Distributed multi-agent system based algorithms for energy management of power systems

Power system is one of the most complex engineering systems in the world. The traditional centralized optimization and control scheme require complicated communication networks to acquire global operating conditions and process the large amount data with a centralized controller. Due to delays with communication and data processing, the centralized solutions might be unable to provide timely response. Also, in a deregulated power market, it is desirable that system participants (generators or loads) have more autonomy in their energy decision. In addition, it is well known that centralized solutions are usually inflexible and susceptible to singlepoint-communication failures. Since distributed optimization and control schemes are more flexible, reliable, and have fewer requirements on communication system, they are promising alternative for future power system.;To address the needs of power systems and problems with existing solutions, this dissertation proposes several fully distributed multi-agent-system (MAS) based algorithms originated from the most recent developments of control and optimization theory, which have been successfully applied to optimal active power dispatch (OAPD), integrated energy management based on social welfare optimization (SWO), day-ahead optimal generation-demand scheduling (OGDS).;A fully distributed solution is proposed for OAPD problem. To balance computational efficiency and feasibility of the solution, DC power flow was integrated to check line-flow-constraint violations. In this way, optimality and feasibility of the solution can both be guaranteed and improved compared to the authors' previous distributed solution. For SWO, a fully distributed solution is proposed that solves economic dispatch (ED) and demand response (DR) problems in an integrated way. Compared with sequentially implementing the two operations, the integrated solution can both maximize benefits of customers and minimize generation cost of generators efficiently and simultaneously. By adjusting generators and dispatchable loads, line-flow constraints can be easier satisfied. For OGDS, a fully distributed solution of day-ahead optimal generation and demand scheduling problem is proposed to achieve optimal energy schedule with demand-side management. According to the proposed solution, 2 dispatchable-demand properties are modeled: non-committed and committed demand. The distributed solution can both preserve customers' autonomy in demand profile scheduling and facilitate their participation in retail power market.;For the above energy management problems, 2-level algorithms are designed and implemented by using MAS: gradient algorithm and consensus algorithm. According to the proposed solution, each bus is assigned with an agent (local controller). An agent has computing and communication capability. Local decision variables (generation or demand) are calculated and updated in bus agents (BA) by using gradient algorithm. All necessary global information required by gradient algorithm is calculated by using consensus algorithm, which is designed to achieve global situation awareness in a fully distributed way.;The effectiveness of all proposed algorithms has been demonstrated through simulations. The corresponding work has produced three papers in IEEE transactions on Information Informatics, Smart Grids and IET Generation, Transmission and Distribution.