A KAN-LSTM combined load forecasting and risk assessment method in distribution network with energy storage

The operation of distribution systems with coal mines faces several technical challenges, including nonlinear load characteristics, strong couplings, and multiple types of disturbances due to the large-scale deployment of power electronics-based auxiliary equipment. Ensuring secure operation relies on accurate load forecasting, timely voltage risks assessment, and the support of energy storage systems. An integrated technical framework was proposed in this paper. A Kolmogorov-Arnold Network-Long Short-Term Memory (KAN-LSTM) hybrid model was developed for short-term load forecasting, which combines the strong nonlinear fitting ability and the long-term time-series dependency capture capability. Comparative results in case studies on the IEEE 33-bus system with practical data demonstrated that the KAN-LSTM model performs better than traditional BP networks and single LSTM model. For voltage risk assessment, a piecewise nonlinear risk evaluation method is designed to consider the voltage-gas risk mechanism in coal mines, which effectively evaluates the operation risks quantitatively with voltage deviations. Moreover, with the support of energy storage systems, which is optimized with a Mixed-Integer Second-Order Cone Programming (MISOCP) model, the proposed framework can identify high-risk nodes and maintain voltage deviations within secured regions. This study provides a technical solution to improve the security level of the distribution systems with coal mines. • A KAN-LSTM hybrid model captures coal mine load nonlinearity and mutations, outperforming BP and LSTM. • A piecewise voltage–gas risk method quantifies safety risk in real time for coal mine grids. • MISOCP optimizes energy storage to reduce voltage violations via active–reactive coordination. • The integrated framework enables closed-loop prediction–assessment–mitigation for safe operation. • Case studies on IEEE 33-bus validate forecasting accuracy, risk identification, and voltage recovery.