Abstract
Due to the rapid development of technology and the rise of industry and economics, the damage to the society caused by the failure of power system becomes larger and larger. How to recover the power system from black-out in the shortest time with minimum cost has become an important issue in recent years. In this thesis, we will treat components of the power grid as intelligent agents, for instance, feeders, switches and power demand areas, and then construct a distributed power restoration system with intelligent agents based on concept of committee-based negotiation. Once the power system of an area fails, the system we proposed will automatically detect and isolate this area, and begin the solution finding processes. Base on the committee-based negotiation mechanism, the switch agents will share their information to the committee members of the committee which they belong to, negotiate with each other, and find out a candidate local solution depending on two major objective functions which are minimizing the number of switch changes and maximizing the amount of power demand to be restored. Finally, we can escape from being trapped in a local optimal solution and find the globally optimal solution with the hierarchical message passing and by the exhaustive constraint satisfaction decision making mechanism we proposed. Besides, we further take the problem of voltage drop into account and implement the mechanism of load-transfer to let the solution be much stable in practice. We also design a fault-tolerance mechanism on our restoration system to assure that our restoration system can work even when limited number of agents failed. Although the solution generated maybe degraded due to the missing of some information, but the system will not directly shut down and let the emergent problem of power system be unsolved. Otherwise, it must cause larger damage to the society. We will take different power grids and benchmark problems from power restoration research literatures as the test samples of our experiments and compare our multi-agent approach to the traditional methods. Besides, we also evaluate the fault tolerance ability of our restoration system to see how it performs under different degrees of failure in power restoration agents.