MULTI AGENT BASED CONTROL AND PROTECTION FOR AN INVERTER BASED MICROGRID

Transcription

1 MULTI AGENT BASED CONTROL AND PROTECTION FOR AN INVERTER BASED MICROGRID Asitha Lakruwan Kulasekera (108020B) Thesis submitted in partial fulfilment of the requirements for the degree Master of Science Department of Mechanical Engineering University of Moratuwa Sri Lanka February 2012

2 DECLARATION I declare that this is my own work and this thesis does not incorporate without acknowledgement any material previously submitted for a Degree or Diploma in any other University or institute of higher learning and to the best of my knowledge and belief it does not contain any material previously published or written by another person except where the acknowledgement is made in the text. Also, I hereby grant to University of Moratuwa the non-exclusive right to reproduce and distribute my thesis/dissertation, in whole or in part in print, electronic or other medium. I retain the right to use this content in whole or part in future works (such as articles or books). Signature: Date: The above candidate has carried out research for the Masters/MPhil/PhD thesis under my supervision. Signature of the supervisor: Date: Signature of the supervisor: Date: i

3 Table of Contents List of figures... v List of tables... vi List of abbreviations... vii 1. Introduction Emergence of smarter electrical grids and microgrids Smart grid Smarter microgrids Use of multi agent systems Thesis statement Methodology Contributions Organization Literature Review A Multi agent system Features of a multi agent system Multi agent systems and applications Toolkits/frameworks for development of multi agent systems JADE and Zeus a comparative overview JADE agent building toolkit JADE architecture Recent applications of multi agent systems in microgrids Distributed control Electricity trading ii

4 Optimization Power restoration Control architectures Conceptual design of dual-layered MAS Dual layered architecture Agents used in the dual-layered MAS Implementation of dual-layered MAS Dual-layered Multi agent system implementation in JADE Agent specification Application analysis Application design MAS algorithm implementation Running multi agent system application Simulations and Results Description of simulation circuit Intentional islanding during a fault Messages exchanged among agents and physical entities Updates to the ontology Results and discussion Protecting critical loads during intentional islanding Messages exchanged among agents and physical entities Updates to the ontology Results and discussion Managing critical loads during islanded operation Messages exchanged among agents and physical entities iii

5 Updates to the ontology Results and discussion Conclusions and Future Directions Conclusions Future work References Section 4: Impact of Research results... Error! Bookmark not defined. Section 5: Publications... Error! Bookmark not defined. iv

6 LIST OF FIGURES Figure 1.1 A smart grid application... 2 Figure 1.2 Conceptual model of a microgrid... 4 Figure 2.1 Example of a multi agent system... 9 Figure 2.2 JADE architecture Figure 2.3 Basic multi agent system structure Figure 2.4 Single layered multi agent system architecture Figure 3.1 Proposed dual layered MAS architecture Figure 3.2 Interaction between agents in the dual layered multi agent architecture.. 25 Figure 4.1 Dual-layered MAS ontology Figure 4.2 Control algorithm for multi agent system operation Figure 4.3 Establishing communication between simulation environment and MAS36 Figure 5.1 Simulated microgrid test-bed Figure 5.2 Test-bed set-up for case study Figure 5.3 Control flow chart for intentional islanding Figure 5.4 Key messages exchanged between agents during islanding Figure 5.5 Line to line voltages during intentional islanding Figure 5.6 Test-bed set-up for case study Figure 5.7 Line to line voltages during securing critical loads Figure 5.8 Test-bed set-up for case study Figure 5.9 Control flow chart for load management during islanded operation Figure 5.10 Line to line voltages during islanded operation Figure 6.1 Proposed microgrid test-bed v

7 LIST OF TABLES Table 2.1 A comparison between JADE and ZEUS platforms Table 4.1 Microgrid domain knowledge, attributes and initial values Table 5.1 Updated concepts, initial and updated values for case study Table 5.2 Updated concepts, initial and updated values for case study Table 5.3 Updated concepts, initial and updated values for case study Table 5.4 Priority revision during islanded mode vi

8 LIST OF ABBREVIATIONS Abbreviation DG DSM FIPA FOSS GHG GUI JADE MAS NCRES OS PC PCC PV PWM SCADA Description Distributed Generation Demand Side Management Foundation for Intelligent Physical Agents Free and Open Source Software Green House gasses Graphical User interface Java Agent Development Framework Multi Agent Systems Non-conventional Renewable Energy Sources Operating Systems Personal Computer Point of Common Coupling Photovoltaic (cell) Pulse Width Modulation Supervisory Control and Data Acquisition vii

9 SECTION 3 1. INTRODUCTION An agent is a software entity which can operate on its own by responding and reacting with its environment. A collection of such agents collaborating to achieve an overall goal can be considered as a multi agent system (MAS). Such systems are ideal for the complex real-time control problems that are faced with today. Multi agent systems are being widely used in computer systems and e-commerce applications [1], [2]. The collaborative social ability of agents is enhanced by the presence of fast communication capabilities in such applications (i.e. web/internet). With the integration of fast communication capabilities into power grids in building future-proof smart(er) grids, the possibility of integrating multi agent systems to power engineering is becoming real [3]. The unique feature set available in multi agent systems is proving to be the correct fit in terms of the requirements in power engineering applications. This section provides an introduction to the research area, which the application of multi agent systems in microgrid applications. The importance of this work is briefly explained and is followed by the thesis statement Emergence of smarter electrical grids and microgrids The ever increasing demand for electrical power has been a major challenge for the electrical supply authorities worldwide [4]. The present grid systems which have mostly been built over the last century are rapidly aging [4]. These legacy grid infrastructures are becoming increasingly congested and are seen as incapable of meeting the future energy needs of an information economy. The renewed interest in climate change mitigation through reduced Green House Gas (GHG) emissions has ushered in a need of greener generation and utilizing renewables in the energy mix. Therefore, these power systems have to become smarter, more reliable and more robust in taking on renewable energy sources without losing stability and efficiency. 1

10 Smart grid A smart grid is the use of sensors, communications, computational capabilities and control to enhance the overall functionality of the electric power grid [4]. This concept is depicted in Figure 1.1. Figure 1.1 A smart grid application Smarter grids are a step forward in overcoming the limitations of legacy electricity infrastructure[4]. They must be automated and be able to behave as a distributed energy delivery system, that incorporates two-way flow of electricity & information including advanced sensing & monitoring [4] along with the following functionalities [5]. Visualizing the power grid in real time Increasing system capacity Relieving bottlenecks Enabling a self-healing network Enabling enhanced connectivity to consumers Allowing incorporation of distributed non-conventional renewable energy sources (NCRES) 2

11 One rising feature integrated into smart(er) grids is the microgrid concept [6]. It allows for far greater control over improving grid reliability [5], resilience towards a wide range of threats to the electrical grid [7], allow for easy incorporation of distributed generation [8], reduce GHG emissions etc. Therefore, in developing the future smarter grid, implementation of successful microgrids are vital Smarter microgrids Smarter microgrids are part of the solution to the problems faced by legacy electrical infrastructure. Their capability to house local embedded generators as distributed generation and be locally controlled offers various advantages over legacy systems. In order to realize the potential of such smarter systems, a high degree of distributed control is required [6]. As a novel method in providing this flexible distributed control requirement, Multi Agent Systems (MAS) are stepping forward. Centralized bulk generation facilities are giving way to, more distributed small scale generation. Distributed generation (DG) introduces the capability of embedding a wide range of low emission and potentially lower cost generation options. Such options can include internal combustion engines, micro-turbines and nonconventional renewable energy sources such as solar photo voltaic, wind, biomass, mini/micro-hydro and even waste-to-energy systems. These smaller generation technologies allow the sources to be placed optimally with respect to the loads, thus reducing emissions and transmission losses, while also allowing for local control instead of central dispatching [6]. Such DG systems provide the potential to take a sub-system approach in working with microgrids. During a disturbance in the main grid, local generation and their loads can be separated from the utility and isolate the local loads from the disturbance, while maintaining the integrity of the main utility grid. This isolation is carried out at a single point of connection to the main utility known as point of common coupling (PCC). This capability, known as islanding, requires high reliability and flexibility from the microgrid. These requirements can be met by having peer-to-peer control and plug-and-play capabilities for each entity in the microgrid. 3

