Transfer Multi-Searcher Q-Learning And Its Application In Power System

With the development of the power system,there are more and more nodes in the power system,and the optimization problem is used more and more frequently.The commonly used solutions are classical mathematical methods and artificial intelligence algorithms.Traditional numerical calculation can solve the optimization problems of some power systems,but it is often difficult to model.Sometimes it takes a long time to optimize a model,and sometimes the problem can not be solved.Intelligent algorithm has low requirements for specific mathematical model and is easier to apply,but for the optimization calculation of high-dimensional,complex and nonlinear problems,the optimization results and optimization time are often difficult to be directly applied to the operation of the actual power system.At present,the penetration of renewable energy power generation systems such as wind and solar energy in the grid is gradually increasing.However,the load uncertainty and the subsequent power production fluctuations will also bring new challenges to the operation and distribution of the power system.Therefore,this paper establishes a collaborative optimal scheduling model including wind turbine and photovoltaic generator.With the concept of low-carbon power put forward,grid companies also need to bear part of the responsibility of carbon emissions in the process of power transmission,and pay part of the economic cost here.The carbon footprint of the generation side is transferred to the grid side and the user side by using the carbon emission analysis method,and the generation side,the grid side and the user side are allocated the carbon emission responsibility in the form of carbon flow,so as to develop a more effective energy conservation and emission reduction strategy.Therefore,in order to realize the low-carbon,energy-saving and economic operation of power system,the carbon energy compound flow is put into the objective function of reactive power optimization.In order to solve the above two models quickly,this paper designs a multi searcher optimization algorithm,and further combines reinforcement learning,deep learning and migration learning,proposes a Q-learning algorithm of migration multi searcher.In the process of initialization,the multi searcher optimization algorithm adopts chaos theory,and contains two different types of searchers: Global searcher and local searcher,which can greatly improve the search efficiency of the problem.The Q-learning algorithm of migration multi searcher uses binary technology to initialize continuous variables and discrete variables,effectively reduces the dimension of action state space,and avoids "dimension disaster".In the process of information matrix transfer,the long-term and short-term memory neural network technology is used to pre learn the information matrix,make full use of historical optimization information to optimize new tasks and reduce the optimization time.Finally,the effectiveness of the proposed algorithm is proved by standard examples and practical engineering examples,which has certain theoretical value and engineering significance.