Hadoop环境下基于神经网络的交通流预测方法研究

发布时间：2018-05-28 06:45

本文选题：BP神经网络 + K近邻　；参考：《北京交通大学》2017年硕士论文

【摘要】：随着经济的快速发展,交通运输已经迅速成为国民经济发展命脉。虽然,交通行业的快速发展给人们带来了巨大便利,但随之而来的就是严重的交通拥堵问题。实时准确的交通流预测是交通引导系统中的关键技术,而交通路线系统是智能交通系统的重要组成部分。由于交通系统是一个有人为因素参与的非平稳的随机系统,传统的线性模型预测越来越不适应于非线性的交通预测了,智能预测和组合优化模型越来越受到人们的关注。本文深入研究符合交通流数据特性的代表性预测方法,经分析选取人工智能中的的经典方法BP神经网络作为交通流预测的基本算法。传统的BP神经网络在进行交通流预测时,训练时间和训练精度往往不能同时得到保证。首先,本文分析传统BP神经网络的预测模式,提出一种将输出层的动态值变为定值的预测模式加速收敛。其次,在定值输出层的基础上,针对神经网络训练时间长的缺点,提出一种利用K近邻算法优化训练数据集的K-BP预测模型。该模型在提前考虑预测数据与训练数据匹配度的前提下,进行BP神经网络的训练数据集筛选。相比于传统神经网络,该模型在缩短训练时间的前提下减小了训练误差。随着信息技术与物联网技术在城市交通领域的广泛应用,城市交通流量的数据已经呈现出大数据的诸多特征。传统的神经网络预测模型在小规模训练样本的前提下,还能满足交通流预测需求。但随着训练样本的维度和数据量不断增多,传统的神经网络在训练样本方面往往会消耗过长时间,不利于实现实时的短时交通预测。本文提出了在Hadoop环境下利用MapReduce的分布式处理框架与BP神经网络相结合的预测模型,该模型利用BP神经网络的MapReduce并行化在保证预测精度的同时减小预测时间,达到预测的实时性。
[Abstract]:With the rapid development of economy, transportation has become the lifeline of the development of national economy. Although the rapid development of the transportation industry has brought great convenience to people, it is followed by serious traffic congestion. Real-time and accurate traffic flow prediction is a key technology in traffic guidance system, and traffic route system is an important part of intelligent transportation system. Because the traffic system is a non-stationary stochastic system with the participation of artificial factors, the traditional linear model prediction is becoming more and more unsuitable for nonlinear traffic forecasting, and intelligent forecasting and combinatorial optimization models have attracted more and more attention. In this paper, the representative forecasting methods which accord with the characteristics of traffic flow data are deeply studied. The classical artificial intelligence method BP neural network is selected as the basic algorithm of traffic flow prediction. The traditional BP neural network can not guarantee the training time and precision simultaneously in traffic flow prediction. Firstly, this paper analyzes the prediction model of traditional BP neural network, and proposes a prediction model which can change the dynamic value of the output layer into a fixed value. Secondly, on the basis of constant output layer, aiming at the disadvantage of long training time of neural network, a K-BP prediction model is proposed to optimize the training data set using K-nearest neighbor algorithm. The training data set of BP neural network is filtered on the premise of considering the matching degree between prediction data and training data in advance. Compared with the traditional neural network, the model reduces the training error on the premise of shortening the training time. With the wide application of information technology and Internet of things technology in the field of urban transportation, the data of urban traffic flow have shown many characteristics of big data. The traditional neural network forecasting model can meet the demand of traffic flow forecasting on the premise of small scale training samples. However, with the increasing of dimension and data volume of training samples, the traditional neural networks often consume too long time in training samples, which is not conducive to real-time short-term traffic prediction. In this paper, a prediction model based on the distributed processing framework of MapReduce and BP neural network is proposed in Hadoop environment. In this model, the MapReduce parallelization of BP neural network is used to ensure the prediction accuracy and reduce the prediction time to achieve the real-time prediction.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183;U491.1

【参考文献】