12 A conceptual microgrid model is shown in Figure 1.2. This can be either a small scale distribution network or an electrical system of a building. Certain loads, considered critical, will have local power sources. Non-critical loads will have no local supply. The critical loads can be isolated from upstream outages by islanding at the PCC and shedding the excess demand of non-critical loads. Figure 1.2 Conceptual model of a microgrid 1.2. Use of multi agent systems Most conventional power systems rely on Supervisory Control and Data Acquisition (SCADA) systems for control and communications. The control and communication architecture provided by such SCADA systems is built upon proprietary communication and signalling protocols. In terms of a microgrid, the aim is to allow for flexible integration of a multitude of different distributed energy resources (DER) and/or non-conventional renewable energy sources (NCRES). However the vendor specific control protocols used in commercially available SCADA systems for microgrids (i.e. RSView32 [9], Networked Distributed Resource and Central Operation Management System) restrict the interconnectivity and interoperability of 4

13 diverse systems. Such restrictions can result in increased deployment costs and prevent microgrid rollout. Use of multi agent systems to provide distributed control capabilities to the microgrid applications offers various advantages over conventional SCADA systems. Multi agent system development platforms are available as Free and Open Source (FOSS) applications and most of them are based on Java, making them platform independent. They can also be combined with external programming to interface with different external hardware and software systems [1]. These advantages make multi agent systems more suited for microgrid control. In addition, multi agent systems also have the following advantages in context of microgrid control: i. Multi agent systems are inherently flexible and extensible. ii. Local loads and sources will have different specifications and systems. Therefore, an intelligent distributed control system is the best solution. iii. The microgrid will have to take rapid autonomous decisions regarding seamless transition to islanded mode from grid connected mode and protecting critical loads by load shedding. iv. Multi agent systems allow for a complex task to be broken down into several smaller tasks assigned to a team of agents. This allows for easier handling of a larger problem Thesis statement The objective of this research is to design, develop and simulate a multi agent system that enables real-time energy management of a microgrid. This will include the management and control procedures during transition from grid-connected to islanding mode; the algorithms to manage priority levels of the connected loads and protect critical loads utilizing available limited capacity from local embedded generators during outages. Novel MAS which provides scalable and robust distributed control architecture for microgrid application is presented in the thesis. The MAS model is designed using 5

14 the JADE (Java Agent Development framework) platform and is implemented on a simulated test-bed in MATLAB/SIMULINK. The steps taken to realize the objectives of the thesis statement are listed below, and are further detailed in Section 1.3.1: 1. Identification of suitable agent building framework 2. Design of Multi agent system architecture 3. Development of Multi agent system 4. Integration of the multi agent system and the microgrid simulation 5. Validation of results using simulation studies Methodology the methodology consists of the following five steps: Step 1: Identification of suitable agent building framework Several commercial and open-source multi agent system building toolkits are currently available today for developing complex multi agent system in which the design phase has been simplified [10], [11]. Such tool kits have been assessed in literature in terms of time required for analysis and design, code generation, integration with external code, response time, the number of messages exchanged among agents etc. Further details regarding the selection of a suitable toolkit is discussed in Section 2.4. Step 2: Multi agent system architecture Design The multi agent system aims to provide control and protection system for an inverter based microgrid. The aim is to break down a larger goal into several smaller tasks and delegate them to individual entities rather than to a single controller. Further details on designing the multi agent architecture is presented in Section 3. 6

15 Step 3: Development of dual-layered multi agent system Developing a multi agent system using a selected tool kit consists of following steps. Initially, the abilities of each agent have to be specified, followed by identifying each of their roles and responsibilities, through building role models and defining social and domain responsibilities. Then the ontology of the system has to be created based on modelling their knowledge or facts. All of these are concluded by creating the agents and generating the code. This process is detailed in Section 4. Step 4: Integration of the multi agent system and the microgrid simulation The developed multi agent system attempts to control physical entities or a simulation of the same. The multi agent system and the simulation are inherently in two domains. The multi agent system runs on a JAVA platform while the simulation is run on a simulating environment (i.e. MATLAB/SIMULINK). Therefore, in order to establish communication between the two domains, for sensing and control TCP/IP is used. A third party TCP/IP server is utilized as an interface to MATLAB and it connects to another middle server which allows multiple simultaneous TCP/IP connections. The middle server is required because; the interface allows only a single connection at a time. Details regarding the creation of this interface are given in Section 4.2. Step 5: Validation of results using simulation studies Three case studies are carried out to verify the implementation of the multi agent system within a simulated microgrid test-bed environment. They demonstrate the functionality of the system in islanding, protecting/securing supply to critical loads and managing load priorities during islanded mode operation. The case studies and their results are presented in detail in Section 5. Details of the microgrid test-bed used for simulations are given in Section

16 1.4. Contributions In summary, the proposed multi-agent system control architecture is designed, developed and simulated in several simulation case studies. It is expected that this research will provide an insight into the design and development of a multi-agent system. Furthermore, this research also contributes novel control architecture to increase effectiveness of multi-agent system in the context of microgrids. Specifically, this comprises control algorithms for managing the available limited supply from local embedded sources to protect local critical loads during fault/s in the main grid Organization Rest of the thesis is organized as follows. Section 2 summarizes related work. Sections 3 and 4 explain the methodology presented in Section 1 in more detail. Section 5 presents the case studies and results, and the final section presents conclusions and future work. 8

17 2. LITERATURE REVIEW An agent system is hardware or software entity that possesses social coordination and communication capabilities and capable of achieving a larger overall goal by tackling smaller individual tasks. Agents have the ability to operate within its environment to achieve its goal by sharing knowledge with other agents or taking initiative. Even a single agent can operate as a system, by reacting and interacting with its environment. Such an agent will require specific programming to provide itself with individual knowledge, actions and goals. Any other agents in the same environment will not have any effect on the operation of such an agent A Multi agent system A collection of social, collaborative agents constitute a multi agent system. All the agents will inherit a set of common goals, and react to change in the environment, make decisions to achieve those goals, or help other agents in achieving their goals [1]. In contrast to single agent systems, other agents will also be a part of any agent s environment. An example multi agent system is described in Figure 2.1. It shows a community of agents in a stock market environment. Figure 2.1 Example of a multi agent system Source: Manage the Complexity with Intelligent Agent Technology ( 9

18 The selling agent only has direct knowledge of potential buyers and maximizing the profit. Other agents will know about their own domains. The trading rules agent knows of all the rules and regulations; the risk assessment agent is able to assess the risk associated with any stock; Data mining agent is able to access and understand patterns in the sales data and the forecasting agent is able to forecast the rise and fall of stock prices. The selling agent can make a more educated decision on trading a particular stock by collaborating with his fellow agents, than on his own. He can communicate with the other agents in his environment and ask for any supporting information that will enable him to maximize his sales and/or profits Features of a multi agent system As the aim of this thesis is to implement a multi agent system for distributed control (intelligence), the following available feature set is ideally suited for such applications. Autonomy agents are capable of operating without human supervision. Social ability agents are able to communicate with each other as well as human operators using a common language. Reactivity Proactivity agents are able to identify and react to changes in the environment. agents take decisions or initiatives based on their assigned objectives/goals. Agents also hold certain knowledge or beliefs regarding their existence or their environment. This knowledge and beliefs enable the agents to take decisions regarding changes in their environment and seek to fulfil their objectives. In addition to these features, agents have all or some of the following capabilities as well. 10

