Allocating defense resources for spatial cyber-physical power systems based on deep reinforcement learning

Allocating defense resources to specific lines can enhance the resilience of power systems against external damages. Considering the impact of information systems, a defense resource allocation model for cyber-physical power systems (CPPS) is developed with the length of power lines as the defense cost. It is assumed that defense resources can reduce the probability of successful attacks. For this nonlinear programming (NLP) problem, an optimization-seeking method based on the deep Q-network (DQN) algorithm is proposed. The model and algorithm are evaluated based on the IEEE-39 bus system. The results show that for small action sets, the method is in general agreement with the results obtained by the optimization solver BONMIN. In addition, the allocation strategies with different scales of resources and action sets are analyzed. These studies can provide ideas for the application of deep reinforcement learning (DRL) in resource allocation for power systems.