【作者】 李波；

【导师】 王喜富；

【作者基本信息】 北京交通大学 ， 交通运输工程， 2012， 硕士

【摘要】 随着我国经济的增长和城市化进展,交通拥堵、交通事故频发、尾气污染等交通问题已经成为当今社会普遍关注的焦点。实时而准确的短时交通流量预测可以为城市交通诱导和控制提供数据支持,是解决多种交通问题的关键和基础。本文在对短时交通流量预测问题分析与总结的基础上,对短时交通流量数据的去噪处理与预测问题进行了分析与研究。本文首先分析了交通流预测的基本概念和交通流的特性,并建立了短时交通流量预测的性能指标。为了提高采集到的交通流量数据的准确性和精度,采用了拉格朗日插值定理对错误数据和丢失数据进行了插值处理,然后根据几种小波母函数的特点,选取db5小波对交通流量数据做3层分解,并进行数据去噪处理。为了提高预测精度和算法的收敛速度,在进行预测前,还对数据进行了归一化处理。针对BP神经网络和遗传算法各自在应用上的缺陷,用遗传算法来优化BP神经网络,建立遗传神经网络的预测模型。在预测仿真阶段,分别采用了BP神经网络、遗传神经网络对原始数据和小波去噪数据进行了预测,并对各组的预测结果的性能指标进行了计算,通过对比分析,得出用遗传神经网络预测的去噪数据拟合度较高,误差较小,在理论上是可行的,能够提高短时交通流量预测的精度和准确性。更多还原

【Abstract】 With the economic growth and urbanization progress of our country, Traffic congestion, frequent accidents, exhaust pollution and traffic problems have become the focus of widespread concern in today’s society. Real-time and accurate short-term traffic flow forecasting can provide data support for traffic guidance and control, and it is the base and key to solve many of traffic problems. This paper first analyses and summarizes of the short-term traffic flow prediction, on this basis, do some research on denoising and prediction of short-time traffic data.This paper first analyses the characteristics of the basic concepts of traffic flow forecasting and traffic flow, and establishs performance indicators for short-term traffic flow prediction. In order to improve the accuracy and precision of the collected traffic data, uses the Lagrange interpolation theorem on the error data and missing data interpolation. According to the characteristics of several mother wavelet, select the db5wavelet decomposes3layer on the traffic flow data and denoising. The data were normalizes in order to improve the prediction accuracy and speed of convergence of the algorithm before making the forecast. Considering BP neural network and genetic algorithm defects in the application, this paper uses genetic algorithm to optimizes the BP neural network and then established genetic neural network prediction model. In predicting the simulation phase, this paper uses BP neural network and genetic neural network to predict the raw data and the wavelet denoising data, and calculates the performance indicators for the prediction results of each group. Through comparative and analysis, this paper comes to a conclusion that the denoising data predicted by the genetic neural network have a higher goodness of fit and small error, and the approach is theoretically possible and can improve the accuracy of short-time traffic flow prediction.更多还原