摘要
【目标】为了探究货车燃油效率的主要影响因素及因素间的交互作用,提高燃油效率预测的精度,基于WOA-XGBoost和SHAP方法,构建了一种可解释货车燃油效率预测模型。【方法】首先,根据燃油效率的定义重新划分样本并计算相关特征。其次,使用鲸鱼算法(WOA)和灰狼算法(GWO)对XGBoost的超参数进行寻优,根据评价指标对不同种群下的模型进行综合排序以找出2种算法下XGBoost的最优参数。随后,使用WOA-XGBoost,GWO-XGBoost,XGBoost,LightGBM,Random Forest和SVR这6种模型进行燃油效率预测并对模型的综合排序进行比较。最后,利用SHAP归因方法对综合排序最高的模型进行解释。【结果】综合排序最高的模型是WOA-XGBoost,其测试集的评价指标MSE,MAE,RMSE,R^(2)值分别为0.2512,0.1457,0.5012,0.9680;巡航时间和平均速度是影响燃油效率的主要因素,平均SHAP值分别为1.62和0.86;巡航时间和平均速度为交互作用最大的特征,平均速度大于40 km/h时对燃油效率有正向影响,反之则有负向影响。【结论】本研究方法在燃油效率预测方面具有优越性,研究结果对优化驾驶行为具有一定指导意义。
[Objective]The study investigated the main factors influencing truck fuel efficiency and their interactions,so as to improve the accuracy of fuel efficiency prediction.An interpretable truck fuel efficiency prediction model based on WOA-XGBoost and SHAP was constructed.[Method]First,the samples were reclassified according to the definition of fuel efficiency.The relevant features were computed as well based on this definition.Then,the whale optimization algorithm(WOA)and grey wolf optimizer(GWO)were used to optimize the hyperparameters of XGBoost.The models with different populations were comprehensively ranked based on evaluation metrics to identify the optimal parameters of XGBoost by using both algorithms.Subsequently,six models,i.e.,WOA-XGBoost,GWO-XGBoost,XGBoost,LightGBM,Random Forest,and SVR,were used for fuel efficiency prediction.A comparative analysis on the models rankings was performed.Finally,SHAP attribution method was used to interpret the model with the highest ranking.[Result]WOA-XGBoost model has the highest comprehensive ranking.Its evaluation metrics,i.e.,MSE,MAE,RMSE,R^(2),are 0.2512,0.1457,0.5012,0.9680 respectively.Cruise time and average speed are the main factors influencing fuel efficiency,with average SHAP values of 1.62 and 0.86 respectively.Cruising time and average speed are the most significant features for interaction.The average speed has a positive effect on fuel efficiency when it is over 4 km/h,otherwise it has a negative effect.[Conclusion]The proposed method demonstrates superiority in fuel efficiency prediction.The study result will provide valuable guidance for optimizing driving behavior.
作者
杜凯
史晴晴
李乐天
宋京妮
肖梅
陈丹
DU Kai;SHI Qingqing;LI Letian;SONG Jingni;XIAO Mei;CHEN Dan(School of Electronics and Control Engineering,Chang’an University,Xi’an,Shaanxi 710064,China;School of Transportation Engineering,Chang’an University,Xi’an,Shaanxi 710064,China;Chang’an Dublin International College of Transportation,Chang’an University,Xi’an,Shaanxi 710064,China)
出处
《公路交通科技》
北大核心
2025年第7期202-213,共12页
Journal of Highway and Transportation Research and Development
基金
陕西省自然科学基金项目(2022JM-426)
陕西省社会科学基金项目(2020D028)
中央高校基础研究基金项目(300102341104)。
