Studies On Online Intelligent Decision-Making And Optimization Of Generator Start-up After Power System Major Blackouts

Along with the development of social economy,power demand continues to grow.Some renewable energy including wind generation are growing in burgeoning power systems on a large scale.The dependence of countries,societies and people on electricity is becoming higher and higher.Large-scale blackouts threaten the lives and property of people and even the security of country and society.Rapid power system self-healing after blackouts is able to alleviate the negative effects.The primary task of power system self-healing is to accomplish generator start-up and establish a backbone network.Especially,generator start-up is the most important task in the initial stage of power system restoration.Hence,Comprehensive studies on generator start-up considering backbone network reconfiguration are very important and valuable for secure,organized and rapid power system self-healing.Reasonable and effective generator start-up schemes can involve different kinds of factors about power system restoration,which can guide dispatchers to accomplish generator start-up and backbone network reconfiguration.Rapidly developing power information systems can be efficiently used to make decisions step by step for generator start-up intelligently,which can handle the conditions that the assumed restoration process of schemes developed offline is different from practical restoration process.For the high wind power penetration power systems with higher risk of power failure,online guidance for restoration of wind farms and conventional generators can make full advantages of the wind farms with high wind speed to promote power system restoration.On the basis of current researches,online intelligent decision-making and optimization for generator start-up are studied in this dissertation.Advanced artificial intelligence techniques including preference multiobjective optimization,evolutionary computation,deep learning,reinforcement learning and Monte Carlo tree search(MCTS)are applied.A comprehensive intelligent decision-making and optimization system for generator start-up including restoration scheme development,online restoration decision-making and generator start-up considering wind farms is established.Main contributions and innovations of the dissertation are described as follows:(1)Integrating different factors about generator start-up,a generator start-up method is proposed based on preference multiobjective optimization to develop generator start-up schemes.On the one hand,considering the influences of generators,networks and loads,three indexes,including total generation capability,average importance of transmission lines and percentage of important load in skeleton network,are proposed as evaluation indicators of generator start-up.Considering the evaluation indicators as objectives and the corresponding constraints,a preference multiobjective optimization model is established based on the preferences for different objectives.On the other hand,a preference-based discrete nondominated sorting genetic algorithm Ⅱ(PD-NSGA-Ⅱ)is designed considering the preference and high discreteness of the suggested model.A preference-based dominance relation and a relaxed Pareto dominance relation are proposed and used in outspread population and normal population during evolution to improve the solution efficiency for generator start-up problem.Following conclusions are drawn from numerical studies.The generator start-up schemes obtained with the proposed preference multiobjective optimization model can emphasize the generator start-up without compromising the influence of skeleton network on subsequent restoration and economy.Compared with similar algorithms,PD-NSGA-Ⅱ is more efficient for solving the generator start-up optimization problem,which can obtain solutions with required quantity and high quality.Furthermore,the number of solutions can be set by DMs in advance,which is convenient for the final decision-making.The approach is practical for generator start-up scheme development.(2)For the uncertainty of initial state of power systems after blackouts and generator start-up processes,a dynamic decision-making method is proposed based on MCTS and sparse autoencoder(SAE)for online generator start-up.A framework for online generator start-up including offline preparatory work and online decision-making is created.Generators can be restored and skeleton network can be established by determining the next restored lines step by step.A generator start-up efficiency indicator is proposed as the evaluation indicator to guide decision-making.SAE is utilized to learn the data about generator start-up offline for a value network.The value network is used in MCTS,which can rapidly estimate the generator start-up efficiency indicator in a certain power system situation.MCTS is introduced for generator start-up decision-making.Some techniques,including modified upper confidence bound apply to tree(UCT)algorithm,move prunning and simulation based on value network,is proposed to improve the search efficiency of MCTS for the next restored lines.Further,parallel computation is applied to guarantee the online realization of generator start-up decision making.Following conclusions are drawn from numerical studies.SAE is a highly desirable method for the offline learning of generator start-up data.The modified UCT algorithm,move pruning technique and value network established offline can improve the search efficiency of MCTS,which guarantees online realization of generator start-up decision making reliably.The proposed algorithm can automatically handle the uncertainties during generator start-up,which include the unexpected line restoration time and line restoration failure.The schemes obtained online have better robustness than the schemes obtained offline.(3)A generator start-up method based on real-time power system situation and wind power forecast information is proposed to determine the next restored lines during gen-erator start-up,which can assist dispatchers to accomplish the restoration of wind farms and conventional generators.An evaluation indicator considering total generation capability and the hot-start of thermal generators is proposed to evaluate the performance of generator start-up considering wind farms.Due to the high diversity of wind power scenarios,a self-learning strategy based on reinforcement learning is proposed for gen-erator start-up data learning to generate a policy network.The policy network can rapidly estimate the restoration probability of alternative lines in a certain situation during generator start-up.The policy network is applied to MCTS to improve its search efficiency.In order to guarantee the active power balance during restoration,model predictive control is adopted for wind power control.A rolling optimization model is proposed to calculate the power generation of every wind farm based on real-time situation and wind power forecast information.Following conclusions are drawn from numerical studies.The self-learning strategy can obtain effective policy network within adjustable time,which improves the efficiency of generator start-up decision-making.The proposed method can make reasonable decisions in different wind power scenarios and spontaneously respond to sudden wind power ramping events.The wind power control strategy can guarantee power balance and provide control redundancy for possible wind power ramping by sacrificing some power generation.