Abstract：With the increasing proportion of clean distributed energy in the power grid, the stable operation of the power grid is challenged by the power prediction error, a control method is required for energy storage system to make a rapid charge and discharge decision to achieve the best effect of peak shaving and load shifting in the real-time scheduling cycle. Considering the constraints such as unit ramp rate and capacity limit of the energy storage system, an improved Q-learning algorithm is proposed, using the penalty term lead the algorithm to obtain optimal decisions by training in phases using the offline data, which has a fast convergence performance and high accuracy; In the real-time scheduling cycle, when the prediction of load power is not accurate, energy storage system don’t need the iterative operation, it just needs to correct the state values and judge its state, then obtain its optimal action value by using the greedy strategy based on the trained Q value table. The simulation results show that the method proposed in this paper has a faster convergence performance compared with the traditional Q learning algorithm, and the trained Q value table can be used for energy storage system to make optimal decisions in the real-time scheduling cycle.
汪波, 郑文迪. 基于改进Q学习算法的储能系统实时优化决策研究[J]. 电气技术, 2018, 19(2): 54-60.
Wang Bo, Zheng Wendi. Research on real-time optimization decision of energy storage system based on improved Q-learning algorithm. Electrical Engineering, 2018, 19(2): 54-60.
2018, Vol. 19 
