SOE Estimation For Lithium-ion Batteries Based On Data Fusion

State of Charge(SOC)is the core state of lithium battery,representing the remaining battery power.On the one hand,SOC cannot be directly measured,but can only be obtained indirectly by means of models,algorithms,measurable values,etc.On the other hand,accurate estimation of SOC plays an important role in battery management system and battery performance maintenance.The difficulty and importance of SOC estimation make it a hot research problem.This paper studies SOC estimation from the perspective of data fusion.Firstly,the battery used in the experiment and the battery test platform were introduced,and the relevant experimental data were obtained through dynamic condition test.The recursive least square method with forgetting factor is used to identify the parameters of the equivalent circuit model,which lays a foundation for the subsequent use of the model.Secondly,Kalman filter based on neural network is proposed,which integrates the estimation results of neural network and amptime-integral method,and improves the estimation accuracy of SOC.A feedforward neural network model is used to map the nonlinear relationship between the measurable values and SOC.In order to map the relationship between historical data and SOC,average current and average voltage are added to the input vector of neural network.It is verified by simulation that the SOC result of Kalman filter fusion has higher accuracy than that of neural network.Then,the multi-measurement Kalman filter is proposed for the first time to estimate SOC.The multi-measurement Kalman filter aims to minimize the variance of the estimation error,and obtains the optimal weight,and integrates the estimation results of the two sub-Kalman filters in a weighted way,thus improving the estimation accuracy of SOC.The two sub-Kalman filters are respectively "Kalman filter based on neural network" and "extended Kalman filter based on equivalent circuit model".The experimental results show that the estimation accuracy of multi-measurement Kalman filter is better than that of sub-Kalman filter and feedforward neural network.Finally,to solve the problem that the estimation accuracy and convergence rate decrease due to insufficient prediction of the system noise,adaptive system noise is introduced into the multimeasurement Kalman filter.In the process of introducing adaptive noise,weighted fusion of subKalman filter system noise is proposed to solve the problem of system noise asynchronism.In order to solve the problem of uncertain covariance,we propose to use pseudo inverse matrix instead of matrix inverse.The experimental results show that the estimation accuracy and convergence speed are improved significantly after the introduction of adaptive noise,the estimation error of SOC is reduced to less than 0.015,and the convergence time is reduced to about 50 seconds.