Life Prediction of Railway Relay Based on an Attention Mechanism TCN-BiGRU Model

Abstract To address the problem of accurate and efficient prediction of railway relay lifespan, this study proposes an electrical lifespan prediction method for railway relays based on the integration of attention mechanisms, Temporal Convolutional Network and Bidirectional Gated Recurrent Unit. Considering the potential variability in the importance of temporal features, we introduce an attention mechanism to reallocate feature weights dynamically. The TCN module is employed to extract local temporal features from relay operation data, while the BiGRU component is utilized to capture long-term dependencies and obtain global temporal information. Experimental validation was conducted using characteristic parameters extracted from a railway DC relay electrical lifespan testing system platform. The results indicate that the proposed model outperforms in terms of prediction accuracy (with an average of 96.85% precision), minimal error rate, and enhanced stability compared with existing models, thereby confirming the practical feasibility of this prediction framework.