19 Data collection Social knowledge Learning Communication an agent contains or will be able to collect data to build-up knowledge regarding its environment, and assist fellow agents making their decisions. the multi agent system contains an agent who helps the other agents to know about each other s existence and share their capabilities (i.e. yellow pages service). agents have learning capabilities to update their knowledge and beliefs based on changes in the environment, and updates from fellow agents. all agents in the multi agent system should have a clearly defined set of protocols regarding their communication with humans or other agents Multi agent systems and applications Multi agent systems are becoming more and more ubiquitous in the world. Recently, multi agent systems are being implemented in a many fields such as mathematics, physics, engineering sciences, and social science communities [12]. They have been used in a wide array of applications ranging from computing and processing [13 16], robotics and control systems [17 21], transportation [22 25], aerospace and nuclear engineering in recent literature. There is a rising trend of applying multi agent systems in power engineering applications for microgrids which is discussed in detail in Section

20 2.3. Toolkits/frameworks for development of multi agent systems A multi agent system can be built from base coding. However, this is not necessary as there are many easily available agent development tool kits where the agent development process has been simplified. There are a number of open-source agent development toolkits available, such as: Aglets software development kit [26], Voyager [27], Zeus [11], JADE [10], Tracy [28], SPRINGS [29], and Skeleton [30]. Performance analysis of these agent development toolkits has been comprehensively discussed in [31 38] in terms of time taken for analysis and design, code generation, integration with external code, response time, the number of messages exchanged among agents etc.. Based on these evaluations JADE and ZEUS are initially short listed as potential contenders. Further analysis carried out to select the best suited platform is presented in the following Section 2.4. Step 1: Identification of suitable agent building framework 2.4. JADE and Zeus a comparative overview In selecting an agent tool kit / platform for an open-source development application it is important to select one which is based on a common standard. The established standard for agent systems is IEEE s standard on Foundation for Intelligent Physical Agents (FIPA) [39]. Based on the development platform of a well-known standard will allow for easier inter-operability between different systems, resulting in universal applicability. Therefore, the main criteria in selecting the best suited toolkit for this application are compliance with FIPA standards. Several studies in the literature [31], [33], [37] have compared the performance of JADE and ZEUS in terms of availability, design and analysis, number of lines of code, response time etc. A comparison of JADE and ZEUS platforms are given in Table

21 Table 2.1 A comparison between JADE and ZEUS platforms Free and open source FIPA Compliant JAVA Based JADE Agent Communication Language (ACL) Communication Provides authentication Decent GUI Lower latency in complex systems ZEUS Free and open source FIPA Compliant JAVA Based ACL and Knowledge Query Manipulation Language (KQML) Some security capabilities User friendly GUI Higher latency in complex systems JADE and Zeus along with JATLite and Skeleton are comprehensively evaluated in [31], grouping them together based on Java and availability of GUI. These four are evaluated for use in a web-based news retrieval system. The author has evaluated their performance based on run-time and software building capabilities. The evaluated parameters in terms of software building characteristics are: time required for analysis and design of multi-agent system; time required for code generation; time required for integration of agent with external code; time required for testing and debugging the system; number of lines in the code and number of new Java classes and reused classes to build news agents in the system. Selected parameters for evaluating run-time characteristics are: response time and number of messages exchanged among agents. While the results show both to be similar in terms of software building, JADE outperforms ZEUS in terms of run-time performance. JADE, ZEUS and Skeleton are compared in terms of number of documents requested from a web server versus response time while varying the number of web-agents in [33], which show JADE outperforms all others for complex applications. 13

22 In applying a multi agent system for a microgrid or power application, complex systems maybe required with (near) real-time performance. Therefore response time and the number of messages exchanged become critical. A delay of 0.1 s caused by excess messages will be visible on the power distribution. Therefore JADE becomes the most suitable tool for the application JADE agent building toolkit JADE (Java Agent Development Framework) is an open source agent development toolkit developed and distributed by Telecom Italia under the LGPL (Library Gnu Public License) license [10]. It is a software development platform which provides middleware functionalities independent of application and simplify agent creation [1]. This middleware functionality [40] is provided by: i. run-time environment where the agents exist ii. vast class library for simplifying agent development iii. suite of graphical tools that allow for simple administration and monitoring of agents JADE is based on the ubiquitous JAVA platform which makes it platform independent and highly versatile. Details of the JADE architecture are presented in Section JADE architecture Each JADE run-time environment in operation is called a container, which can contain several agents. A set of containers are called a platform. Each platform must have a primary or main container, and all other secondary containers must register with it as soon as they are created / initialized. These secondary containers are called normal containers and must be told where to find the main container in their platform. If another main container is created it will constitute a different platform. The agents in these containers can communicate with one another, even if they are in 14

23 the same container, different containers in the same platform or on different platforms, using the network protocol stack provided by JADE. These notions are depicted in Figure 2.2 [40]. When the main container is started, it also includes the following two special agents (these are automatically started): AMS (Agent Management System) Agent: this agent provides a unique name (Agent ID or AID) to each agent in the platform (naming service) and represents the authority in the platform to create/kill agents on remote containers. DF (Directory Facilitator): this agent provides a Yellow Pages service, using which an agent can find other agents with capabilities / information it requires in order to achieve its goals. Figure 2.2 JADE architecture In Figure 2.2, A1 to A5 denote agents in two platforms 1 and 2. Agents A2 and A3 are in the same container, Agents A1 and A4 are in different containers, while all of 15

24 them are in the same platform. Agent A5 is in the main container of a different platform. Each main container hosts its own AMS and DF. Agents may need to carry out parallel negotiations and each negotiation will need to proceed at its own pace. Normally for a software entity/program this can be done by assigning a thread. Modern OS designs limit the number of concurrent threads. Therefore, in order for efficient parallel operation of agents, JADE provides a concept called behaviours [41]. Behaviour represents a task that an agent can carry out and is implemented as an object of a class that extends jade.core.behaviours.behaviour. A behaviour is essentially an event handler [42], a method which defines how an agent will react to a given event. An event can be defined as a change of state in the environment. In JADE, behaviours are classes and any code written to handle a given event will be placed in a method called action. As behaviours are methods, they are processed one after the other by an agent s thread. Normally, this requires complex programming to allow threads to pause in the middle of execution to listen for a response and continue without losing context (active and passive phases). JADE simplifies this process by allowing the easy creation of different behaviours for all active phases in an activity, and arranging them to be created and triggered in the correct sequence [43]. Each behaviour class must implement the following abstract methods: Action( ) actual task to be accomplished by the specific behaviour class Done( ) returns true when the behaviour has finished and false when the behaviour has not and re-executes action Reset( ) restarts a behaviour from the beginning An agent can be created by extending the agent class and providing agent specific capabilities by programming one or more behaviour subclasses [41]. JADE provides a wide range of ready-made behaviours for common agent tasks such as sending and receiving ACL messages, and breaking down complex tasks as collections of simpler tasks. Behaviours can be added to the agent by the addbehaviour( ) method of the agent class. Behaviours can also be added at any time when an agent starts up [using the setup( ) method] or from within other behaviours. 16

25 As FIPA specifications [39] define, in order for agents communications to be successful, they must share the same language, vocabulary and protocols. JADE simplifies this by allowing the creation of Agent ontologies. A JADE ontology defines the vocabulary and semantics for the content of messages exchanged between agents, and is built up of three main interfaces (components); Concept, AgentAction and Predicate [43] Recent applications of multi agent systems in microgrids Analysis of recent literature revealed that the majority of the work has focused on distributed microgrid control, while the other categories included: electricity trading (market model analysis), optimization and restoration Distributed control The work concentrated on distributed control amounts to more than half of the recent literature [44 66]. This depicts the recent trend in using multi agent systems for development of distributed control for microgrids. Among the research, various architectures and frameworks are presented with simulations showing their effectiveness in controlling a microgrid. A basic hierarchical MAS which will act as a basis for further development is presented in [44] to demonstrate the applicability of MAS as a distributed control framework. Artificial neural networks and fuzzy systems have been used for generation planning and load forecasting for operation planning and an adaptive control based on traditional PI (Proportional-Integral) control that has been used in MAS-based microgrid control model [57]. A real-time power management and control system proposed in [52] is capable to optimize microgrid systems based on multiple objectives, such as power demands, fuel consumption, environmental emissions, costs and dispatchable loads. In [49] a control architecture is given which is based on a hierarchically equal agent structure without a central agent able to resolve real and reactive power mismatch within microgrids. Though there is no 17

