State of charge estimation of lithium-ion batteries using PSO optimized random forest algorithm and performance analysis

The State of Charge (SOC) is a vital parameter for Lithium-ion batteries driving electric vehicles and an accurate knowledge of SOC is mandatory for the effective operation and longevity of Battery Energy Storage Systems (BESS). The Data-driven methods are largely employed in recent days towards the SOC estimation. However, various parameters of data-driven methods need to be judicially chosen to accurately estimate SOC. The Random Forest (RF) method is a popular machine-learning algorithm that has been extensively employed for several applications. There are five hyperparameters in the RF algorithm that need to be optimized for reliable prediction of SOC. Towards this goal, this paper proposes a Particle Swarm Optimization (PSO) assisted Random Forest (RF) method for SOC estimation of Lithium-ion batteries. The convergence characteristics of PSO-based optimization are presented and show a faster convergence rate. The optimized RF algorithm is then employed for a public dataset of dynamic profiles namely Federal Urban Driving Schedule (FUDS) and Dynamic Stress Test (DST) and the SOC estimation is carried out. The computed results are presented and observed to be promising. Further, the proposed strategy is compared with Differential Search Algorithm (DSA) based parameter tuning and the proposed method is shown to exhibit superior performance.