基于维度规约的快速路全时段排放模型适应性研究

针对城市快速路汽车污染物排放控制需要，紧扣不同排放模型在映射不同时段排放影响因素与排放率关系方面的差异，以排放测试车辆实际工况排放序列为数据源，分别将反向传播神经网络(Back Propagation Neural Network, BPNN)、广义回归神经网络(General Regression Neural Network, GRNN)和径向基函数神经网络(Radial Basis Function, RBFNN)与平均影响值(Mean Impact Value, MIV)方法相结合，构建维度规约模型。以95%累计贡献率为阈值对排放预测模型输入维度进行降维的基础上，分析神经网络在维度规约前后在不同时段的预测污染物排放率适应性。结果表明：维度规约后BPNN和GRNN模型的R2及MSE在全时段排放数据集中的预测性能提升1.19%、10.14%、6.51%、15.56%,RBF模型对维度规约不敏感；全时段GRNN模型的R2和其余两个模型相比提高10.18%和7.68%,MSE和其余两个模型相比降低0.0396和0.0446，同时MAPE显著降低7.38%和3.86%，揭示GRNN模型在...

Adaptability of Expressway Full Time Emission Model Based on Dimensionality Reduction

To meet the needs of urban expressway vehicle pollutant emission control and consider the differences of emission models in mapping the relationship between emission influencing factors and emission rates in different periods, this paper uses the real condition emission rate data and develops a dimensionality reduction model based on the Back propagation neural network (BPNN), General regression neural network (GRNN), Radial Basis Function (RBFNN) and Mean Impact Value (MIV). To reduce the dimension of the input of the emission prediction model with 95% cumulative contribution rate as the threshold, the study analyzes the adaptability of the neural network in different periods before and after the dimension specification. The results indicate that after dimensionality reduction, the prediction performance of BPNN and GRNN models in R2 and MSE evaluation dimensions in the full-time emission dataset is respectively improved by 1.19% , 10.14% , 6.51% and 15.56% . The RBF model is not sensitive to dimensionality reduction. The Full time GRNN model R2 is respectively 10.18% and 7.68% higher than BPNN and RBFNN, the MSE is respectively 0.0396 and 0.0446 lower than the other two models, and the MAPE is respectively 7.38% and 3.86% lower than the other two models. It also reveals that the GRNN model is more robust than BPNN and RBFNN in predicting expressway pollutant emissions. By analyzing the prediction performance of GRNN in different periods, the predicted R2 in the normal peak period is respectively 3.10% and 4.37% higher than that in the early peak and late peak. The MSE is respectively 0.0303 and 0.0157 lower than that in morning peak and evening peak, and the MAPE is respectively 0.4117 and 0.2857 lower than other two periods. The abnormal fluctuation of emission time series under the influence of traffic status, traffic flow and driving behavior in different periods of expressway has a significant impact on the robustness and generalization ability of the emission model, which provides a basis for the emission model to include the emission period, traffic status and other parameters in the future.