Construction method of temporal deep learning model for aging assessment of power hard composite insulators

Abstract Hard composite insulators are essential for reliable power transmission, but aging caused by environmental and electrical stresses increases leakage current and failure risks. Existing assessment methods face challenges in prediction accuracy, adaptability, and temporal feature analysis. This research proposes a novel ModABi-LSTMO framework that integrates DWT-based feature extraction, BiLSTM temporal learning, and ModAO optimization for accurate and intelligent aging assessment of power hard composite insulators. The Power Hard Composite Insulator Aging Dataset (Kaggle, 9775 records) containing simulated operational and environmental sensor observations designed to emulate aging behaviour of high-voltage composite insulators under representative operating conditions. In the preprocessing step, missing sensor data was imputed and feature values standardized using Min–Max normalization to improve training stability. Feature extraction utilized Discrete Wavelet Transform (DWT) to obtain time-frequency features from leakage current signals, facilitating the detection of hidden aging patterns in the insulation system. A DL system named Modified Archimedes Bi-directional Long Short-Term Memory Optimizer (ModABi-LSTMO) is created based on extracted temporal features. The model combines a Bi-directional LSTM (BiLSTM) network to efficiently represent forward and backward temporal relationships in the degradation process, and the modified Archimedes optimization (ModAO) strategy is employed to optimize the network parameters and enhance convergence behavior. Experimental analysis in Python demonstrates that the ModABi-LSTMO model achieves a 98.9% accuracy in predicting aging states, outperforming traditional methods. This model enhances the reliability and efficiency of power transmission systems through smart monitoring and predictive maintenance for power hard composite insulators.