融合多维超声时频域特征的锂离子电池 荷电状态估计

The state of charge (SOC) of lithium-ion batteries is an essential parameter of the battery management system. Accurately estimating the SOC of lithium-ion batteries is crucial for the safe operation of electric vehicles. However, SOC cannot be measured directly, and it can only be estimated by parameters related to the working state of the battery. Meanwhile, due to the highly nonlinear and time-varying characteristics of the battery, the accurate estimation of battery SOC has become a difficult issue. Conventional methods estimate the SOC by using battery voltage, current, temperature, and other parameters. However, the acquisition of these parameters depends on the measurement of electrode behavior. And they are susceptible to factors such as impedance and charge-discharge rate. Compared with conventional signals, ultrasonic signals can discriminate minor changes in the battery materials' physical properties, so they can characterize the battery states accurately. At present, the research on estimating SOC by ultrasound only utilizes the time-domain features of ultrasonic signals, which lacks multi-dimensional analysis. Moreover, due to the nonlinear and non-stationary characteristics of ultrasonic signals, using the time-domain features cannot reflect the changes of ultrasonic signals at different scales, which will reduce the accuracy of SOC estimation. In order to solve the above problems, this paper proposes a lithium-ion battery SOC estimation method that integrates multi-dimensional ultrasonic time-frequency domain features. Multi-dimensional ultrasonic time-frequency domain features which have high correlations with SOC are extracted through joint time-frequency domain analysis of signals. The lithium-ion battery SOC estimation model is proposed by long-short-term memory neural network (LSTM), which realizes the accurate estimation of battery SOC. Firstly, the propagation process of ultrasonic waves in the battery was studied through the continuous uniform layered medium model. The influence of the battery materials' physical properties on the ultrasonic propagation characteristics was analyzed. Secondly, the ultrasonic testing platform for lithium-ion batteries was built and the multi-dimensional ultrasonic time-frequency domain features were extracted. Based on the ultrasonic features, the electrochemical process inside the battery was explained. Finally, considering the special processing ability of LSTM for time series data, the SOC estimation model of lithium-ion batteries fused with multi-dimensional ultrasonic time-frequency domain features was proposed by LSTM. The effects of different fusion features on SOC estimation accuracy were compared. Experimental results show that the accuracy of SOC estimation can be effectively improved by the integration of multi-dimensional ultrasonic time-frequency features. During battery charging and discharging, the root mean square error of SOC estimation is within 0.85% and 0.37% respectively, and the mean absolute error is within 0.58% and 0.24%. At the initial stage of battery charging and discharging, the error of SOC estimation is relatively and neural networks [J]. Chinese Journal of Automotive Engineering, 2021, 11(1): 19-24. error of SOC estimation could decrease at less than 1%, while the maximum error of SOC estimation is more than 22% without modification. Meanwhile, with different capacity modification accuracies, the maximum errors of SOC estimation could be ensured to be less than 1.5%, which reduced the estimation errors and improved the timelines with the joint SOC-SOH estimation model. The following conclusions can be drawn from the analysis: (1) Compared with EKF, the actual dynamic noise covariance is considered in AEKF algorithm proposed. It is more appropriate to establish SOC model to effectively improve SOC estimation accuracy. (2) Fractiona order model can better reflect the polarization characteristics of lithium battery. With the health factors extracted based on charging conditions and fractional order model parameters, SOH estimation model established can reduce the estimation error. (3) AEKF algorithm is used to adaptively monitor the charging and discharging state of lithium battery to acquire accurate health factors. SOH estimation value is used to modify capacity parameters instead of fixed capacity parameters because of actual capacity attenuation. The joint estimation model designed is more suitable for the actual change. It has stronger timeliness and robustness.