Deep Reinforcement Learning based Operation and Maintenance Strategy for Renewable Energy Sources Projects in a Retail Electricity Market

With the trend of utilizing renewable energies to reduce the usage of fossil fuels and to fulfill the goal of carbon neutral in China by the year of 2060, nearly 20 projects with renewable energy sources (RES) have been built or expanded in the year of 2020 in the Guangdong Province, China. Also, the original wholesale electricity market in the Guangdong Province is based on the time-of-use policy, which has been transformed into a two-settlement pool-based electricity market, where RES projects can participate in through substations. In this paper, an innovative operation and maintenance strategy for major components in RES projects is proposed based on realworld data sets. In addition, a comprehensive satisfaction level evaluating the operation and maintenance strategy is developed based on 22 indices regarding the performance of the wind turbines/photovoltaic in the RES projects. Our objective is to minimize the operation and maintenance costs of RES projects in a two-settlement pool-based retail electricity market while maximizing the comprehensive operation and maintenance satisfaction levels based on the proposed indices. A novel deep reinforcement learning algorithm is developed to provide optimal operation and maintenance strategies for operators of RES projects. The uncertainties associate with the operation and maintenance strategy such as wind speed, solar irradiance, distributed wind turbine/solar panel conditions, maintenance crew conditions, and day-ahead and real-time electricity prices are handled through the training process by the deep Q-network (DQN). Extensive simulation results based on realworld data sets prove that RES projects can maximize the revenue by selling power to the retail electricity market with the proposed optimal operation and maintenance strategy, while maintaining the comprehensive satisfaction indices above a certain level.