Reinforcement Learning-Based Active Distribution Management for Reducing the Risk of Cascading Failure

In this paper, a controller for reducing the risk of cascading line outages in an active distribution network (ADN), has been developed. The controller is based on adaptive critic design (ACD), which receives inputs from a distribution management system regarding the states of the distribution network. These states are then evaluated by the controller for critical contingencies in the network. Accordingly, the controller takes coordinated control actions by giving recommendations for adjusting the optimal power flow (OPF) set points for voltage and active power of the distributed energy resources (DERs) in the network. These control actions avoid considerable deviation from the optimal set points for generation cost and line losses. The effectiveness of the developed ACD controller is demonstrated with a test case of IEEE 30 bus system. The technique for formulation of this ACD algorithm, so as to attain a faster but accurate convergence, has been elaborately described in this work.