Enhancing Battery Capacity Estimation Accuracy through the Neural Network Algorithm

Abstract

Accurate estimation of battery metrics, such as state of health (SOH), is crucial for effective battery management systems (BMS) due to capacity degradation over time. This paper proposes a methodology to enhance battery capacity estimation accuracy by addressing uncertainties related to state of charge (SOC) estimation and measurement. The methodology employs the Neural Network Algorithm (NNA), an optimization algorithm inspired by artificial neural networks (ANNs). The NNA generates an initial population of pattern solutions and iteratively updates them using a weight matrix, bias operator, and transfer function operator. By combining the advantages of ANNs and optimization techniques, the NNA aims to find an optimal solution considering interdependent variables and incorporating global and local feedbacks. Leveraging the capabilities of the NNA, our objective is to identify the candidate that minimizes a specified cost function, ensuring up-to-date cell capacity through a memory forgetting factor. The algorithm's precision was validated using NASA's Prognostic Data, demonstrating outstanding performance by surpassing two aggressive algorithms in terms of accuracy. In the most severe case scenario, the algorithm achieved a peak error of less than 0.4%. Furthermore, the algorithm consistently demonstrated predictive performance measures that were superior to those of the compared algorithms.