A deep learning-based method for estimating the state of charge and predicting the remaining life of energy storage batteries
Year: 2026
Type: article
Abstract: INTRODUCTION: In order to improve the precision of state estimation and forecast of long term deterioration tendency in complicated charging and discharging conditions. OBJECTIVES: This paper adopted a collaborative modeling approach for state of charge (SOC), state of health (SOH), and Remaining Useful Life (RUL) for SOC estimation. METHODS: By integrating the Huber loss and state continuity constraints, this method can capture both the local and long-term temporal dependencies; The SOH estimation uses multiscale convolutions, gated temporal coding, and residual regression to represent cross-cycle degradation characteristics, and generates future health trajectories using an encoder-decoder architecture, which enables RUL threshold search prediction. RESULTS: The results show that the average absolute error of SOC is 0.63%, and the root mean square error is 0.86%, and the SOH estimate reaches a mean absolute error of 0.58%, and the RUL estimate reaches 13.8 cycles. CONCLUSION: Under joint learning, the SOC root mean square error, SOH mean absolute error, and RUL mean absolute error are further reduced to 0.79%, 0.55%, and 12.9 cycles, respectively, confirming that shared degradation representation and task-coordinated optimization improve state estimation and lifetime prediction.
DOI: 10.4108/ew.14866
Authors: Yuanjing Qian, Ning Zhao, Zhenrong Yang, Chang Liu
Language: English
Subfield: Automotive Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7236730238
