基于Spark的改进SA-SVR短时交通预测研究
发布时间：2018-10-31 15:58
【摘要】：科技的快速发展为人们的生活带来了便捷,但同时也带来了一些负面影响。交通事故、道路拥堵、以及车辆尾气排放带来的全球变暖等,这些交通问题作为经济发展的负面附属品,是众多问题中急需解决的一个。自交通问题出现之时,对于交通问题的解决方案的研究从未停步,随着智能时代的到来,智能交通系统的概念被提了出来。智能交通系统作为目前解决交通问题的首选,而短时交通流作为智能交通系统的一部分更是得到了研究人员的重视。但是交通流并非是一成不变的,它是一个非平稳的易受外界环境干扰的非线性系统,并且拥有海量的交通流数据,随着时间的推移这些数据也在不断的增长。如何处理这些海量的数据并达到交通流预测的精确性和实时性要求成为近年来的主要研究方向。本文从研究提高短时交通流预测的准确性和实时性出发,主要研究内容包括:(1)研究了适用于处理小样本非线性的支持向量回归机(SVR)。在已有的基础上,对交通流和交通流的数据特点进行研究,基于交通流和交通流数据的特点研究了比较实用的短时交通流预测模型,经过研究对比和实验,验证SVR作为短时交通流预测的可行性和实用性。(2)改进了适用于处理大型组合优化的模拟退火算法(SA),将其应用于支持向量回归机进行参数优化。在选择支持向量回归机的基础上,对支持向量回归机的研究发现支持向量机的参数对于整个模型的预测结果有着至关重要的作用,为了达到建立基于最优参数的短时交通流预测模型,本文研究对比其他传统参数优化算法,确立并改进了适用于处理大型组合优化的模拟退火算法,基于改进后的模拟退火算法对支持向量回归机进行参数优化,并基于最优参数建立了预测模型,解决了短时交通流预测中的预测准确性问题。(3)建立了Spark平台下的SA-SVR预测模型。随着交通流数据量的增加,在处理海量的交通流数据的过程中,单机模式下的预测模型由于物理因素的限制无法满足短时交通流预测对于预测实时性的要求,为了解决预测时间的问题,本文在大数据时代的背景下研究对比选择具有分布式并行处理能力的Spark技术对支持向量回归机做大规模的并行算法训练,并结合了支持向量回归机处理非线性小样本事件的特性和Spark的并行处理时间短的优点,建立了Spark平台下的SA-SVR预测模型。实验证明,此模型在保证了预测精度的前提下缩短了预测的时间,同时满足了短时交通流预测对于精确性和实时性的要求。本文基于预测模型进行了三组对比实验,分别是RBF神经网络与支持向量回归机模型、网格算法与模拟退火算法及改进后的模拟退火算法参数优化模型、单机模式下与Spark并行模式下的预测模型实验对比。这三组对比实验结果证明了基于改进的模拟退火算法对支持向量回归机进行参数优化后的模型在Spark环境下比传统的算法及单机模式下的预测更具有竞争力,Spark平台下的该模型在预测过程中不仅解决了短时交通流预测的精确性问题,也解决了短时交通流预测中的预测效率问题,总体提高短时交通流预测中处理交通流数据的能力及预测的精确性和实时性。本文的主要创新点是将支持向量回归机的稀疏性特点与分布式集群Spark系统的并行处理能力相结合,在Spark平台下进行大规模SVR训练,由此建立了Spark平台下的SA-SVR短时交通流预测模型,该模型很好地解决了短时交通流预测的精确性和实时性问题。
[Abstract]:The rapid development of science and technology brings convenience to people's life, but at the same time has some negative effects. These traffic problems, such as traffic accidents, road congestion and global warming caused by vehicle exhaust emissions, are one of the many problems in the economy. With the advent of the traffic problem, the research on the traffic problem has never stopped. With the advent of the smart age, the concept of the intelligent transportation system has been raised. Intelligent transportation system is the first choice to solve traffic problem, while short-time traffic flow is regarded as part of intelligent transportation system. But the traffic flow is not immutable, it is a non-linear system that is non-stationary and easily disturbed by external environment, and has massive traffic flow data, and these data are constantly increasing over time. How to deal with these massive amounts of data and achieve the accuracy and real-time requirements of traffic flow prediction has become the main research direction in recent years. In order to improve the accuracy and real-time performance of short-time traffic flow prediction, the main research contents include: (1) the support vector regression machine (SVR) suitable for processing small sample non-linearity is studied. Based on the existing data characteristics of traffic flow and traffic flow, a practical short-time traffic flow prediction model was studied based on the characteristics of traffic flow and traffic flow data. (2) The simulated annealing algorithm (SA), which is suitable for processing large-scale combination optimization, is applied to support vector regression machine for parameter optimization. On the basis of selecting the support vector regression machine, the research of support vector regression machine has found that the parameter of support vector machine plays an important role in the prediction result of the whole model, in order to reach the short-time traffic flow prediction model based on the optimal parameters, Compared with other traditional parameter optimization algorithms, this paper establishes and improves the simulated annealing algorithm suitable for processing large-scale combination optimization, optimizes the parameters based on the improved simulated annealing algorithm, and establishes a prediction model based on the optimal parameters. and solves the problem of prediction accuracy in short-time traffic flow prediction. (3) The SA-SVR prediction model under the Spark platform is established. with the increase of the amount of traffic flow, in the process of processing mass traffic flow data, the prediction model in the stand-alone mode cannot satisfy the requirement of short-term traffic flow prediction to predict the real-time performance due to the limitation of physical factors, and in order to solve the problem of the prediction time, In this paper, in the background of the large data era, we study the Spark technology with distributed parallel processing ability to train the