For both the SOC and SOH estimation methods, shortening the amount of voltage and charge/discharge process data is conducive to the determination of the model parameters. For instance, Park et al.  and Ahn et al.  proposed a modified data-driven method based on the empirical mode decomposition for SOC estimation. Since the battery internal resistance is the sum of the ohmic resistance of the electrolyte and the electrode, Li et al.  proposed a multi-model collaborative learning method to estimate the internal battery state, which is more accurate than the one-model-based estimation method. In addition, Zhang et al.  combined the multi-model-collaborative learning method with the adaptive back propagation algorithm to improve the robustness of the system and the accuracy of the parameter estimation. In this technology, the parameter estimation model is connected to the SOC model to determine the optimal parameters of the model. Li et al.  proposed a one-model-based estimation method of battery SOH. To overcome the problem of the mutual consistency between the two parameters, the proposed methods consider the difference between the two parameters, suitable for a change point estimation of SOC as well as under the application of lithium ion batteries.
In order to know the battery health status information in real time and further improve the data-driven battery model, effective parameters constraints are needed. Previously, Tian et al.  established a cache memory model to predict the internal battery temperature that is mainly represented by the electrolyte, which is warm during the first 75 minutes, and the formation time and discharge time of the electrolyte need to be predicted to prevent the battery from overheating. Similarly, Li et al.  considered the temperature, the SOH and the battery voltage to predict the temperature of the electrolyte during the charging process.
By considering the change in the discharge direction when the battery is at different states, Chang et al.  proposed the position relationship of the battery state and the discharge direction and the parameter constraint of the models are based on the model properties, thereby greatly reducing model error. Zuo et al. d2c66b5586