对于目前插电式混合动力汽车能量管理功率分配的实时性与准确性问题，现有的离线功率预测模型未能全面考虑混合动力汽车蓄电池的健康状态（SOH），以及功率波动等动态性能多目标变化对预测结果的影响，同时现有的机器学习算法在混合动力汽车的动力分配计算及预测方面也存在性能不足问题，故通过离线动态规划+在线XGBoost算法，实现对混合动力汽车的功率分配建模及预测。首先搭建插电式混合动力汽车的动力系统模型，并且通过聚类分析获取车辆行驶的典型混合工况，其次使用动态规划算法离线计算该工况下发动机与锂电池的最优功率分配比例，最后 XGBoost算法以动态规划计算结果作为训练数据进行模型训练验证。计算结果表明：离线阶段的动态规划考虑了多目标优化，使得在线阶段的模型训练有足够的数据支撑。对比随机森林算法，XGBoost算法将最大误差降低了28%，同时将计算速度提升了62%，可以实现对插电式混合动力汽车功率分配的精确估计。

The real-time and accuracy problems of power distribution exist in plug-in hybrid electric vehicle energy management. The existing off-line power prediction model fails to fully consider the influence of multi-objective changes in dynamic performance of hybrid electric vehicle battery such as SOH and power fluctuation on the prediction results, and the existing machine learning algorithm has insufficient performance in power distribution calculation and prediction of hybrid electric vehicles. In order to solve these problems, the off-line dynamic programming and online XGBoost algorithm was used to model the power distribution of hybrid electric vehicles and realize the power distribution prediction. Firstly, the power system model of plug-in hybrid electric vehicle was built, and the typical mixed driving conditions of vehicle were obtained by cluster analysis method. Secondly, the optimal distribution ratio of engine power and lithium battery under the working condition was calculated off-line by dynamic programming algorithm. Finally, XGBoost algorithm was used as training data to verify the model. The calculation results show that the considered multi-objective optimization in the off-line dynamic planning makes the model training in the online stage have sufficient data support. Compared to the random forest algorithm, the XGBoost algorithm reduces the maximum error by 28% and increases the computational speed by 62%, which enables accurate estimation of the power distribution of plug-in hybrid vehicles.