Smart Estimation of Low-Voltage Capacitor Life for Predictive Maintenance

Low-voltage shunt capacitors, as effective components for reactive power compensation, have been widely implemented in power systems. However, when their capacitance degrades beyond a critical threshold, leading to failure, this can significantly compromise system operational safety. This study investigates the remaining service life prediction of low-voltage shunt capacitors and develops a data-driven predictive model incorporating various environmental factors. Building upon conventional long short-term memory (LSTM) neural network prediction, we propose an enhanced bidirectional LSTM approach integrated with convolutional neural networks and the whale optimization algorithm. This hybrid WOA-CNN-BiLSTM model demonstrates improved prediction accuracy, computational efficiency, and robustness. Comparative simulation analyses of root mean square error (RMSE) and mean absolute error (MAE) before and after optimization reveal that the proposed model achieves reductions of 0.0117 in RMSE and 0.0063 in MAE compared to traditional LSTM. The results indicate that the WOA-CNN-BiLSTM model offers superior accuracy and stability, enabling effective mitigation of power quality deterioration caused by abnormal operations in reactive power compensation equipment. This advancement contributes to enhanced operational efficiency and extended service life for power system components.