State of charge estimation for Li-ion battery based intelligent algorithms

State of charge (SOC) is a crucial index for a battery’s energy assessment. Its estimation is becoming an increasing challenge in order to assure the battery's safety and efficiency. To this end, many methods can be found in the scientific literature with various accuracy and complexity. However, accurate SOC is highly dependent on the adopted methodology. This paper investigates five methods for estimating battery SOC for lithium-ion (Li-ion) manufacturers. For this purpose, five methods were selected and then used in practice, including the modified Coulomb counting method, the extended Kalman filter, the neural network (NN), and two other techniques based on machine learning known as the support vector machines and the K-nearest neighbor algorithm (KNN), respectively. A detailed analysis based on statistical assessment is performed on an experimental test that covers multiple cycles of charge and discharge modes. The KNN method proved to be more accurate than the EKF approach, which is extensively used for estimating the SOC of Li-ion batteries. The algorithm demonstrated great predictive accuracy, with most predictions having a relative error near zero and a maximum error value of roughly 0.26%. All of the findings validate the methodology's reliability and efficiency.