Research On An Adaptive Control Method For Multi-Energy Structure Power Grid Based On Reinforcement Learning

Operational control of a power grid is a difficult task as it requires complex and reliable decision-making.The current control mode is mainly based on manual control mode,which requires regulators to have a deep understanding of the operation mode of the power grid and the operation status of the power grid equipment,and to make plans for possible events that affect the operation of the power grid.However,when the penetration rate of renewable energy in the grid reaches a certain level,the intermittency and volatility of renewable energy generation will lead to changes in the power flow pattern in the grid,and its distributed characteristics will also change the original passive radial grid.structure,which complicates grid energy scheduling,and the manual control model that relies on fully predictable models will be unsustainable.To this end,this thesis proposes an adaptive control method for multi-energy structure grid based on reinforcement learning,which can adopt different strategies for different proportions of renewable energy in the multi-energy structure grid,in order to maximize the consumption caused by renewable energy replacement.fluctuations in grid operation.The main work of this thesis is as follows:(1)Design a control framework based on hazard warning.The framework includes two parts: the danger early warning mechanism and the auxiliary exploration mode.The danger early warning mechanism can selectively make the reinforcement learning algorithm intervene in the control according to the current operating conditions of the power grid,effectively improving the fault tolerance of the automatic control.The auxiliary exploration mode provides an optimized topological action set for the reinforcement learning algorithm on the basis of exploratory actions,and introduces an expert system to assist the exploration process of the reinforcement learning algorithm,so as to improve the learning efficiency of the reinforcement learning algorithm.(2)An intelligent grid control algorithm based on reinforcement learning is proposed.The key links such as the control action,reward function and update mechanism required by the control method are defined in detail,and the pre-state of the power grid environment is defined,which eliminates the changes of the power grid environment caused by random viriation factors,and then constructs a complete Markov decision process.This algorithm can autonomously try grid control actions in the training process and continuously improve its own control strategy.The experimental results in the power grid simulation environment with a single energy structure show that after the algorithm is integrated into the control framework based on danger warning,the power outage accidents during the operation of the power grid are reduced by 1 times.(3)An adaptive multi-energy structure grid intelligent control algorithm is proposed.In order to solve the problem of algorithm "decision forgetting" caused by the change of grid energy structure,this thesis proposes an adaptive grid intelligent control algorithm based on the grid intelligent control algorithm.By introducing an adaptive mechanism,the algorithm enables the agent to make corresponding decisions based on the similarity between the current environment and general environmental characteristics in the power grid control process of the multi-energy structure,effectively improving the adaptability of the agent in the control process.The experimental results in the multi-energy structure power grid simulation environment show that,compared with the general reinforcement learning algorithm,the adaptive multi-energy structure power grid intelligent control algorithm shows good adaptability when the power grid energy structure changes,and its comprehensive control success rate is improved.about 2.2 times.