State of Health Estimation for Lithium-Ion Batteries Based on Feature Fusion and Multilayer Perceptron

Abstract Electric vehicles have become an important component of modern green ecosystems, and the safe operation of electric vehicles largely depends on the state of health (SOH) of lithium-ion batteries. However, battery aging involves complex physical and chemical processes, making accurate estimation of SOH challenging. To improve the accuracy of SOH estimation and its generalization ability in different datasets, this paper introduces a new model that utilizes Principal Component Analysis (PCA) fuse multiple different features as model inputs, and establishes a Multilayer Perceptron (MLP) model optimized by Grey Wolf Optimization (GWO) algorithm for SOH estimation. MLP is an artificial neural network widely used in various artificial intelligence problems, and optimization algorithms were added in this paper to address the inherent limitations in model training, thereby improving the stability and performance of the model. This paper extracts four different features from battery voltage data, verifies the effectiveness of the features through correlation coefficients, and uses PCA algorithm to fuse all features as model inputs. The fused features not only reduce the input dimension of the model but also improve feature correlation and model efficiency.