基于SVM与数据降噪方法结合的交通流预测

Traffic Flow Prediction Based on Combination of SVM and Data Noise Reduction Method

摘要: 准确预测道路交通流是智能交通系统的重要组成部分。由于受到一些外部因素的干扰，含有噪声的原始交通流数据可能会导致预测性能下降，为此，本研究提出了一种称为经验独立模态分解(Empirical Independent Mode Decomposition, EIMD)的降噪方法。模拟实验结果表明EIMD方法优于其他降噪方法。在此基础上，本文提出了一种结合EIMD与支持向量机(Support Vector Machine, SVM)的组合预测模型，以提高预测精度。通过使用贵阳市路段的交通数据，评估了不同模型(SVM、小波神经网络(Wavelet Neural Network, WNN)和卷积神经网络(Convolutional Neural Networks, CNN))结合降噪方法的预测性能。预测结果表明，结合EIMD降噪方法的模型预测结果均优于未采用降噪方法的模型，其中，EIMD结合SVM的组合模型预测性能高于其组合模型。研究结果表明，本研究提出的方法能够降低噪声对预测性能的影响，准确地预测道路交通流。这对于智能交通系统具有重要意义。

Abstract: Ensuring precise forecasting of road traffic flux holds utmost importance in the advancement of intelligent transportation systems. Nonetheless, the presence of disturbances in the original traffic flux data may impede the accuracy of predictions. In this investigation, we propose an Empirical Independent Mode Decomposition (EIMD) technique to effectively diminish disturb-ances. Our simulated outcomes showcase the superiority of the EIMD approach when compared to alternative techniques for reducing disturbances. Building upon this, we present a combined forecasting model that integrates EIMD and Support Vector Machine (SVM) to further enhance prediction accuracy. Through the utilization of traffic data from Guiyang City, we assess the performance of diverse combined models, which include Wavelet Neural Networks (SVM), Convolutional Neural Networks (WNN), and Convolutional Neural Networks (CNN), while employing various methods for reducing disturbances. The outcomes unveil that the model incorporating EIMD outperforms the model without disturbance reduction, and the combined model with SVM yields the most favorable prediction performance. These discoveries underline the efficacy of our proposed approach in mitigating the impact of disturbances and accurately forecasting road traffic flux, thereby making a significant contribution to the advancement of intelligent transportation systems.