Early Prediction of Remaining Useful Life for Li-ion Batteries Using Transformer Model with Dual Auto-Encoder and Ensemble Techniques

The accurate early prediction of the Remaining Useful Life (RUL) of lithium-ion batteries plays an important role for battery health management and predictive maintenance. The insights derived from early RUL prediction can be harnessed to enhance battery lifespan and also for the early detection of potential battery failures, which is particularly critical in the context of the widespread use of lithium-ion batteries in various applications, such as electric vehicles, renewable energy storage, and portable electronics.To address this challenge, we introduce an novel neural network model, the Dual Auto-Encoder Transformer, combined with ensemble techniques for early RUL prediction. Raw battery data often contains noise. To mitigate this issue, we employ a Dual Auto-Encoder(DAE) to denoise the data, capture crucial features, and reconstruct the information effectively. Our approach goes beyond traditional methods by integrating both data preconditioning (denoising) and early RUL prediction into a unified framework, eliminating the need for separate processes. We conduct our experiments on the MIT Dataset, consisting of 124 lithium-ion batteries data points. The result demonstrate that our proposed method outperforms existing approaches, showcasing a significant improvement in early prediction of the RUL with a Mean Absolute Error (MAE) reduced to 0.0372 and Root Mean Square Error (RMSE) to 0.0451, highlighting the model's accuracy, efficiency, and its potential impact in real-world applications.