Research On State Prediction Of Lithium-ion Battery Based On Fusion Data-driven Method

As a key part of the energy storage system,lithium-ion battery performance will deteriorate with the decrease of capacity and the increase of impedance during continuous charging and discharging,and may also lead to equipment and system failure.Therefore,predicting its state of energy(SOE),state of health(SOH)and remaining useful life(RUL)is of great significance to the reliability of the system.However,it is difficult to directly measure the energy and capacity that characterize the degraded state of lithium-ion batteries.Therefore,based on the data-driven method,this thesis makes indirect predictions of battery pack SOE,battery SOH and RUL by extracting easily measured parameters such as voltage,current,and temperature.For the public dataset,this thesis establishes a fusion model of temporal convolutional network-Gaussian process regression for the prediction of battery pack and cell state.In terms of battery pack,by extracting voltage,current and temperature,which can indirectly reflect the state of the battery pack,the model is trained with 30% of the early data,the SOE of the battery pack in the next cycle is extrapolated and the experimental results are analyzed,the root mean square error is 0.0203,the average absolute error is 0.0154,and the prediction effect is good.In terms of battery cells,this thesis adopts the strategy of indirect prediction,constructs indirect health factors from voltage,current,temperature and other parameters,and finds three more suitable indirect health factors through correlation analysis.Considering that the future indirect health factors are also unknown,the temporal convolutional network is used to predict them,and the updated health factors are used as the input of the subsequent Gaussian process regression model,and only 30% of the early data is used for training,which realizes the early prediction of battery SOH and RUL.