Joint Estimation Of State Of Health And State Of Charge For Lithium-ion Battery

Lithium-ion batteries have been widely used in various energy storage devices due to their high energy density,long cycle life and low cost,such as electronic mobile devices,electric vehicles,aerospace,power grids and defense industries.But the aging of batteries during operation is the main factor LIB s’life.In order to make full use of lithium ion batteries,it is necessary to track the changes of State of Health(SOH)and State of Charge(SOC).So far,it is hard to directly detect the available capacity of LIBs,and the degradation model of LIBs is a complex nonlinear process.At present,state estimation has some problems such as poor accuracy of cold start prediction,dependence on large amount of historical data and poor adaptability of single state estimation.Therefore,this thesis propose a joint SOH-SOC estimation model based on auto-regressive recurrent neural network(DeepAR)for lithium-ion batteries.The model can accurately predict the state of health and state of charge of the battery by using the external monitoring data of the battery.Combined with adaptive white noise unified ensemble empirical mode decomposition(CEEMDAN),the capacity diving point is identified.The main research results are shown as the following four aspects:Ⅰ.Factor analysis of capacity degradation of LIBsTo study the battery characteristics of LIBs during decay,and detailedly describe the working principle and characteristics of LIBs.Then evaluate and analyse the factors affecting the actual capacity decay rate of power battery.It is pointed out that the cycle times,discharge rate,discharge temperature and other parameters are the main factors affecting the battery capacity.The fading trend of LIBs is analyzed by charge-discharge curve,and the influence of different factors on the battery state is studied.Ⅱ.Research on SOH-SOC joint estimation algorithm for LIBs based on DeepARTo solve the problem of poor prediction accuracy of lithium-ion battery state cold start by traditional neural network,an auto-regressive recurrent neural network model that can selectively integrate multi-layer RNN,LSTM or GRU is used.The battery degradation mechanism is analyzed from the perspective of data and the correlation between SOC and SOH is expounded.Six groups of different aging characteristics are extracted from the voltage,temperature and time changes during charging and discharging.Three groups of aging characteristic values with high correlation with capacity state are selected as input to predict the health state of the battery.The obtained SOH is used as the performance index in a period of time to predict the state of charge of the battery with the current,voltage and temperature input model.The RMSE and MAE of the combined estimates were all less than 2.5%and less than 2.1%,which was nearly 3.0%higher than that of the single SOC estimates.Ⅲ.Research on LIBs’ capacity diving prediction algorithm based on CEEMDAN-DeepARIn view of the problem that the discharge capacity of LIBs’ decreases sharply in severe environment,the CEEMDAN-DeepAR combined model is constructed by analyzing the capacitance trend,OCV-SOC curve and capacity increment curve.The adaptive noise complete ensemble empirical mode decomposition theory is used to decompose the original battery capacity signal into multiple sets of modal components.The capacity trend component is used to identify the potential trend of capacity degradation in time.The moving average input method is used to input the representation information.The constructed model has better fitting performance in all validation data and can identify capacity diving points.Ⅳ.Experimental design of LIBs capacity estimationThis thesis designed a simple LIBs capacity estimation experimental device,including controller,charging module,signal conditioning and analog-to-digital conversion module,temperature detection module,constant current charging and discharging module and other hardware circuits.The software flow and display interface of Raspberry PI 4B are also designed.By setting different charging and discharging ratios,the corresponding degradation data of battery state is obtained,and then the obtained battery data is jointly predicted,and the diving point is identified.It proves that the model can achieve reliable prediction on the self-measured data under different charging and discharging ratios.In summary,a SOH-SOC joint estimation model based on DeepAR is proposed in this thesis.The prediction algorithm,prediction accuracy,diving point estimation and experimental prototype are mainly designed and studied.The battery capacity estimation experimental platform meets the requirements for 1.6 Ah single lithium-ion battery life verification.