An Emergency Load Shedding Optimization Algorithm Based on Differential Evolution Driven by Deep Transfer Learning

Abstract

Emergency load shedding (ELS) optimization by the meta-heuristic algorithm is confronted with a heavy computing burden as time-domain simulation (TDS) is needed to evaluate individuals. In this paper, an ELS optimization algorithm based on differential evolution driven by deep transfer learning is proposed to decrease TDS times. First, an optimization model of ELS is established in a single scenario based on differential evolution. Then, a Deep Belief Network (DBN) model is deployed to evaluate load shedding schemes to speed up the optimization. Moreover, when scenarios are changed, transfer learning is utilized to update the DBN model, avoiding massive samples regeneration by TDS. Finally, the optimal load shedding scheme is rolling updated following scenarios change. A simplified provincial power grid verifies the effectiveness of the proposed algorithm.