26 central agent, a neighbour-to-neighbour three-step communication algorithm is used to determine the real and reactive power mismatches. Autonomous operation in island mode is proposed using a Multi-Agent Reinforcement Learning Algorithm that utilizes the concept of layered learning in [53], [56]. The concept of layered learning is used to group various controls and actions of the agents depending on their effect on the environment and for the agents to ultimately coordinate in achieving their goals. MAS controller for the MAS control architecture is given in [47], [50]. An intelligent load controller is described in [53] based on the same system. Further, the ability of MASs to achieve efficient use of NCRES and green technologies is also described. In [65], a three tiered hierarchical coordinated control strategy which aims to utilize the MAS most efficiently for voltage stability and power balance in the microgrid has been proposed. A controller for bilateral switching between grid-connected operation and islanded operation is simulated in [55] for monitoring and control of a microgrid. Intelligent Distributed Autonomous Power Systems (IDAPS) of Virginia Tech [63] illustrates the capability of MAS s to be considered as a software alternative to traditional hardware-based zonal protection that can be used for effective islanding. Their system demonstrates the MAS s ability to disconnect and stabilize the microgrid while being able to allow for redefinition of zonal boundaries on the fly [63]. A control algorithm comprising of an intelligent connection agent and grid synchronization system is used in [59] to ensure the appropriate microgrid disconnection from the main grid. This is used to improve the performance of microgrid and its interaction with the main network, or another microgrid, under grid voltage transients. MAS based DG controllers have been presented in [46], [51], [54] and [61]. A multiagent based DG microgrid control framework is presented in [46] for a DC distributed energy system that can be used in a microgrid as a modular power generation unit. Two-layered control architecture is presented in [51] to achieve local 18

27 autonomy and global optimization respectively, operating in both grid-connected mode and islanded mode. A multi-agent models used in [54], [61] take each element in the microgrid as a separate autonomous intelligent agent and are simulated using the Real Time Digital Simulator (RTDS) and the PowerWorld Simulator respectively. A smart grid (SG) should ideally be built up of microgrids. Therefore a multimicrogrid system is essentially a SG system. Multi-microgrid systems inherit selfhealing, intentional islanding, and the anticipation of the entire distribution network from SGs and the ability for dynamic behaviour from microgrids. MAS based control systems have been proposed for multi-microgrids in [60], [64]. The incorporation of Service Oriented Architecture (SOA) approaches such as web services into MASs are proposed in [58], [66]. The architectures presented are able to dynamically adapt to changes, work together to solve complicated problems and be flexible to accommodate for new behaviours with greater agility. A hybrid MAS is presented in [45] where the controller is able to intelligently choose the operation model, self-control the optimal operation of microgrid, monitor realtime data and provide autonomous local protection. The several MAS control architectures found in the literature contained a similar basis or basic arrangement. A summary of these control architectures is discussed in Section Electricity trading MASs have been applied for electricity trading or market model analysis in [67 72]. A power market model is introduced in [67] for the effective operation of the microgrid. A MAS electricity trading algorithm is proposed in [68] to maximize the revenue from the microgrid. A case study is used in [69] on price determination based on demand and supply side bidding strategies to propose a pricing mechanism for Micro Grid energy in the competitive electricity market. The same mechanism is 19

28 developed in [70] by using a microgrid central controller to participate in the bidding process to settle the market-clearing price (MCP). The need to include economic and political measures, rather than only technological parameters in the future electricity architecture is elaborated in [71]. This is done using a MAS which is able to incorporate the mechanisms of a prevailing electricity market for high integration of NCRES into microgrids. A retailing spot market of electric energy proposed in [72] is able to interoperate with other stakeholders in the utility grid using a cooperating MAS Optimization MAS based controllers have been used in [48], [73 76] for optimizing microgrid operations. As loads and sources within a microgrid can be diverse and distributed, real-time reaction and DG source management is critical in preventing local power outages. It is also essential that this is done most efficiently and cost effectively as possible for the microgrid to be economically viable. The systems presented in [48], [76] consider the characteristics of the source or load types and self-regulates with other agents in order to globally optimize in terms of cost and efficiency. The optimization MAS in [73] is able to determine the optimal operation of a solarpowered microgrid considering the load demand, environmental requirements and PV panel and battery capacities. A competitive pricing mechanism is also presented, using auction theory and two bidding techniques (i.e. single bidding and discriminatory bidding) in order to serve the consumers at a reduced price and to provide better revenues. In [74], power production is maximized with respect to production cost and DG unit constraints. Demand Side Management (DSM) is also utilized via a load control agent. The MAS is able to monitor energy resources and schedule generation (dispatch control) for optimized microgrid operation. An artificial immune system based algorithm is used by the MAS in [75] to optimize the production of the local DGs in order to optimize the microgrid operation. 20

29 Power restoration MASs have been used for power restoration of microgrids in [77 80]. The load restoration algorithm in [79] consists of agents that make synchronized load restoration decisions according to information learned from their direct neighbours. Though only the direct neighbours are contacted, global information is discovered based on the Average-Consensus theorem [81], [82]. Thereafter, the load restoration problem is modelled and solved using algorithms for the 0-1 Knapsack problem. Power restoration in the case of a large-scale blackout in a microgrid with DGs is discussed in [77], [80]. The models used the available capacity of DGs to reduce the blackout area. A MAS using the Agent-Environment-Rules (AER) ontology and the Constraint Satisfaction Problem (CSP) based model is used in [77]. An asynchronous backtracking algorithm is used to solve the AER model based on CSP. In [80], a multi-agent immune algorithm is used for rapid restoration of power. A method called a random tree method is used to code antibodies into the immune algorithm to prevent unfeasible solutions. High frequency mutation is carried out for the antigens to be dynamically changed in order to rapidly converge to the solution. A hierarchical control strategy for power restoration is given in [78] targeting swift power restoration to the most critical loads Control architectures Intelligent agents can be assigned to each entity in a microgrid; i.e. circuit breaker, generation unit, loads, users etc. This assignment will determine the MAS architecture. Several different architectures for microgrid applications have been proposed up to now. Almost all available architectures can be termed single layered architectures. Especially, as an example, single layered architectures have been proposed for microgrid applications such as resolving real and reactive power mismatch [49] and load restoration [79]. Several multiple level control architectures have been presented in the literature [51], [65]. However, in these multi layered topologies, the layers address different strata of the distribution grid and not of the control architecture. 21

30 A summarization of the basic MAS structure used in distributed control approaches in the literature is given in Figure 2.3. Three main agents shown are common in all related work. A DG agent handles the addition, operation and control of all local DG units. A user agent acts as or connects to a user interface where utility/consumer users interact with the system. Last, but not least, a control agent will oversee all these and any other related control tasks, while mainly connecting to/operating circuit breakers. Two main types of control agents are seen; overall MAS controllers [47], [50] and load controllers [53]. Figure 2.3 Basic multi agent system structure This basic architecture can be implemented over a conceptual microgrid consisting of several different DG units and critical and non-critical local loads as shown in Figure 2.4. Database agent facilitates message parsing (exchange) and storage, acting as the access point for data for all the agents and users. Here the control agent is tasked with islanding as well as load management. Some drawbacks of the single layered architecture can be listed as: Certain agents get overloaded with objectives Such agents become critical, and their failure can be catastrophic Lower flexibility in terms of expanding the microgrid and/or controller 22

