State of Health Prediction Method for Ship Propulsion Batteries Based on a Parallel TCN-LSTM Architecture

This study proposes a State of Health (SOH) prediction method for ship propulsion batteries. First, three health-related features are extracted from battery charging voltage curves. Next, a Temporal Convolutional Network (TCN) and a Long Short-Term Memory (LSTM) network are constructed to capture the complex nonlinear relationships between these features and battery SOH, producing high-dimensional feature representations. Finally, a cross-attention mechanism fuses the output features of the TCN and LSTM into a parallel TCN-LSTM model, thereby improving SOH prediction accuracy. Experimental results demonstrate that the proposed TCN-LSTM model achieves a prediction error of 0.0111-representing a 37.99% reduction compared to the standalone TCN model and a 45.32% reduction compared to the standalone LSTM model, which highlighting the superiority of the proposed approach.