Neural Network Certification Informed Power System Transient Stability Preventive Control with Renewable Energy

Abstract

Existing machine learning-based surrogate modeling methods for transient stability constrained-optimal power flow (TSC-OPF) lack certifications in the presence of unseen disturbances or uncertainties. This may lead to divergence of TSC-OPF or insecure control strategies. This paper proposes a neural network certification-informed power system transient stability preventive control method considering the impacts of various uncertainty resources, such as errors from measurements, fluctuations in renewable energy sources (RESs) and loads, etc. A deep belief network (DBN) is trained to estimate the transient stability, replacing the time-consuming time-domain simulation-based calculations. Then, DBN is embedded into the iterations of the primal-dual interior-point method to solve TSC-OPF. To guarantee the robustness of the solutions, the neural network verifier $\alpha, \beta$-CROWN to deal with uncertainties from RESs and loads is proposed. The yielded certification results allow us to further adjust the transient stability safety margin under the iterated TSC-OPF solution process, balancing system security and economics. Numerical results on a modified western South Carolina 500-bus system demonstrate that the proposed method can efficiently and quickly obtain the safety-verified preventive control strategy through RES curtailment and generator dispatch with only a slight increase in cost.