OptNet-Embedded Data-Driven Approach for Optimal Power Flow Proxy

Abstract

Solving AC-optimal power flow (AC-OPF) in real-time is crucial for further power system operation and security analysis. To this end, data-driven methods are employed to directly output the OPF solution. However, due to the prediction error, it is a challenge for data-driven methods to provide a feasible solution. To address this issue, different feasibility-enhanced methods are proposed. However, they are either computationally expensive or cannot provide feasible solutions. In this paper, we propose an OptNet-embedded data-driven approach for AC-OPF proxy to provide a feasible solution efficiently. This approach designs a three-stage neural network architecture to represent the OPF problem, where the first stage is used to lift the dimension of the input, the second stage is used to approximate the OPF problem using OptNet, and the third stage is used to decouple the high-dimensional solution to acquire the OPF solution. Finally, to expedite the solving process, a two-step pruning method is proposed to remove the unnecessary inequality constraints and values. Numerical experiments on the IEEE 4- and 14-bus test systems validate that the proposed approach can provide a “good enough” feasible solution.