31 Figure 2.4 Single layered multi agent system architecture 23

32 3. CONCEPTUAL DESIGN OF DUAL-LAYERED MAS This section presents the conceptual design of multi agent architecture in detail. Step 2: Multi agent system architecture conceptual design 3.1. Dual layered architecture The single layered multi agent systems have the drawback of agents needing cater multiple objectives. This drawback can be overcome by properly delegating the multiple objectives to different agents. The proposed multi agent architecture achieves this task delegation by creating multi-layered multi agent system architecture. This novel architecture is presented in Figure 4.1. Strategic (Primary) Level Makes run-time decisions Control Agent Tactical (Secondary) Level Execution of run-time decisions I/O operations User Agent DG Agent LV Agent Load Agent Utility (Background) Operations Stores, and shares data Data access point for all DB Agent Figure 3.1 Proposed dual layered MAS architecture The control hierarchy is split in to two, creating dual-layered control architecture. The primary layer (or the strategic layer) comprises the Control agent who makes the run-time decisions. The secondary (tactical) layer contains the User Agent, DG 24

33 Agent, LV Agent and Load Agent. This layer carries out the execution of the runtime decisions of the strategic level, which mainly comprises input/output operations. Both these layers are ultimately supported by a service (utility) provided by a DB (Database) agent, running in the background. This utility agent stores system information and data and messages shared between the agents and acts as a data access point for all agents. The agent assignment of the novel dual-layered architecture is given in Figure 3.2. This introduces the new secondary control layer in comparison to the single layered architecture given in Figure 2.3. The agents depicted in the figure are briefly introduced in Section 3.2. Figure 3.2 Interaction between agents in the dual layered multi agent architecture The process of developing this dual-layered multi agent system and implementing it on the simulation test-bed is given in Section 4. The different case studies carried out and their respective results are given in Section 5. 25

34 3.2. Agents used in the dual-layered MAS The proposed dual-layered MAS shown in Figure 3.2 comprise the control agent, LV agent, Load agent, DG agent, User agent and DB agent: Control agent: This agent monitors the main grid for faults and/or outages and controls the secondary layer in the MAS. It oversees the operation of the secondary layer which is tasked with handling loads and the low voltage (LV) connection. It also informs the agents in the MAS whether the system is islanding capable or not. Ultimately, the control agent will be the centre of the primary layer holding influence over the DG agent and controlling the LV (Low Voltage) agent and Load agents. LV agent: This agent controls the islanding operation by opening the circuit breaker at the PCC. It stores the islanded or grid-connected status of the PCC. The LV agent can also be given metering capabilities if a net-metering scheme is used for the microgrid. Load agent: Each individual load has an individual load agent. It stores information regarding each load and control the connect/disconnect operation. Information such as load identification number, load priority and current demand are stored for each load. The LV agent can also be given metering capabilities if DSM (Demand Side Management) schemes are to be used within the microgrid. DG agent: This agent is responsible for storing information regarding the DG unit, metering and controlling the output and connecting/disconnecting the unit to/from the microgrid. The information stored regarding the DG includes; i. Identification number ii. Connection status (connected/disconnected) iii. Type of equipment (IC engine, micro-turbine, Solar PV, Fuel cell etc.) iv. Rated power (in kw) v. Output power (in kw) vi. Local energy source (Fuel) availability (given as infinite) vii. Unit availability (given as always available not under maintenance) viii. Energy cost (in Rs./kW) 26

35 User agent: This agent behaves as the gateway for the user to interact with the system, to obtain real-time information and set system goals. This agent collects information from other agents regarding demand and supply and makes it available to the system users. The user agent can also interface with a GUI (Graphical User Interface) for easy interaction. Database Agent: stores system information and data and messages shared between the agents. The priority list used for load management is also hosted by the database agent. The database agent acts as the access point for data for all the agents and users. These agents operate collaboratively to achieve the overall goals of the system (i.e. island the microgrid during a fault, protect supply to critical loads and provide dynamic load management.) 27

36 4. IMPLEMENTATION OF DUAL-LAYERED MAS Following section describes the process of implementing the dual-layered architecture introduced in Section 3.1. The methods and process of developing and implementing the dual-layered multi agent system using JADE is explained including features of multi-agent systems and the JADE toolkit. This section also describes the process of establishing TCP/IP communication for running the multi agent system on the simulated test-bed. Step 3: Developing the dual-layered multi agent system 4.1. Dual-layered Multi agent system implementation in JADE The multi-agent system implementation using JADE can be broken into the following main steps: i. Agent specification ii. Application analysis iii. Application design iv. Implementation Agent specification Here, the specifications, requirements and tasks for the agents in multi agent system are defined. The agents to be used in the multi agent systems were introduced in Section

37 Application analysis The operation of the multi agent system is defined in this step. The target is to develop a multi agent system that achieves the objectives put forward in Section 1.3. The MAS is expected to provide the following capabilities to the microgrid: i. Dynamic interaction with the microgrid (Rapid islanding in case of an upstream fault) ii. Sharing limited capacity of local DG unit/s between multiple loads iii. Load management and prioritization In order for the agents to follow the objectives, how they collaborate with one another must be clearly defined. The agent collaboration process is described below. Agent collaboration 1 Main-grid / microgrid Stimulus control agent receives information from main grid and microgrid 2 Respond to microgrid main Control agent responds the microgrid 3 Inform all agents control agent informs all registered subscribers about the information from microgrid 4 Command loads sends control commands to its corresponding loads 5 Command DG send control commands to the DG unit 6 Load information receives load information from their corresponding loads 7 Control DG Power production control agent informs DG agent of how much power to produce 8 Information on DG output DG unit informs DG agent about the amount of power produced 9 Manage loads control agent manages its corresponding load circuitry based on locally produced power 29

38 Application design Using the JADE toolkit to allow for agent to communication with each other and reason about facts and knowledge in relation to the microgrid domain is carried out by the following steps [1]: i. Define an ontology including the schemas for the types of predicate, agent action and concept that are pertinent to the microgrid domain. ii. Develop proper Java classes for all types of predicate, agent action and concept in the ontology. iii. Register the defined ontology and the selected content language with the agent. iv. Create and handle content expression as Java objects that are instances of the classes developed in step ii and let JADE translate these Java objects to/from strings or sequences of bytes that fit the content slot of ACL Messages. Details of the ontology creation and concept identification are discussed as the first step in application design. 30

39 Ontology creation The detailed structure of the ontology for the dual-layered MAS is given in Figure 4.1. The ontology, as an example, has a concept named LV (LV Agent), with a predicate called Islanded (addressing whether grid connected or not) and an agent action called Disconnect (to open the circuit breaker at PCC). Figure 4.1 Dual-layered MAS ontology Concepts can either be a role of an entity or a common property. Here the concepts represent the agents that make up the multi agent system. A concept can include subconcepts as shown in Figure 4.1 (i.e. the concept, Load, includes the subconcepts, Load ID, Demand, Priority, and Connected). Predicates house the status of different concepts that can either be true or false. Four predicates are defined in the ontology; Islanded, Connected, SupplyGTDemand and 31

40 CapacityGTDemand. Islanded and Connected predicates are used by LV agent and Load agents respectively to define the state of their circuit breakers as open or closed. SupplyGTDemand verifies whether supply greater than or equal (>=) demand in the case of controlling DG output during load management in islanded mode. CapacityGTDemand is used for load shedding for load management to verify if the available local capacity is greater than or equal to the total local demand. Agent actions are special concepts that indicate actions that can be performed by particular agents. For example, the agent actions, connect and disconnect are used by LV and Load agents to close and open their circuit breakers. Software based development of the ontology is specific to the toolkit and detailed information regarding this process can be found in the online tutorial [43]. The process consists of the following steps. i. Define the vocabulary of the agents communication space ii. Define the java class that specifies the structure and semantic of the object MakeOperation ( ) iii. Define the schema of the object iv. Register the ontology and the language that are used for assembling and parsing (or coding and decoding) the content of messages with the agent's content manager 32

