Research On Day-ahead Trading Strategy Of Wind Power Based On Deep Reinforcement Learning

Wind energy is one of the fastest growing renewable energy sources.With the rapid growth of wind power,countries all over the world are rapidly integrating wind power into the electricity market.However,due to the strong uncertainty of wind energy,wind power producers are often punished due to improper bidding strategies when they participate in short-term power market wind power bidding transactions.Therefore,how to optimize the trading strategy of wind power producers in the short-term power market wind power day-a-day trading will be the main research content of this article.The research work is based on the National Key Research and Development Program-China and Egypt Joint Research Project(2018YFE0127600)-Research on key technologies for planning and stability control of weakly interconnected hybrid renewable energy systems.This article is based on deep reinforcement learning to study the day-ahead trading strategies of wind power in the short-term power market.The main reason is that the transaction entities in the power market are complex and involve multi-party competition and cooperation.Deep reinforcement learning can recognize complex environments.And learning can well solve the decision-making problem of multi-party interactive games.First,this article analyzes the basic theories and models of reinforcement learning,including the basic framework and constituent elements of its learning,and summarizes two types of reinforcement learning algorithms and their processes based on value functions and stochastic gradient strategies: one type of simulation comparison used in this article,One type is the asynchronous actor-critic algorithm used in this article,which is the basis of the A3 C algorithm.Then this article takes the Nordic electricity market as the background,analyzes its structural characteristics and the functions of each market member,and based on the relationship between wind energy and electricity prices,establishes two bidding trading models for wind power generators in the short-term electricity market: one for only considering Electric energy trading in the traditional electricity market,the other also considers energy trading in the energy storage market,and uses the A3 C algorithm for two wind power bidding trading models,constructing two types of day-a-day wind power trading in the electricity market based on deep reinforcement learning.Strategy.Finally,a wind power plant in western Denmark was selected to simulate two bidding trading strategies.The simulation results were analyzed and compared with the two traditional reinforcement learning optimization strategies and the original strategies adopted by wind power producers.The results show that the A3C-based wind power bidding trading strategy based on deep reinforcement learning can optimize the income of wind power generators in both power market trading scenarios.The strategy used in the traditional power market transaction is higher than the daily income of the two traditional reinforcement learning optimizations.Strategies 2.2% and 3.3%;in the electricity market transaction scenario that considers the energy storage market,the proposed strategy reduces the cost by29.5% compared to the original strategy adopted by wind power producers.In summary,the wind power bidding strategy based on deep reinforcement learning A3 C proposed in this paper can not only improve the profits of wind power producers,but also cope with the uncertainty of wind power forecasting and dynamic changes in the power market.