State of Charge Estimation of Li-Ion Batteries Using Random Forest Regression Model with Modified Parameters for Multiple Cycles

This paper proposes the use of Random Forest Regression (RFR) algorithm for accurate estimation of State of Charge (SOC) of Lithium Ion Battery (LIB) over multiple charge–discharge cycles. LIB discharge cycle parameters, i.e., voltage and current are used to get the basic relationship between open circuit voltage (OCV) and SOC. A normalization curve is used for mapping the OCV and SOC of LIB. The normalization curve parameters obtained from the experimental and standard dataset, RFR model has been suitably trained. Nonlinearity in OCV-SOC relationship and fall in SOC over the successive discharge cycles are the major sources of error in SOC estimation. To address these issues, the RFR model has been modified using SOC slope over successive discharge cycles as an additional parameter. Performance check and validation of trained RFR model is done in terms of mean squared error (MSE) score. Results indicate that SOC estimation of battery using proposed RFR model can productively curb the error below 0.15295 for whole range of SOC and 0.04995 for partial (10–90%) range of SOC.