Deep learning-driven evolutionary algorithm for power system voltage stability control

Abstract

The voltage stability problem has strong nonlinearity and sensitive to dimensionality, which makes it hard to balance accuracy and efficiency in its control. To enhance voltage stability margin (VSM) precisely and rapidly, a surrogate control algorithm with back propagation neural network (BPNN) embedded is carried out. Firstly, BPNN is adopted to parameterize a nonlinear voltage stability boundary mapped by operational variables. Further, benefited from the speed and precision ability of BPNN calculation, BPNN is embedded into preventive control model to replace the time-consuming calculation process of VSM. In doing so, an efficient and nonlinear simplified control model oriented to voltage stability can be constructed to enable efficiency for load margin enhancement. Besides, non-dominated sorting genetic algorithm (NSGA-II) is proposed to solve the VSM control model with the advantages of fast running speed and good convergence of solution set. At last, numerical studies on 118-bus systems testify that BPNN can achieve VSM evaluation with high accuracy and rapidly assist in NSGA-II improving the static voltage stability margin.