Deep-Q-Network-Based Intelligent Reschedule for Power System Operational Planning

Because there are too many controllable variables and multi-objective problems in the reschedule of power grid operational planning, the traditional reschedule method relying on artificial experience can not meet the demand. Considering the advantages of DRL in high-dimensional data perception and multi-objective optimization, this paper proposes an intelligent reschedule algorithm of grid operational planning based on DQN, which approximates the value function of the reschedule action through the Q network. It uses the state feature vector composed of grid data as the input of the Q network, and outputs the action value of the reschedule action. Then it selects actions to execute according to the $\varepsilon-$ greedy strategy. And calculate the power flow to obtain the state feature vector under new operation for giving the reward value. Finally, after training, the algorithm gives the optimal discrete action strategy to reach the target operational planning. Case studies on the IEEE39-bus system and the comparison result with OPF shows that this algorithm has good convergence in the multi-objective optimization of the operation and a fast calculation speed after offline training.