基于多时间尺度双扩展卡尔曼滤波的电池峰值功率估计方法

State of Power (SOP) estimation is one of the core functions of battery management system (BMS). Current SOP estimation methods are mainly divided into three categories: characteristic map method, data-driven method and multi-constraint model-based estimation method. However, in practical applications, traditional SOP estimation methods are generally difficult to obtain accurate and realistic peak power. Therefore, a large number of battery cell, module and battery pack experiments are carried out under different temperatures, different working conditions and different aging states in this paper. A battery cell model and battery pack model with dual-characteristic cells uniformly distributed are established to implement the expansion of the cell model to the system algorithm, which is high prediction accuracy and low calculation amount. In addition, a double Kalman filter (DEKF) algorithm based on multi-time scale sliding windows is proposed. The model parameter library is updated based on the peak power test results, which realizes the slow time-varying estimation of the parameters and improves the accuracy and robustness of the algorithm to estimate the peak power. Firstly, an equivalent circuit model (ECM) is established as the battery model in this paper. Then, by comparing 12 common ECMs in terms of modeling accuracy and computational complexity, the Thevenin model is finally selected to simulate battery characteristics, due to the relatively high voltage prediction accuracy and relatively small amount of calculation. Based on the battery cell model, a battery pack model with dual characteristic cells uniformly distributed is established. The two characteristic cells are the real-time highest cell voltage and the lowest cell voltage in the battery pack. The cell numbers of characteristic cells are hence changeable. At the same time, the parallel link is regarded as a large cell, and the model parameters change with the battery characteristics. Based on the assumption of uniform distribution of battery cells, the highest and lowest cell voltage and battery pack voltage are used as input. The calculation formula of the battery system is simplified, and the transformation from the single model to the system model is realized. For the Thevenin model, the open-circuit voltage at time k+1 is taken as the first-order Taylor expansion at time k, and the power battery voltage calculation model is obtained. Considering that the peak power has a positive correlation with the terminal voltage, the dichotomy method is used to try to select a certain peak power. By comparing the corresponding terminal voltage with the target terminal voltage, the dichotomy method is used to correct the peak value Power level, to achieve the purpose of constantly approaching the real peak power. Finally, a hardware-in-the-loop (HIL) simulation platform was built. The experimental objects mainly included 50 A·h three-cell battery cells and 12-string battery modules. On the basis of the initial characteristic test of power battery cells and modules, representative cells and modules that need to be tested are classified and screened, and then a half-year full-life and multi-temperature power prediction is completed in accordance with the experimental requirements and procedures verification and aging cycle test experiments are conducted, and a large amount of experimental data is obtained. The results show that the improved DEKF algorithm is effective, has high accuracy and robustness, and the SOC estimation error is less than 3%. Effective voltage error is less than 40mV.