基于IPSO-BP的轻型汽油车道路排放预测

On-road emission prediction of light-duty gasoline vehicles based on IPSO-BP neural network

结合轻型汽油车RDE测试方法,采用便携式排放测试系统(Portable Emission Measurement System,PEMS),对某国六轻型汽油车在市区、市郊和高速路段上进行排放特性测试;建立双隐含层反向传播神经网络,并加入改进粒子群算法(Improved Particle Swarm Optimization Algorithm,IPSO)对BP结构的初始阈值及权值进行寻优;利用主成分分析对输入参数进行降维,CO与NO x排放预测值为目标输出,用试验数据进行训练及验证.结果表明:所建立的IPSO-BP排放预测模型的泛化能力较好,CO与NO x排放预测值能与试验值高度吻合,其平均相对误差分别为10.58%和13.76%;整体排放水平上,CO与NO x排放因子相对误差分别为4.81%和6.4%,该预测模型对监测轻型汽油车实际道路排放具有一定的工程价值.

Combined with the RDE test method for light-duty gasoline vehicles,a Portable Emission Measurement System(PEMS)is adopted to test the emission characteristics of a light-duty gasoline vehicle on urban,rural and high-speed road.Double hidden layer back propagation neural network is established,and Improved Particle Swarm Optimization Algorithm(IPSO)is added to optimize the initial threshold and weight of BP structure.Dimension reduction of input parameters is realized by principal component analysis,the predicted values of CO and NO x emission are used as target output,and the test data are used for training and verification.The results show that the established IPSO-BP emission prediction model had good generalization ability,and the predicted values of CO and NO x emissions are highly consistent with the experimental values,with the mean relative errors of 10.58%and 13.76%,respectively.In terms of the overall emission level,the relative errors of CO and NO x emission factors are 4.81%and 6.4%,respectively.This prediction model has certain engineering value for monitoring the actual road emissions of light-duty gasoline vehicles.