Advanced RUL Estimation for Lithium-Ion Batteries: Integrating Attention-Based LSTM with Mutual Learning-enhanced Artificial Bee Colony Optimization

In this investigation, we address the critical challenge of forecasting the remaining useful life (RUL) of lithium-ion batteries, a crucial aspect for preventing battery failures and optimizing their overall lifespan. To accomplish this goal, we introduce innovative methodologies according to attention-based LSTM (long short-term memory), a specialized architecture particularly suited for predicting values in time-series data. Our LSTM model incorporates 1D dilated convolution layers, which facilitate the simultaneous extraction of feature vectors, enhancing subsequent prediction processes by combining these vectors. Furthermore, we present a novel approach called ML-ABC (Mutual Learning-based Artificial Bee Colony) for the starting weight pre-training of the proposed model. ML-ABC dynamically adapts the optimal “food source” for potential fixes, merging elements of reciprocal learning that are associated with the starting weights. In order to provide a comprehensive dataset for accurate RUL estimation, we leverage multiple measurable parameters obtained from the battery management system, including temperature charging profiles, voltage, and current. Unlike conventional models featuring one-to-one input–output structures, our method employs a many-to-one structure. This flexible approach accommodates various types of input while also reducing the number of parameters, consequently enhancing the model’s ability to generalize its predictions. We conducted thorough evaluations using NASA lithium-ion battery datasets and achieved remarkable results. Our model demonstrated superior performance compared to other deep learning models, achieving MAPE (mean absolute percentage error) values of 0.8521 for Battery #5, 1.0230 for Battery #6, 0.0930 for Battery #7, as well as 1.0126 for Battery #18. This success underscores the efficacy and superiority of our proposed approach in accurately forecasting what is left of the usable lifetime of lithium-ion batteries. This advancement carries significant implications for diverse industries reliant on these energy storage systems.