PCA-BP在城市汽车保有量预测中的应用研究

Research on Urban Car Ownership Prediction Based on PCA-BP Neural Network

城市汽车保有量影响因素众多,且存在复杂的相关关系,传统数学预测模型和神经网络模型,无法消除影响因素之间的相关性,从而导致预测精度较低。为提高城市汽车保有量预测精度,提出了一种基于主成分分析的BP神经网络预测模型。通过对城市汽车保有量影响因子进行主成分分析,消除各因子间的冗余信息,降低BP神经网络的输入维数,简化神经网络拓扑结构,提高城市汽车保有量的训练速度与预测精度。对南京市2006-2009年南京市汽车保有量进行仿真,实现结果表明,PCA-BP模型的训练速度快、预测精度高,可为城市汽车保有量预测提供参考依据。

Variation of urban car ownership is influenced by many factors, and relations between them are nonlinear. Because redundant information existing in the factors can not be eliminated effectively, prediction accuracy of the traditional mathematical model and neural network model is low. To improve the prediction accuracy of urban car ownership, the PCA - BP was proposed. The redundant information among the various factors was removed through principal component analysis on impact factors of urban ownership, the neural network topology structure was simplified, and the training speed and prediction accuracy were improved. The implementation results show that compared with ARIMA, BP and multiple regression analysis, the prediction accuracy of PCA - BP neural network mode is higher and the speed is faster. The method provides a new way for the urban car ownership production prediction.