«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

30

期数:

2020年01期

页码:

135-139

栏目:

应用开发研究

出版日期:

2020-01-10

文章信息/Info

Title:

Short-term Traffic Flow Prediction Based on Wavelet and Neural Network

文章编号:

1673-629X(2020)01-0135-05

作者:

李佩钰

长安大学 信息工程学院,陕西 西安 710064

Author(s):

LI Pei-yu

School of Information Engineering,Chang’an University,Xi’an 710064,China

关键词:

短时交通流预测; 小波变换; 去噪; BP 神经网络

Keywords:

short-term traffic flow prediction; wavelet transform; denoising; BP neural network

分类号:

U491.14

DOI:

10. 3969 / j. issn. 1673-629X. 2020. 01. 024

摘要:

针对实际交通流变化的不稳定性和复杂性的特点,应用交通流预测模型获取更准确的交通流信息,是智能交通领域的一个研究热点。 提出一种基于小波分析与神经网络结合的预测模型。 模型主要思想是通过小波多分辨率分析和Mallat 算法对原始交通流数据进行平滑降噪处理,处理过程选用 db10 小波和软阈值去噪函数使得交通流曲线更加平滑稳定,更能真实反映交通流的真实情况;再采用激活函数为 Tan-Sigmoid,训练函数为 trainlm,各层神经元节点数为 1-12-1 的三层 BP 神经网络对消噪后的交通流数据进行训练,用训练好的预测模型对实际交通流信息进行预测,最后获取准确的交通流信息。实验结果表明, 采用小波分析与 BP 神经网络结合的方法得到的预测结果平均相对误差为 0.03%, 最大相对误差为0.39, 拟合度(EC)达到0.96。 仅使用 BP 神经网络预测模型对交通流数据进行预测后得到的预测结果的平均相对误差为 0郾 08%,最大相对误差为 0.89%; 实验对比采用 BP 神经网络预测模型和卡尔曼滤波、GM(1,1)预测模型对交通流的预测,BP 神经网络预测模型的误差指标大大减小,拟合度大大提高,有较好的准确性和可行性,能较准确地反映交通流真实情况。而经过小波去噪与 BP 神经网络结合的预测模型提高了预测精度,为交通流的实时动态预警提供了更加准确真实的情况。

Abstract:

Aiming at the instability and complexity of actual traffic flow changes,it is a research hotspot in the field of intelligent transportation to apply traffic flow prediction model to obtain more accurate traffic flow information. We propose a prediction model based on wavelet analysis combined with neural network,whose main idea is to smooth the noise reduction of the original traffic flow data by wavelet multi-resolution analysis and Mallat algorithm. The db10 wavelet and soft threshold denoising function in the process are used to make the traffic flow curve more smooth and stable,and more realistically reflect the traffic flow. The three-layer BP neural network with activation function as Tan-Sigmoid,training function as trainlm and the number of neuron nodes in each layer as 1-12-1 is adopted to train the traffic flow data after denoising. The trained prediction model is used to predict the actual traffic flow information,and finally the accurate traffic flow information is obtained. The experiment shows that the average relative error of the prediction obtained by wavelet analysis combined with BP neural network is 0.03%,the maximum relative error is 0.39,and the fitness (EC) is 0.96. The average relative error of the prediction obtained by using BP neural network prediction model to predict traffic flow data is 0.08%,and the maximum relative error is 0.89%. The experimental comparison uses BP neural network prediction model and Kalman filter,GM(1,1) prediction model for traffic flow prediction. BP neural network prediction model has greatly reduced the error index,greatly improved the fitting degree,with better accuracy and feasibility,and can accurately reflect the real situation of traffic flow. The prediction model combined with wavelet denoising and BP neural network improves the prediction accuracy and provides a more accurate and true situation for the real-time dynamic warning of traffic flow.

常用功能

更新日期/Last Update: 2020-01-10