Q-learning algorithm based method for enhancing resiliency of integrated energy system

The stochastic dynamic optimization problem of integrated energy system is modeled as a Markov decision process, and Q-learning algorithm is introduced to solve this complex problem. In order to overcome the disadvantages of Q-learning algorithm, two improvements are made to the typical Q-learning:the Q table initialization method is improved and the upper bound convergence algorithm is adopted for the action selection. Simulative results show that Q-learning algorithm ensures better convergence while solving the problem, and the improved initialization method and the upper bound convergence algorithm can significantly improve the computational efficiency and make the results converge to a better solution. Moreover, compared with the conventional mixed integer linear programming model, Q-learning algorithm achieves better optimization results. © 2020, Electric Power Automation Equipment Press. All right reserved.