41 Concept identification The next step in the application design consists of modelling the knowledge that will be used by the agents. This knowledge is prescribed as the concepts used in the ontology, their attributes and initial values. The multi agent system carries physical as well as several abstract concepts. Being in islanded mode is an abstract concept, and power demand/available power is a physical concept. A list of concepts defined is given in Table 4.1. Table 4.1 Microgrid domain knowledge, attributes and initial values Concept Attributes Initial Value islanded islanded: Boolean pcc_state: Boolean islanded: false pcc_state: true DG_info dg_id: String connection_state : Boolean type : String rated_power : float local_fuel : Integer unit_availability : Integer energy_cost : float dg_id: 01 connection_state : true type : IC rated_power : 15 local_fuel : 100 Unit_availability : 100 energy_cost : 0 Load_x _info (x = 1 or 2) load_id: String load_priority : float power_demand: float connected : Boolean load_id: 1 / 2 load_priority : 0.7 / 0.4 power_demand: 7.5 / 7.5 connected : true / true DG_Ctrl output_power : float output_power : 7.5 Ctrl_DG power_reqd: float power_reqd: Ctrl_Load_x shed_loadx : Boolean shed_loadx : false CB_A (Circuit Breaker A) cba_state : Boolean cba_state : true CB_B cbb_state : Boolean cbb_state : true CB_C cbc_state : Boolean cbc_state : true CB_D cbd_state : Boolean cbd_state : true The technical/programing aspects of next steps (steps ii iv) are not discussed here and they can be found in detail with examples in pages in [1]. 33

42 MAS algorithm implementation The multi agent system implementation is discussed in this section. The multi agent system allows the microgrid to operate in grid connected mode as well as islanded mode. The multi agent system is implemented by developing the control algorithm depicted below in Figure 4.2. Figure 4.2 Control algorithm for multi agent system operation Assumptions: It is assumed that unlimited power is available from the grid and local embedded generator has unlimited fuel to withstand an outage of any duration. The embedded generator is assumed to be always online supplying part of the local demand, and capable of instantly changing the output depending on changes in local critical demand. 34

43 The control agent measures the main grid voltage and frequency, and detects any fault or outage based on the measurements. If the measurements exceed the standard thresholds in ±10% for voltage and ±0.5 Hz for frequency, the control agent switches the microgrid to islanded mode operation. When the decision to switch to island mode is taken, all the agents are notified and they cooperate and coordinate with the control agent to achieve the objectives set forth. Based on the prescribed MAS_Ontology, the agents can easily start their interactions and accomplish their goals through the exchange of ACL messages. Three main objectives the multi agent system has to meet (which are defined in Section 1.3) can be simplified as; i. dynamic interaction with the microgrid (Rapid islanding in case of an upstream fault) ii. sharing limited capacity of local DG unit/s between multiple loads iii. load management and securing supply based on load priority Operation of each of the objectives is simulated and the success of the multi agent system is verified in Section 5. The control algorithm is explained in detailed in the respective simulation section. 35

44 Step 4: Integration of the multi agent system and the microgrid simulation 4.2. Running multi agent system application The multi agent system attempts to control physical entities or a simulation of the same. This puts the multi agent system and the simulation into two different domains. In order for the multi agent system to work on the simulated environment establishing communication between the two domains is required. The multi agent system runs on a JAVA platform while the simulation is run on a simulating environment (i.e. MATLAB/SIMULINK). Therefore, in order to establish communication between two domains, for sensing and control TCP/IP is used. A third party TCP/IP server [83] is utilized as an interface to MATLAB and it connects to another middle server which allows multiple simultaneous TCP/IP connections. Each agent requires a separate connection to the test-bed. The interface allows only a single connection at a time; therefore a separate middle server is required. This process is depicted in Figure 4.3. Figure 4.3 Establishing communication between simulation environment and MAS The middle server allows various agents in the multi agent system to establish multiple simultaneous TCP connections and multiplexes all of them through the single TCP connection to the TCP server. Both servers can be implemented on the same PC running the simulation or on a separate PC on a connected network. 36

45 5. SIMULATIONS AND RESULTS Step 5: Result validation using simulation studies Three case studies are developed to verify the capability of the proposed MAS to provide control and protection for an inverter based microgrid. Within conventional power systems, frequency control is handled by rotating inertial masses of large generators. This causes a problem within smaller microgrids. Therefore, an inverter based system is selected, where the frequency control is handled by the inverter interface. Details of the simulation test-bed used in the case studies are presented in Section 5.1. Case study 1: Intentional islanding during an upstream fault In this case study, an upstream fault is introduced on the main grid system at 0.05 s into the simulation. Upon the detection of the fault, the MAS switches from grid connected mode to islanded mode of operation. The transition is carried out within 0.02 s of detecting the fault. This verifies the capability of the MAS in rapidly initiating a seamless transition. This is further discussed in Section 5.2. Case study 2: Protecting critical loads during intentional islanding During an outage or fault on the main grid, the microgrid has to cater local demand based solely on the local sources. In order to meet the local demand and maintain the stability of the system at the same time, some loads have to be shed. This is done by, shedding lower priority non-critical loads and protecting the higher priority critical loads. This is completed within 0.02 s of detecting the fault and initiating islanding transition. This verifies the capability of the MAS in rapidly protecting critical loads during intentional islanding. This is further discussed in Section 5.3. Case study 3: Managing critical loads during islanded operation In the islanded mode, as the locally available DG capacity is limited, some of the local loads have to be shed in order to maintain/protect the supply to higher priority (critical) loads. During such islanded operation, if power needs to be re-routed to a lower priority load by raising its priority, the MAS system allows for on-the-fly 37

46 priority revisions. This allows power to be re-connected to a previously shed load. This is also completed within 0.02 s of detecting priority revision. This verifies the capability of the MAS in rapidly managing load priorities during islanded operation. This is further discussed in Section Description of simulation circuit The simulations are carried out on an inverter-based microgrid test-bed developed using MATLAB/SIMULINK. The test-bed is built mainly using the SimPower Systems block set available in SIMULINK. The test-bed based on the simplified block diagram, shown in Figure 5.1, is used for the simulation case studies. The test-bed comprises an embedded generator, acting as a DG unit, supplying part of the local demand, and some critical and non-critical loads. It is assumed that the DG unit can operate at full capacity without fuel limitations during any outage. It is also assumed that the DG unit can instantly increase output using battery storage. As an inverter based microgrid is considered, a grid interface unit comprising an inverter, low pass filter, PWM (Pulse Width Modulation) and circuit breaker is used in the case of a DC source. The microgrid is connected to a main utility grid (11 kv) across a transformer (11kV/400V) and a main circuit breaker. Figure 5.1 Simulated microgrid test-bed 38

47 5.2. Intentional islanding during a fault The first case study is carried out to verify the effectiveness of the MAS in rapidly islanding the microgrid upon the detection of an upstream outage on the main grid. The ability of MAS to sense the upstream outage by detecting the voltage droop and initiate agent actions to seamlessly transition to islanded mode from grid-connected mode is addressed here. The test-bed used for the simulation is shown in Figure 5.2. Figure 5.2 Test-bed set-up for case study 1 The test-bed comprises a 15kW DG unit producing 480V, connected to a 400V, 50Hz microgrid, a transformer (480V/400V) connecting the DG to the microgrid, two 7.5kW local loads, with a connection to a 11kV main utility grid fed through another transformer (11kV/400V), all of which are connected across circuit breakers. Total local demand is 15 kw, 50% of which (7.5 kw) is supplied by the DG unit, while the rest is supplied by the main grid. It is assumed that the DG unit can instantly increase output using battery storage. An upstream fault is introduced to the system at t = 0.05 s into the simulation. At the beginning of the simulation breakers A, B, C and D are all closed. 39

