Dynamic Economic Dispatch Research Of Power System Considering Uncertainties Of Renewable Energy Forecast Error

In 2020,China made a commitment of peak carbon dioxide emissions and achieve carbon neutrality,which indicating that the development of renewable energy will be bolstered.However,the inherent volatility,intermittence and randomness of renewable energy make the forecast the output difficult.The uncertainties of forecast error are the challenges the dispatching department face.How to arrange the scheduling plan to adapt to the uncertainties of renewable energy forecast error is an urgent problem to be solved by the dispatching department.In recent years,Machine Learning technologies,represented by Deep Learning and Reinforcement Learning,have emerged continuously,providing new methods and new ideas for solving practical engineering problems of power systems.In this paper,Machine Learning technology is applied to the dynamic economic dispatch scenario of power system,and the specific work arrangement is as follows:Based on the Deep Reinforcement learning algorithm,this paper constructs a dynamic economic dispatch model to deal with the uncertain factors of power system.Firstly,the dispatching decision center is regarded as an agent,the power system as an environment,and the dynamic economic dispatch model is transformed into a multi-stage decision model.In view of the continuous state and action space in the model,this paper combines the Proximal Policy Optimization algorithm and the Deep Neural Network to solve the decision model.Finally,the feasibility and applicability of the proposed method are verified in the case study.The case study shows that the Deep Reinforcement Learning model can adapt to the uncertainties of power system based on its exploration-interactive-feedback-learning mechanism,avoid the construction of probability model for complex random variables,and this method is not limited by the system scale,shortens the online decision time.The complex reward function in Reinforcement Learning model increases the difficulty of modeling and the lack of prior knowledge at the initial stage of training leads to a long convergence time of the algorithm.To solve these problems,a dynamic economic scheduling model based on Generative Adversarial Imitation Learning is proposed in this paper.Firstly,this paper introduces Imitation Learning,and takes the optimal scheduling strategy generated by perfect dispatch as the expert strategy to guide the agent to explore,so as to improve the convergence of the algorithm.Secondly,this paper uses the game characteristics of Generative Adversarial Networks to construct Generative Model to generation strategy,and Discriminative Model to recognize generation strategy and perfect dispatch strategy.Generative Adversarial Networks can avoid the artificial definition of reward function.Finally,a case study is used to verify the superiority of the proposed model over the single Reinforcement Learning model.The model can realize the end-to-end dispatch strategy learning,avoid the meaningless action exploration of the agent,shorten the offline training time,and enhance the ability of the algorithm to deal with high-dimensional complex problems.In this paper,dynamic economic dispatch of power system is innovatively regarded as a decision model,and data-driven Machine Learning algorithm is used to fully mine the uncertain information contained in the data,which breaks through the bottleneck of model-based algorithm and achieves good calculation results.In the future,with the increasing application scenarios of Artificial Intelligence technology,the application of Machine Learning algorithm to dynamic economic dispatch will make a greater breakthrough.