Joint Optimization Control Of Energy Storage System Management And Demand Response

Most of the power generation of distributed new energy sources is intermittent and fluctuant. Connecting the distributed new energy generation systems to the power grid directly will cause fluctuations in the frequency and voltage of the grid, which will even lead to blackouts. The introduction of energy storage system can effectively smooth the intermittent power fluctuations, increase back-up capacity of the grid and regulate the difference between peak load and valley load. The reasonable energy scheduling for energy storage system has become one of the key technologies in smart grid. Under real-time pricing mechanism, optimal scheduling of the energy storage system and demand response can bring considerable economic benefits to users.In this thesis, we consider the system consisting of the external grid, distributed power sources and multiple intelligent appliances, and establish the model of joint optimization control of energy storage system management and demand response. According to the random nature of solar photovoltaic power generation system, load’s demand for electricity and price information, we control the charging or discharging action of the energy storage system and the accessing time of the load, and define the optimization objective function for the system, which includes both electricity economic and user satisfaction, model the joint optimization problem as an infinite-horizon Markov decision process, and then Q-learning based algorithm is presented. Simulation results show that with our proposed joint optimization control mechanism, system obtains a higher long-term return than using single battery operating control or demand response control.Considering the time characteristics of both the power generation of distributed energy sources and the price of electricity, we further establish a finite horizon optimization model for the system, by using time as one of the state vectors. And simulated annealing based Q-learning is presented. Simulation results show that the joint optimization controlling improves reward of the users.