Learning Optimization Of The Dispatch For Regional Active Distribution System Considering The Peak Load Regulation Demand Of Power Grid

Nowadays,with the rapid development of science and technology,the level of industrial automation constantly improves,intelligent technology and optimized dispatching mode are becoming increasingly maturing.Photovoltaic(PV),wind power and other kinds of renewable energy have been applied largely,and users' electricity consumption patterns have become more intelligent and diversified.The active distribution system(ADS)which can improve the consumption capacity of renewable energy and realize energy saving and emission reduction came into being.However,the random change of new energy output and the uncertainty of users' electricity consumption lead to the increasingly severe peak regulation form of power grid,and affect the security and power supply reliability of the power system.The peak load regulation task of power grid needs to be reasonably allocated by dispatching institution,and the distributed peak load regulation resources such as new energy,energy storage and flexible load in regional ADS are coordinated and optimized in order to improve the capacity of new energy consumption,meet the demand of peak load regulation,maintain the economy of ADS,enhance the reliability and controllability of power system.The ADS combined with PV,vanadium redox battery(VRB)energy storage device and multiple types of flexible load in industrial parks is focused in this dissertation.Considering the stochasticity of source and load,the dynamic dispatch optimization problem that response to random peaking demand instructions real-time is researched.First,the random dynamic variations of photovoltaic,multiple loads demand and peak load regulation demand are described as continuous Markov processes,and the VRB energy storage system is modeled based on its charge-discharge characteristics.Then,decision epoch,outputs level of photovoltaic,multiple load demands level,peak operation demands level and SOC level of VRB are defined as states of the system,the adjustment level of VRB and multiple types of flexible load are set as the actions.Based on practical operation requirements including the power balance constraint,the dynamic optimal dispatch problem for the system was described as a stochastic dynamic programming model,which aims to meet the random peak load regulation demand of power grid and realize the safe and economic operation of the system.Finally,the optimal policy was obtained by using Q learning optimization method,and the comparison and analysis on strategies obtained by traditional Q learning,double Q learning,simulated annealing Q learning and simulated annealing double Q learning was given.Simulation results show that the peak load regulation demand of power grid is basically satisfied and the operational efficiency is significantly enhanced by the optimal policy,which verifies the effectiveness of the optimal method.In addition,based on the research of ADS of industrial park,the dynamic dispatching problem of multi-region ADS is studied considering the peak load regulation demand of power grid.Firstly,the dynamic dispatching problem of ADS was described as a twolayer scheduling optimization model by adopting the layered control mode.The upper level formulates the corresponding peak load adjustment task allocation plan for each region,while the lower level formulates various flexible load adjustment plans and charging and discharging plans for energy storage devices in each regional ADS.Then,the double-layer Q learning algorithm is adopted to solve the problem,the simulation results and analysis is shown in this thesis,which indicates the effectiveness of proposed strategy.