An Adaptive LSTM Network With Atangana-Goufo Difference Operator for Enhanced State of Charge Estimation of Lithium-Ion Batteries

Abstract

The accurate estimation of the State of Charge (SOC) for Lithium-Ion Batteries (LIBs) remains a critical challenge for Battery Management Systems (BMS). Due to complex and dynamic charging/discharging conditions, the inherent fixed memory structure of conventional Long Short-Term Memory (LSTM) networks fails to adequately bridge the discrepancies between historical data and real-time measurements, constraining their adaptability to rapid operating scenarios. To address this, this paper proposes a novel adaptive LSTM network based on the Atangana–Goufo (AG) fractional-order difference operator, named LSTM-AG network. The key innovation lies in the integration of the AG operator, which develops non-local and fractional-order properties well-suited for modeling complex dynamic systems with the LSTM framework. This integration establishes a fractional-order memory gating mechanism that dynamically and self-adaptively balances the contribution of long-term historical information against current inputs, overcoming the memory rigidity of conventional LSTM. This adaptive capability effectively enhances the modeling flexibility and responsiveness to fluctuating operating conditions. Comprehensive experimental validations under various operating conditions demonstrate that the proposed LSTM-AG network outperforms the standard LSTM network with significantly higher SOC estimation accuracy, stronger robustness, and better generalization ability.