is aimed at controlling distributed generation and performing demand side management to provide robust and effective grid with lower emissions and costs. Furthermore, appropriate strategy used in combination with the multi-agent energy management system would decrease peak demands in the wider grid. The proposed solution consists of single-type agents who control one or more gird entities which are represented as generic sub-agent elements. The agent applies one control algorithm across all elements and uses cost function to evaluate the suitability of the element as a supplier. This study shows the ability of the multi-agent energy management system to control grid entities represented by generic sub-agent elements and select suppliers according to the selection behaviour set by the user. C entralised generation and control are prone to losses in transmission and hence, lower efficiency, causing more emissions. Growth in peak demand on the grid also means a greater stress on the transmission network. One of the most popular solutions is to decentralise and increase the penetration of generation across the grid. DG has the potential to improve the efficiency and minimise the cost of energy generation [1]. However, the maximum effect can be achieved only by applying suitable control strategies and methods on DG. Furthermore, a load shifting/shedding strategy coherent with DG control would decrease peak demands in the wider grid and contribute to the reliability of the system. Multi-agent system is a suitable candidate for controlling DG and performing demand side management, as the nature of the agents is autonomous and pro-active [2] Keywords- Current trends in the development of power generating devices show anticipation of a high penetration of distributed micro-generation that will increase the complexity of the distribution network Recently developed multi-agent solutions are based on different types of agents. These include agents specific for control functions This structure would allow more than one entity of a grid to interface with an agent and to apply a single control algorithm across all agents in the LV distribution network. The rest of the paper is organised as follows: In section 3, the architecture and organisation of the MAS used for this study is presented, including explanation of the sub-agent model and an example of sub-agent elements. Section 4, describes the case study and agents used to evaluate the suitability of sub-agent elements to represent grid entities in the multi-agent energy management system. Section 5 presents the results of the test scenarios. Finally, the paper is ended with sections future work and conclusions. III. MUL II-AGENT ENERGY MANAGEMENT SYSTEM The purpose of the multi-agent energy management system is to apply decentralised control and optimisation of the grid by the means of autonomous agents, where agent is designed to control an object within the grid. The object can be an entity that has only generating capabilities (e.g. distributed generation station), is an electricity consumer (e.g. a typical household) or is a hybrid of generator and consumer (e.g. household with micro-generation and local storage facilities). The task of the agent is to maintain supply/demand balance within the object whilst minimising operational cost of the elements and trading energy with neighbouring agents. Furthermore, by giving preference to the DG located within Agent LV distribution network and utilising local storage, agents are able to minimise the peak demand on the main grid and operate in islanded mode if needed. In case when demand exceeds available supply, agent can also apply demand side management by performing load shifting/shedding and hence, reducing the demand. When agent is searching for the power supplier, the participants of the bidding process are only the neighbouring agents, but the generators, located within the same agent as the load, are also consider as suppliers. If such generator exists within an agent it will have higher chance of being selected as supplier due to smaller transition cost -hence, better efficiency. However, this requires a special structure of the agent. A. Agent structure As mentioned in the introduction, present multi-agent approaches consist of different types of agents for each element. The proposed model of a multi-agent energy management system has only one type of agent, where each entity connected to the agent is represented as a generic element The generic nature of the elements gives the possibility to apply dynamic control strategies on different agents within the network. The user of the agent specifies the type of the control strategy and the agent sets the corresponding weights to the Initially agent estimates the internal balance, which includes internal consumption and generation as well as power exports and imports to the neighbouring agents. Once the internal balance is negative, the agent evaluates the generating capabilities of internal generators and requesting costs of generation from the neighbours. During the bidding process the agent communicates with neighbours in the L V network to exchange the information regarding power requirements and power availability. Once the initial information exchange is complete, each agent creates a new element which represents the power generating and consuming part of the neighbouring agent. The communication is provided through an ACL module in JADE B. Sub-Agent element The main purpose of introducing a generic sub-agent element is to make an agent versatile in interfacing to various grid entities and applying the same control algorithm to all attached elements. Therefore, the parameters of an element must be common for major types of entities in the grid. The parameters of an element and their description are given in The capacity of the element is mostly used to determine the most suitable energy source; a local storage device with very low power capacity is unlikely to be selected as a suitable supplier. Most types of generators can have a near infinite capacity, given the fuel is provided on time. However, for some renewable power sources, the capacity term could provide the predicted availability based on, for example, the forecasted wind speed for wind turbines or remaining daylight time for photo-voltaic cell. The list of parameters can be extended to accommodate new entities of the grid and new parameters can be easily mapped to the existing entities. Furthermore, additional parameters can be added to the sub agent element in order to apply a new control strategy that requires additional information about the elements (e.g. reactive power control, scheduling or applying social factors) C. System model and cost/unction/or a sub-agent element Since the purpose of this study is to evaluate the feasibility of using generic elements for multi-agent control of grid entities, the model representing these entities is assumed to be a perfect input/output system. The model has the following form: Where Yt,e = ut,e ' y � ax (1) The price per kWh, �e ' and impact factor (measured in kg of C0 2 per kWh), Ee are weighted according to the behaviour set by the user. Currently there are three behaviour options: "normal", "cheap" and "green". For the "normal" behaviour the weights for the parameters are normalised and produce equal effect on the cost function. For the other two behaviours the appropriate parameter has a greater weight in order to increase its effect on the cost function. Once the costs for all available elements are calculated, the best supplier element is given by: If the supply/demand balance remained negative after all available generators are operational, the agent will perform load shedding function and switch off load with the lowest The multi-agent energy management system is implemented using JADE. The framework is designed to simplify the implementation and development of multi-agent systems by providing a software platform that complies with FIPA specifications [13]. JADE also provides tools for debugging and deployment phases and supports ACL as the basis for agent communication architecture. The agent consists of two main components: GUI and agent's body. The purpose of the GUI (see figures 2 and 3) is to provide control over the elements attached to the agent, set the supplier selection behaviour and monitor the power exchange with neighbours as well as other statistics. The body of the agent deals with all of the communications, servicing elements and maintaining book keeping. Thanks to the ACL module in JADE communications are easy to handle and requires only performing appropriate actions to informative and content of the message. Since all the agents are identical in structure, every message that agent sends it must produce the appropriate reply and consequently process the reply. Servicing of the elements includes estimating the demand and generating capabilities within the agent and complying with the element model when operating an element (i.e. current output does not exceed maximum output, load does not operate beyond specified duration period etc) Book keeping maintains the record of the major actions of the agent. These include: changing operational states of the