Unit Commitment and Economic Dispatch via Graph Attention Neural Network–Based Parallel Distributed Coordination Algorithm

Abstract

The joint optimisation of unit commitment and economic dispatch (ED) is one of the key issues in smart grid scheduling and control. Integrating the discrete on/off statuses of units in the unit commitment problem with the continuous active power outputs in ED significantly increases the overall complexity of the combined optimisation problem. We propose an innovative distributed algorithm based on a graph attention network to address this challenge. The graph neural network is used to extract the inter-unit relational features and predict the future power dispatch schedule of each unit, while the parallel distributed coordination algorithm (PDCA), acting as the power dispatch algorithm, schedules and controls the output power of the units, including their start-up and shut-down states. Experimental results show that our algorithm performs well on both the IEEE 30-bus and IEEE 118-bus test systems, achieving a 1559 times speed boost compared to advanced solvers, and reaching economic optimality while satisfying all critical constraints to obtain an industrial acceptable solution.