A Fast Neural Network-Based Battery SOH Estimation Using Load Surge Response Characteristics

Abstract

Monitoring the performance of lithium-ion batteries is crucial for the manufacture and operation of various industrial applications. The state of health shows the health status of the batteries. This paper proposes a novel method for fast estimation of battery state of health by leveraging response characteristics during load surges and optimizing the process through a genetic algorithm-extreme learning machine model. Traditional estimation techniques often rely on complete charge/discharge profiles, which are inefficient for online monitoring and real-time applications. The proposed method extracts key features from voltage response curves during inrush currents, thereby eliminating the need for full charge/discharge data. Techniques such as discrete wavelet transform and differential voltage analysis are employed to capture vital health indicators. The genetic algorithm-extreme learning machine algorithm significantly reduces computational complexity while ensuring high estimation accuracy by optimizing the parameters of the extreme learning machine. Experimental results demonstrate that the model achieves high accuracy (within 3%) in estimating state of health across various states of charge. This method is particularly suitable for applications requiring rapid and non-invasive battery health assessments, such as backup power systems and electric vehicle startups.