48 Messages exchanged among agents and physical entities The control agent detects the fault in the upstream, takes the strategic decision to switch to island mode and informs the other agents to switch to islanded mode. The control flow chart used in the process is given in Figure 5.3. This is a reduced version of the algorithm given in Section Upon receiving this island command, the LV agent opens the main circuit breaker at PCC (point A) to isolate the microgrid from the utility. Figure 5.3 Control flow chart for intentional islanding 40

49 The control agent then queries local demand and available capacity data from the Load and DG agents respectively. As the current local DG supply is enough to cater the local demand the DG agent is asked to run at the present demand (increase output to 15kW from 7.5 kw). The messages exchanged between the agents are shown in Figure 5.4. Figure 5.4 Key messages exchanged between agents during islanding 41

50 Updates to the ontology The concepts known to the multi agent system is updated with every step. The agents are able to identify the changes in the environment and react accordingly based on these changes to the ontology. The updates to the ontology are: Table 5.1 Updated concepts, initial and updated values for case study 1 Concept Initial Attribute/s Updated Value islanded islanded: true pcc_state: true islanded: true pcc_state: false Load_1_info load_id: 1 load_priority : 0.7 power_demand: 7.5 connected : true load_id: 1 load_priority : 0.7 power_demand: 7.5 connected : true Load_2_info load_id: 2 load_priority : 0.4 power_demand: 7.5 connected : true load_id: 2 load_priority : 0.4 power_demand: 7.5 connected : true DG_Ctrl output_power : 15 output_power : 15 Ctrl_DG power_reqd: 15 power_reqd: 15 Ctrl_Load_1 shed_load1 : false shed_load1 : false Ctrl_Load_2 shed_load2 : false shed_load2 : false CB_A (Circuit Breaker A) cba_state : true cba_state : false 42

51 Results and discussion The test-bed shown in Figure 5.2 is simulated for 0.1 s. The simulation results presented in Figure 5.5 depicts the successful islanding process. During grid connected mode: While the microgrid is in grid-connected mode, the total demand is 15 kw. During the grid-connected mode the DG unit only provides 7.5 kw output, 50% of the total demand. Another 10 kw is required from the main grid to supply the microgrid. The main grid experiences a fault at t = 0.5 s. (a) Main grid voltage supply (b) Load voltage During transition: Figure 5.5 Line to line voltages during intentional islanding (a) Line to line voltage across the non-critical load at A, (b) Line to line voltage at breaker B When the upstream outage is detected at t = 0.05 s the control agent informs the LV agent and the Load agents to switch to island mode operation. Upon receiving the islanding order the LV agent trips the main circuit breaker A, at the PCC, isolating 43

52 the microgrid from the utility (Figure 5.5(a)). The Control agent then queries the DG agent and the load agents regarding their available capacities the DG can provide to the microgrid and the power requirements of the connected loads. As the available maximum capacity of 15 kw is able to meet the total demand of 15 kw, the control agent commands the DG agent to increase DG output to match the demand. Soon as the microgrid is put to island mode the supply to the loads are maintained (see Figure 5.5(b)). Islanded mode: At t=0.05s the microgrid is separated from the main grid and the local loads are secured. After the microgrid switches to island mode the total local demand is met by the DG unit by supplying 15 kw. All agent operations are carried out rapidly, from detecting fault, opening the main breaker, connecting the local source and shedding loads, to stabilize the microgrid within 0.02 s. Therefore the system is able to disconnect from the main utility grid and maintain the supply to the critical loads without suffering a brownout and/or blackout. 44

53 5.3. Protecting critical loads during intentional islanding The second case study addresses the capability of the MAS to protect/secure power to critical loads during intentional islanding. This should happen immediately following the transition to islanded mode. If power supply from available local DG unit is limited and insufficient to supply the total local demand, load shedding has to be carried out rapidly in order to secure supply to the most critical loads. The testbed used for the simulation is shown in Figure 5.6. Figure 5.6 Test-bed set-up for case study 2 The test-bed is same as the one used in case study 1, with only the following changes. Two local loads are 15 kw each with assigned priorities of 0.7 and 0.4 respectively. The total local demand is 30 kw; initially the DG unit supplies 10 kw while the other 20 kw is provided from the main grid. Similar to case study 1, an upstream fault is introduced to the system at t = 0.05 s in to the simulation. At the beginning of the simulation breakers A, B, C and D are all closed. 45

54 Messages exchanged among agents and physical entities The control process and messages exchanged between agents are based on the details given in Section 5. Further details regarding the control process can be found in Figure 5.3. Messages exchanged between the agents are as shown in Figure 5.4. As the local DG unit is unable to meet the total demand, load shedding is initialized based on the assigned priorities. This makes load connected to point C with the higher priority, the critical load. Therefore, a load_shed command is issued to the load agent for the lower priority, non-critical load at point D Updates to the ontology The concepts known to the multi agent system is updated with every step. The agents are able to identify the changes in the environment and react accordingly based on these changes to the ontology. The updates to the ontology are: Table 5.2 Updated concepts, initial and updated values for case study 2 Concept Initial Value Updated Value islanded islanded: false pcc_state: true islanded: true pcc_state: false Load_1_info load_id: 1 load_priority : 0.7 power_demand: 15 connected : true load_id: 1 load_priority : 0.7 power_demand: 15 connected : true Load_2_info load_id: 2 load_priority : 0.4 power_demand: 15 connected : true load_id: 2 load_priority : 0.4 power_demand: 15 connected : false DG_Ctrl output_power : 10 output_power : 15 Ctrl_DG power_reqd: 10 power_reqd: 15 Ctrl_Load_1 shed_load1 : false shed_load1 : false Ctrl_Load_2 shed_load2 : false shed_load2 : true CB_A (Circuit Breaker A) cba_state : true cba_state : false CB_D cbd_state : true cbd_state : false 46

55 Results and discussion The test-bed shown in Figure 5.6 is simulated for 0.12 s. The simulation results shown in Figure 5.7 depict the successful islanding and subsequent securing of critical loads. Grid connected mode: While the microgrid is in grid-connected mode, the total demand is 30 kw, with 15 kw critical loads and 15 kw non-critical loads. During the grid-connected mode the DG unit provides 10 kw output and the other 20 kw is provided by the main grid. The main grid experiences a fault at t = 0.5 s. Transition period: When the upstream outage is detected at t = 0.05 s the control agent informs the LV agent and the Load agents to switch to island mode operation. Upon receiving the islanding order the LV agent trips the main circuit breaker A, at the PCC, isolating the microgrid from the utility (Figure 5.7(a)). The control agent then queries the DG agent and the load agents regarding their available capacities that the DG can provide to the microgrid and the power requirements of the connected loads. As the total load of 30 kw exceeds the available maximum capacity of 15 kw, the control agent commands the load agents to shed the 15 kw non-critical loads to match the DG capacity and the DG unit to increase output to 15 kw. Soon as the microgrid is put to island mode the supply to the critical loads is maintained (see Figure 5.7(b)). The load agent at the non-critical loads sheds them from the system by opening breaker D (see Figure 5.7(c)). Islanded mode: At t=0.05s the microgrid is separated from the main grid and the load agents balance the local demand. After the microgrid switches to island mode the total local demand is met by the embedded generator supplying 15 kw. 47

56 (a) Main grid voltage supply (b) Critical load voltage (c) Non-critical load voltage Figure 5.7 Line to line voltages during securing critical loads (a) Line to line voltage across the non-critical load at A, (b) Line to line voltage at breaker C, (c) Line to line voltage at breaker D All agent operations are carried out rapidly; from detecting fault, opening the main breaker, connecting the local source and shedding loads, to stabilize the microgrid within 0.02 s. Therefore the system is able to disconnect from the main utility grid and maintain the supply to the critical loads without suffering a brownout and/or blackout. 48

