Study On The State-of-charge And State-of-health Estimation Algorithms For Lithium-ion Battery

In electric vehicles,the battery management system can control the power usage status of the entire electric vehicle,and the state of charge(SOC)estimation and the state of health(SOH)estimation are the core of battery management system.Based on accurate SOC estimation,the remaining capacity of the electric vehicle could be known in time.Accurate SOH estimation could ensure that the electric vehicle was in healthy state.In this paper,the 18650 lithium-ion power batteries commonly used in electric vehicles were used as experimental samples,whose charging and discharging cycle experiments were carried out under the dynamic state test condition(DST),the standard test condition(QCT),the mixed condition of DST and QCT and constant current condition.Meanwhile,charging and discharging experiments were conducted at three different temperatures(0?,25?,45?)under above conditions.Then,the parameter identification method for equivalent circuit model is explored.Finally,the SOC and SOH estimation algorithms have been studied.(1)A linear neural network is established to identify the parameters of the second-order equivalent circuit model of lithium-ion batteries.The results show that the electrochemical polarization resistance has the severe impediment to the discharge process of lithium-ion batteries,and the electrochemical polarization resistance and concentration polarization resistance were changed abruptly with the sudden change of the current.Ohmic resistance increased significantly with the decline of battery lifetime,while electrochemical polarization resistance and concentration polarization resistance would not increase strictly with the decline of battery lifetime,so ohmic resistance could reflect the change of battery life decay.(2)According to the time series relationship of lithium-ion battery charging and discharging data,long-short term memory(LSTM)recurrent neural network is selected as the estimation model of SOC.By analyzing the influences of temperature,working conditions and battery life decay on SOC estimation,voltage,current,residual capacity,previous estimated SOC and ohmic resistance are selected as input parameters of LSTM network.Meanwhile,in order to enhance the robustness of estimation,the parameters of consecutive 3 seconds are selected as an input vector of the LSTM network.The results show that the average estimation error is less than 1.6% and the maximum estimation error is less than 2.5% under different temperature,life decay and working conditions.At the same time,the LSTM network is proved to have high accuracy and robustness on estimating SOC of lithium-ion battery by comparing with BP neural network.(3)According to the influencing factors of SOH estimation of lithium-ion batteries,more data are needed for SOH estimation of lithium-ion batteries to analyze the characteristics comprehensively under current cycle.A gated recurrent unit(GRU)recurrent neural network is proposed to estimate SOH of lithium-ion battery.Compared with LSTM cell,GRU cell has simpler structure and is more suitable for estimating long time series data.Current,residual capacity,SOC and ohmic resistance are selected as input parameters of the model.Through optimization,it is concluded that the parameters of 200 sampling times should be used as an input vector of the GRU network.The results show that the average error of the estimation is less than 2.2% under four different working conditions,and the average error decreases with the decline of battery life approximately.