基于Python的汽车油耗多参数回归模型构建方法

A Method for Constructing Multi-parameter Regression Model of Vehicle Fuel Consumption Based on Python

利用OBD检测仪获取的车辆运行状态参数及油耗的实时检测数据,使用Python机器学习库中的相关算法,对构建汽车运行油耗多参数回归模型的方法进行了相关研究。以怠速时间、加速度、负荷率、冷却液温度、车速、发动机转速、节气门相对位置与进气管绝对压力等参数作为自变量,平均油耗作为因变量,利用Python的特征选择库对影响油耗的运行状态参数进行排序。得出结果依次为(1)冷却液温度(2)车速(3)节气门位置(4)发动机转速(5)怠速时间(6)负荷率(7)加速度(8)进气管绝对压力。基于提取出的特征参数数据并利用Python中的机器学习库,建立的多元线性回归模型的平均绝对误差为3.01,均方根误差为4.80;多层感知器(MLP)神经网络回归模型的平均绝对误差为0.30,均方根误差为0.48;集成回归模型的平均绝对误差为0.23,均方根误差为0.38。对各个模型进行十折交叉验证的结果为:多元线性回归模型的平均得分为0.68,多层感知器神经网络回归模型的平均得分为0.84,集成回归模型的平均得分为0.86。由此可见,车辆运行油耗与车辆运行状态参数之间的线性特征并不明显,而更适合于建立非线性回归模型,且能够为进一步阐明汽车运行油耗与车辆运行状态参数之间的关系提供理论依据。

According to the vehicle operating state parameters obtained by OBD detector and the real-time test data of fuel consumption,and using the related algorithm in Python machine learning library,the method for constructing the multi-parameter regression model of vehicle fuel consumption is studied.Taking the parameters such as idle time,acceleration,load rate,coolant temperature,vehicle speed,engine speed,throttle relative position and absolute pressure of intake pipe as the independent variables and the average fuel consumption as the dependent variable,the operating state parameters affecting fuel consumption is sorted out by Python’s feature selection library.The sorting result is:(1)coolant temperature,(2)vehicle speed,(3)throttle position,(4)engine speed,(5)idle time,(6)load rate,(7)acceleration,and(8)absolute pressure of inlet pipe.Based on the extracted characteristic parameter data and the machine learning library in Python,the average absolute error of the established multiple linear regression model is 3.01,and the rms error is 4.80;the average absolute error of the multilayered perceptron(MLP)neural network regression model is 0.30,and the rms error is 0.48,the average absolute error of the integrated regression model is 0.23,and the rms error is 0.48.The result of a ten-fold cross-verification of each model shows that the average scores of the multiple linear regression model,the MLP neural network regression model,and the integrated regression model are 0.68,0.84 and 0.86 respectively.Thus,it can be seen that the linear characteristics between vehicle fuel consumption and operating state parameters are not obvious,it is more suitable for establishing nonlinear regression models,which could provide a theoretical basis for clarifying the relationship between vehicle fuel consumption and operating state parameters further.