Laplacian Kernel Ridge Regression for Novel Non-Linear Battery State-of-Health Estimation in Electric and Hybrid Vehicles

Due to the increase of greenhouse gases causing climate change, many industries are beginning to make changes to their designs and processes. The automotive industry has responded with (hybrid) electric vehicles (EV’s) which either improve gas mileage or forgo it all together. Opting to use grid power stored by lithium-ion batteries (LIB’s) chosen for the energy density, lifespan, and reliability. These LIBs are operated by a battery management system (BMS) which measures and reports battery data while maintaining operational ranges for the LIB. State-of-Health (SOH) is one key metric in BMS that represents a percentage which simply relates the maximum capacity of a LIB versus it’s rated value. Many different methods of SOH estimation have been developed using electrochemical models, equivalent circuit models, open-circuit voltage, and coulomb-counting, all of which struggle due to constant need for recalibration to maintain accuracy. Recent research has come to apply machine learning (ML) to the SOH problem utilizing numerous algorithms reliant on neural networks (NN), gaussian process regression (GPR) or Support Vector Machines (SVM). This paper introduces Laplacian Kernel Ridge Regression (LKRR) as a novel solution to estimate the SOH. The research results show that the proposed method is robust, accurate, and requires lower training costs than other ML techniques for medium-sized datasets.