End-to-End Framework for Predicting the Remaining Useful Life of Lithium-Ion Batteries

TL;DR

This work addresses RUL prediction for lithium-ion batteries by presenting an end-to-end framework that fuses a novel signal preprocessing pipeline with a hybrid neural architecture. The preprocessing constructs a rich feature tensor, including the derived capacity derivative $\u0307{Q}_i(Q,I)$ and delta features, to capture short-term degradation, while the CNN+Attentional-LSTM and ODE-LSTM branches model both local and continuous-time dynamics. Across two large public datasets, the method achieves state-of-the-art RMSE and MAPE and demonstrates robust transfer learning capabilities under limited target data. The approach holds promise for proactive maintenance and reliable battery management in real-world applications, with potential extensions to other chemistries and real-time edge deployment.

Abstract

Accurate prediction of the Remaining Useful Life (RUL) is essential for enabling timely maintenance of lithium-ion batteries, impacting the operational efficiency of electric applications that rely on them. This paper proposes a RUL prediction approach that leverages data from recent charge-discharge cycles to estimate the number of remaining usable cycles. The approach introduces both a novel signal processing pipeline and a deep learning prediction model. In the signal preprocessing pipeline, a derived capacity feature $\dot{Q}(I, Q)$ is computed based on current and capacity signals. Alongside original capacity, voltage and current, these features are denoised and enhanced using statistical metrics and a delta-based method to capture differences between the current and previous cycles. In the prediction model, the processed features are then fed into a hybrid deep learning architecture composed of 1D Convolutional Neural Networks (CNN), Attentional Long Short-Term Memory (A-LSTM), and Ordinary Differential Equation-based LSTM (ODE-LSTM) blocks. This architecture is designed to capture both local signal characteristics and long-range temporal dependencies while modeling the continuous-time dynamics of battery degradation. The model is further evaluated using transfer learning across different learning strategies and target data partitioning scenarios. Results indicate that the model maintains robust performance, even when fine-tuned on limited target data. Experimental results on two publicly available large-scale datasets demonstrate that the proposed method outperforms a baseline deep learning approach and machine learning techniques, achieving an RMSE of 101.59, highlighting its strong potential for real-world RUL prediction applications.

Figures (6)

• Figure 1: Overall architecture of our proposed end-to-end framework.
• Figure 2: Smoothed incremental capacity (dQ/dV) curves during charging for selected cycles on a battery cell from the second dataset: full voltage range (left) and zoomed-in region (3.4–3.6 V, right).
• Figure 3: RMSE comparison of different denoising methods (left) and the impact of the window length parameter on Savitzky-Golay method (right).
• Figure 4: RMSE comparison of feature combinations (left) with delta set to 9, and the effect of delta on predictive performance (right).
• Figure 5: RMSE Comparison Across Configurations of the Proposed Prediction Model.