Study On SOC Estimation Of Lithium Battery Of Electric Vehicles

Due to the asymptotic depletion of fossil energy especially petroleum, as well as the increasingly deterioration of urban atmospheric environment, people are urged to explore novel traffic tools which are more energy-saving and environmental friendly to replace conventional fuel ones, therefore the battery powered electric vehicles have entered their rapid development. The stage of charge(SOC) of the battery is the significant parameter to estimate battery performance and mileage ranges of electric vehicles, of which the estimation accuracy directly affects the driver learning the battery status and scheduling drive plan, even impacts the public acceptance of electric vehicles. However, the study of SOC estimation still stagnates in the initial stage and all kinds of estimation algorithms are lack of applicability and practicality. Aiming to the SOC research of electric vehicles, this paper focuses on the widely used lithium iron phosphate battery to conduct specific study on SOC estimation.Primarily, this paper summarizes and analyzes the existing SOC estimation methods, then establishes two kinds of network model for SOC estimation correspondingly according to the static charging status and dynamic discharging status of lithium iron vehicle battery. For the static charging process, the paper adopts the Particle Swarm Optimization(PSO) algorithm used in Least Square Support Vector Machine(LSSVM) predictive model. Based on the fast convergence and global optimization of particle swarm optimization algorithm the PSO-LSVVM prediction model has solved the parameter optimization problem of the nonlinear model to implement the battery SOC estimation. In view of that the lithium battery discharging of electric vehicles could be influenced by environment, and the battery parameters may change dynamically, a novel recurrent fuzzy neural network named improved self-organizing recurrent fuzzy neural network(SRFNN) is adopted to predict and identify the battery SOC estimation online. Unlike TSK-type fuzzy neural network, the proposed SRFNN adds a functional link neural network(FLNN) to the consequent part of fuzzy rules for promoting the mapping ability. The SRFNN learning starts with an empty rule base, and all of the rules are generated through the simultaneously online learning of structure and parameter.This paper carries on verifications to the SOC simulation both in static charging status and dynamic discharging status with two algorithm net models proposed by MATLAB. Results show that the PSO-LSSVM algorithm is capable to ideally estimate the static battery SOC whether considering the accuracy or stability of the model. Meanwhile, the improved SRFNN algorithm also has promising learning abilities, as well as an excellent approximation accuracy simultaneously guarantee the required generalization, which may realize the online SOC estimation for lithium battery.