SOH Estimation for Li-ion Batteries Using Gramian Angle Field and ResNet Neural Network

State-of-health (SOH) prediction of Li-ion batteries is vital for extending the life of the battery and improving system safety. The complex nonlinear recession of the battery and manual extraction of features make the traditional SOH estimation methods deficient in prediction accuracy and robustness. For accuracy improvement, a method for SOH estimation is proposed in this paper based on Gramian angle field (GAF) image transformation and ResNet neural network. The method converts the battery charging data into GASF and GADF image data respectively. Normalization of the images are performed as model input. ResNet neural network is also utilized to extract key health features during battery usage. The features are used to train and test the network for SOH estimation. Our experimental results indicate that this approach can effectively improve the SOH accuracy of estimation and deliver a solution for battery well-being monitoring in a more accurate and efficient way.