Solving Chance-Constrained AC-OPF Problem by Neural Network with Bisection-based Projection

Power grid security faces multiple threats, including cyber attacks, renewable generation uncertainty, and load variations. These uncertainties challenge grid operators to maintain both security and efficiency. The chance-constrained AC optimal power flow (CC-ACOPF) problem offers a valuable approach for maintaining grid reliability while optimizing operational costs under complex uncertainties. Despite its importance for reliable grid operation, existing solution methods either compromise model accuracy or face computational barriers that limit their practical implementation in security-critical environments where rapid response is essential. We present a neural network (NN) based approach to efficiently solve CC-ACOPF through two key phases: (i) an approximation phase leveraging NN techniques to obtain initial deterministic AC-OPF solutions; and (ii) a projection phase employs our recently developed bisection-based algorithm to recover solutions satisfying chance constraints with a pre-specified confidence level, leveraging random sampling to evaluate solution feasibility. We establish solution feasibility guarantees and optimality bounds of our approach. Extensive simulations on IEEE 30-/118-/200-bus test systems demonstrate that our method generates feasible CC-ACOPF solutions under diverse settings with minor optimality loss, and achieves a two-order-of-magnitude speedup compared to state-of-the-art alternatives. © 2025 held by the owner/author(s). Publication rights licensed to ACM.