Research On Auxiliary Decision Strategy Of Power Dispatching Based On Deep Reinforcement Learning Algorithm

Power dispatch is an important means to ensure the safe operation of the electric power system.In recent years,the proportion of renewable energy in the electric power system has increased significantly.The randomness,intermittency,and volatility of renewable energy have added difficulty to power dispatch,posing new challenges for power system dispatch operations.Therefore,it is particularly important to formulate reasonable auxiliary power dispatch decision-making strategies based on renewable energy output forecasts and load forecasts.Deep reinforcement learning,by constantly interacting with the environment and accumulating experience,uses feedback information from the environment to continuously optimize and improve strategies.It has the advantages of experience-based learning,the ability to deal with complex environments,the ability to deal with long-term rewards,and the ability to learn adaptively.Applying deep reinforcement learning to the dispatch of a complex power grid can effectively optimize the accuracy of power dispatch.Therefore,this thesis proposes a power dispatch auxiliary decision-making strategy based on a deep reinforcement learning algorithm,mainly focusing on the following aspects:The auxiliary power dispatch decision strategy problem is normalized based on the Markov decision process into a sequence decision problem,and a high proportion of dispatch models for renewable energy power systems are established,clarifying dispatch goals and operational constraints.On this basis,the dispatch decision center is regarded as an intelligent agent and the power system is the environment.The three elements of state space,action space,and reward function of deep reinforcement learning are designed.The near-policy optimization algorithm of deep reinforcement learning is used to optimize power system dispatch,and experience data is continuously obtained through the interaction between the intelligent agent and the environment,and the dispatch auxiliary decision-making strategy is continuously optimized.The case study results show that in a power system with a high proportion of renewable energy units connected to the grid,the dispatch strategy given by the intelligent agent can meet the load demand under different load change trends,and can maximize the absorption of renewable energy in the dispatch cycle with different renewable energy output rates,ensuring the safe and economic operation of the grid.Provide effective and reliable auxiliary decision-making strategies for grid dispatch.In order to improve the exploration-exploitation dilemma of deep reinforcement learning algorithms in the initial training period,an improved power dispatch auxiliary decision-making strategy based on deep reinforcement learning algorithms is proposed.Through a supervised pre-training method,the initial deep reinforcement learning policy network clones the expert policy to improve the performance and training speed of the algorithm,and solves the problem of low sample utilization rate and long algorithm convergence time of deep reinforcement learning.Simulation results show that this method can not only improve the training speed of the deep reinforcement learning algorithm,but also improve the sample utilization rate and performance of the algorithm,which can effectively improve the absorption rate of renewable energy in power transmission.The power dispatch optimization problem can be transformed into a sequential decision-making problem.By using behavior cloning technology to improve deepreinforcement learning algorithms and applying the proximity policy optimization algorithm of deep reinforcement learning to further optimize power system dispatch,prediction accuracy can be improved.This can effectively address the challenges posed by the difficulty and randomness of forecasting renewable energy output on the dispatch operation of the power system.By using the improved deep reinforcement learning algorithm and supervised pre-training,the sample utilization rate and training speed of the deep reinforcement learning algorithm are optimized,and the algorithm performance is effectively improved.It can providean effective auxiliary decision-making strategy for power system dispatch in high proportion renewable energy power systems.Simulation results have verified the effectiveness of the method.