SOC and SOH Estimation of Battery Energy Storage Systems Based on Heterogeneous Deep Learning Algorithm Fusion

Introduction: With the wide application of batteries in the field of power storage, it is essential to accurately estimate and evaluate their SOC and SOH to ensure the safe operation of batteries. At the same time, due to the coupling relationship between SOC and SOH, this paper proposes a joint estimation method of SOC and SOH based on deep learning heterogeneous algorithm fusion, aiming to improve the accuracy of state estimation to ensure the safe and stable operation of the battery. Methods: Firstly, an Ampere-hour Integral-based model is developed for state of charge estimation, and a capacity decay model is constructed for state of health estimation. Secondly, the LSSVM and the UKF fusion algorithm are applied to estimate the SOC of the battery. Simultaneously, the CNN and LSTM fusion algorithms are used to estimate the current SOH. Finally, the maximum available capacity of the battery, as estimated by the SOH module, is integrated into the physical model of SOC. Results: This integration leads to a refined SOC estimate, enhancing the estimation accuracy. This paper uses a 150mAh button-type sodium-ion battery as an example for experimental verification. Compared with the measured values, the SOC estimation error for sodium-ion batteries at chargedischarge rates of 0.1C, 0.5C, and 1C is less than 1.1%, and the error accuracy estimate for the SOH estimation model is below 1.3%. Overall, this method has good applicability and accuracy. Discussion: This study achieves joint estimation of SOC and SOH for energy storage batteries through a deep learning heterogeneous algorithm, demonstrating promising accuracy and stability. However, the SOH estimation exhibits declining fitting performance in later cycles, indicating insufficient relevance of the current health indicators. Future work should integrate multi-dimensional features to enhance model generalization capabilities. Conclusion: This study proposes a high-precision joint modeling framework for battery state estimation and validates the collaborative effectiveness of heterogeneous deep learning algorithms. This achievement not only significantly improves the estimation accuracy of SOC and SOH but also provides an effective solution for enhancing battery usage efficiency and extending battery lifespan, thereby strongly advancing the development of energy storage technology and the electric vehicle industry.