关键词: 液力机械综合传动装置, 功率损失预测, 堆叠集成学习, 贡献解释

Abstract: Power loss is one of the critical parameters for evaluating the performance of the hydro-mechanical integrated transmission device (HMITD) in tracked vehicles. By predicting power losses under different operating conditions, a more comprehensive assessment of the HMITDs can be achieved, thereby ensuring the good mobility of tracked vehicles. This paper proposes a method based on stacked ensemble learning for predicting power losses of the HMITDs under various working conditions. The study utilizes efficiency data from 75 tracked vehicles under multiple working conditions and integrates multiple algorithms, including Random Forest, LightGBM, AdaBoost, CatBoost, and XgBoost, through stacking to effectively predict power losses of the HMITD. Additionally, SHapley Additive exPlanations (SHAP) values are employed to analyze the impact of various factors and models on power loss prediction. The experimental results show that on the training set, the root mean square error (RMSE) of power loss prediction is 6.6, and the goodness of fit (R²) reaches 0.976; while on the test set, these metrics are 8.92 and 0.961, respectively. Further analysis reveals that input torque is the primary factor affecting power losses, and within the stacked ensemble framework, the Random Forest algorithm contributes the most to improving prediction accuracy. Keywords: hydro-mechanical integrated transmission device; power loss prediction; stacking ensemble learning; contribution explanation

Key words: hydro-mechanical integrated transmission device, power loss prediction, stacking ensemble learning, contribution explanation