An online prediction method for static voltage stability based on lightgbm-shap analysis

Rapidly solving the load stability limit power of a system is of great significance for the online assessment of static voltage stability in power systems. According to relevant regulations, under normal operation or maintenance conditions, the static voltage stability margin Kp of the system should not be less than 8%. To monitor the system’s load stability limit power when Kp is less than 8%, this paper proposes a LightGBMSHAP analysis and prediction method capable of offline training and online prediction of the system’s load stability limit power.This method uses offline samples based on the grid’s operational state features and employs the LightGBM algorithm for online prediction of the system’s load stability limit power. SHAP theory is then used to interpret the LightGBM prediction results, analyzing the impact of each operational state feature on the prediction of static voltage stability limit power to identify the critical factors.Testing results on the New England 39-bus system demonstrate that, for sample sets of 1,000 and 10,000, the model’s average relative prediction errors are 1.19% and 1.69%, respectively, with computation times of 0.179s and 4.42s. This method exhibits superior prediction accuracy, computational speed, robustness, and generalization ability. Moreover, the impact factor analysis based on SHAP theory provides a foundation and support for the prediction of system load stability limit power.