57 5.4. Managing critical loads during islanded operation The third case study addresses the capability of the MAS to provide load management capabilities during islanded operation. As the available local capacity is limited and insufficient to cater total local demand during islanded mode, non-critical (lower priority) loads have to be dynamically shed in order to protect the stability of the microgrid. This simulation shows the ability of the MAS to allow for dynamic update of load priority levels. The test-bed used for the simulation is shown in Figure 5.8. Figure 5.8 Test-bed set-up for case study 3 The same test-bed used in Section 5.3 is used to simulate case study 3. The local loads are given two priority levels as shown in Figure 5.8. Local load 01 is given a priority level of 0.7, and load 02 is given a priority level is 0.4. This initial setting designates load 01 as a higher priority load while load 02 is considered a lower priority, by their respective load agents. At the beginning of the simulation breakers A, B, C and D are all closed. An upstream fault is introduced to the system at t = 0.05 s into the simulation, and the microgrid switches to islanded operation. During islanded operation the load 02 priority is raised to 0.9 at t = 0.14s. 49

58 Messages exchanged among agents and physical entities The objective of the test setup is to demonstrate the ability of the proposed MAS to island the microgrid after an upstream fault is detected, and maintain supply to the most critical loads with dynamic priorities. The control flow chart used in the process is given in Figure 5.9. Figure 5.9 Control flow chart for load management during islanded operation 50

59 Updates to the ontology The concepts known to the multi agent system is updated with every step. The agents are able to identify the changes in the environment and react accordingly based on these changes to the ontology. The updates to the ontology are: Table 5.3 Updated concepts, initial and updated values for case study 3 Concept Initial Attribute/s Updated Value islanded islanded: false pcc_state: true islanded: true pcc_state: false Load_1_info load_id: 1 load_priority : 0.7 power_demand: 15 connected : true load_id: 1 load_priority : 0.7 power_demand: 15 connected : false Load_2_info load_id: 2 load_priority : 0.4 power_demand: 15 connected : true load_id: 2 load_priority : 0.9 power_demand: 15 connected : true DG_Ctrl output_power : 15 output_power : 15 Ctrl_DG power_reqd: 15 power_reqd: 15 Ctrl_Load_1 shed_load1 : false shed_load1 : true Ctrl_Load_2 shed_load2 : false shed_load2 : false CB_A (Circuit Breaker A) cba_state : true cba_state : false CB_C cbc_state : true cbc_state : false CB_D cbd_state : true cbd_state : true 51

60 Results and discussion The test-bed shown in Figure 5.8 is simulated for 0.2 s. the simulation results shown in Figure 5.10 depict the successful islanding and subsequent securing of critical loads. Initially the microgrid is operating in the grid-connected mode, with the embedded generator is supplying only part of the local loads while the rest is supplied by the utility grid. The demands supplied initially by the DG and the grid are depicted in Figure 5.8. Figure 5.10 Line to line voltages during islanded operation (a) Main grid voltage at breaker A, (b) load 01 measured at C : switches from high priority to low priority, (c) load 02 measured at D : switches from low priority to high priority. 52

61 Grid connected mode: While the microgrid is in grid-connected mode, the total demand is 30 kw, comprising two 15 kw local loads. During the grid-connected mode the embedded generator only provides 10 kw, while the other 20 kw is supplied by the main grid to the microgrid. Transition period: The proposed dual layered MAS is able to successfully island the microgrid during the upstream outage within 0.02 s of detecting the fault in the main utility grid. As the local generation capacity is unable to meet the total local demand, the MAS initiates load shedding upon initially set priority levels of the loads as in case study 2 (see Section 5.3). This enables the MAS to maintain supply to the most critical load, load 01, by shedding the least critical load, load 02. Islanded mode: At t = 0.05 s the microgrid is separated from the main grid and the load agents balance the local demand as in case study 2 (see Section 5.3). After the microgrid switches to island mode the total local demand is met by the embedded generator supplying 15 kw. During the island operation, the user agent can revise the initial priority assignment. This can be pre-initiated or a user can change the priorities during the islanded mode. At t=0.14, the user agent revises the priority value of load 02 from p=0.4 to p=0.9 and communicates the revise_request to the load agent at load 02. This update is forwarded to other agents via the directory agent, and the control agent reinitiates load management procedures. As the DG unit is still unable to meet the total demand of 30 kw, and has to shed non-critical (lower priority) loads to maintain power to the new highest priority load, load 02. Therefore, the control agent commands the load agents to shed the load 01 of 25 kw to match the DG capacity. Thus, the supply to the new most critical load, load 02 is 53

62 provided by re-closing D (see Figure 5.10(c)). The load agent at the new lower priority load, load 01, sheds it from the system by opening breaker C (see Figure 5.10(b)). This process is shown in Table 5.4. Table 5.4 Priority revision during islanded mode Grid Connected mode Initial priority Islanded mode Revised priority Load 01 Priority = 0.7 Connected Priority = 0.7 Connected Priority = 0.7 Disconnected Load 02 Priority = 0.4 Connected Priority = 0.4 Disconnected Priority = 0.9 Connected When the load priorities are revised by a user via the user agent, the MAS is able to reconfigure the system, in order to provide power to the new most critical load. The reconfiguration is also done within 0.02 s of the revision and successfully reconnects the new critical load. The results show the capability of the MAS to safely island and maintain the supply to its critical loads, while allowing for the critical loads to be dynamically revised. 54

63 6. CONCLUSIONS AND FUTURE DIRECTIONS 6.1. Conclusions The developed MAS based control architecture for control and protection of an inverter based microgrid has been successfully verified using a simulated microgrid test-bed in MATLAB/SIMULINK. The MAS has been developed using the JADE platform using its available programming techniques and communication facilities. The control system has been endowed with seamless islanding, protection of critical loads and load management capabilities during islanded operation. Existing MAS based controllers for microgrid applications relied on a single layered architecture which tasked agents with multiple objectives. This drawback was overcome by developing novel dual layered architecture, where all tasks were properly delegated to individual agents. The MAS controller development phase involved application specification, analysis, design and implementation. This process created the agents that sense the changes in the environment and take initiative to achieve the overall goal put before them. This was achieved by creation of five main agents; control agent, user agent, DG agent, LV agent and load agent. These agents were supported by the DB agent running in the background. The agents were then connected to a simulated microgrid test-bed in MATLAB/SIMULINK over a third party TCP/IP server. Information regarding agent development and server implementation has been discussed. The MAS controller was simulated on a test-bed to demonstrate its ability to seamlessly island the microgrid upon the detection of an upstream fault, protect critical loads during islanding and manage critical and non-critical loads during islanded operation. This incursion into multi agent based control system development will act as a stepping stone to develop more intelligent distributed control systems in the future. The novel concept of multi layered control architectures will increase the impact MAS has in microgrid control. 55

64 6.2. Future work Multiple DG units and storage elements (i.e. battery storage, water pumping, hydrogen generation etc.) can be integrated into the microgrid to increase reliability, reduce emissions and to improve sustainability. As most NCRES tend to be of a highly fluctuating output, energy storage would smooth-out the variation. With the integration of multiple/various DG units, the user (utility or consumer) can select the source which offers the optimal energy solution. These improvements will require the development of DG agent s responsibilities. The following improvements can also be incorporated in future. i. Agent behaviours can be improved by utilizing fuzzy / neural control logic ii. The Effect of increasing number of agents on controller stability and service latency can be studied to evaluate long-term extensibility of MAS iii. With multiple DG units vying to service any given microgrid, a MAS based energy bidding system can be developed for energy auctioning from independent power producers (IPPs) iv. The requirements of communication network redundancy and aspects of how to improve data network security and preventing cyber-attacks and intrusions should also be addressed in future work. The developed multi agent system is to be implemented on the physical test-bed shown in Figure 6.1 comprising rotating machines and solar PV panels. Figure 6.1 Proposed microgrid test-bed 56