support vector regression machine in a large scale, The SA-SVR prediction model under the Spark platform is established by combining the advantages of the support vector regression machine to deal with the nonlinear small sample events and the short parallel processing time of Spark. Experimental results show that this model can shorten the forecast time on the premise of ensuring the prediction accuracy, and meet the requirement of short-time traffic flow prediction on the accuracy and real-time performance. In this paper, three groups of contrast experiments are carried out based on the prediction model, which are the RBF neural network and the support vector regression model, the mesh algorithm and the simulated annealing algorithm and the improved simulated annealing algorithm parameter optimization model, which is compared with the prediction model in the Spark parallel mode under the stand-alone mode. Compared with the traditional algorithm and the single-stand model, the model based on the improved simulated annealing algorithm is more competitive in the Spark environment than in the traditional algorithm and the stand-alone mode. The model not only solves the accuracy problem of short-time traffic flow prediction in the prediction process, but also solves the problem of prediction efficiency in short-time traffic flow prediction, and improves the capability of processing traffic flow data and the prediction accuracy and real-time performance in the short time traffic flow prediction. The main innovation point in this paper is to combine the sparse features of support vector regression machine with the parallel processing capability of distributed cluster Spark system, carry out large-scale SVR training under the Spark platform, and set up the SA-SVR short-time traffic flow prediction model under the Spark platform. The model well solved the accuracy and real-time problem of short-time traffic flow prediction.
【学位授予单位】：东华理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U491.14
[Abstract]:The rapid development of science and technology brings convenience to people's life, but at the same time has some negative effects. These traffic problems, such as traffic accidents, road congestion and global warming caused by vehicle exhaust emissions, are one of the many problems in the economy. With the advent of the traffic problem, the research on the traffic problem has never stopped. With the advent of the smart age, the concept of the intelligent transportation system has been raised. Intelligent transportation system is the first choice to solve traffic problem, while short-time traffic flow is regarded as part of intelligent transportation system. But the traffic flow is not immutable, it is a non-linear system that is non-stationary and easily disturbed by external environment, and has massive traffic flow data, and these data are constantly increasing over time. How to deal with these massive amounts of data and achieve the accuracy and real-time requirements of traffic flow prediction has become the main research direction in recent years. In order to improve the accuracy and real-time performance of short-time traffic flow prediction, the main research contents include: (1) the support vector regression machine (SVR) suitable for processing small sample non-linearity is studied. Based on the existing data characteristics of traffic flow and traffic flow, a practical short-time traffic flow prediction model was studied based on the characteristics of traffic flow and traffic flow data. (2) The simulated annealing algorithm (SA), which is suitable for processing large-scale combination optimization, is applied to support vector regression machine for parameter optimization. On the basis of selecting the support vector regression machine, the research of support vector regression machine has found that the parameter of support vector machine plays an important role in the prediction result of the whole model, in order to reach the short-time traffic flow prediction model based on the optimal parameters, Compared with other traditional parameter optimization algorithms, this paper establishes and improves the simulated annealing algorithm suitable for processing large-scale combination optimization, optimizes the parameters based on the improved simulated annealing algorithm, and establishes a prediction model based on the optimal parameters. and solves the problem of prediction accuracy in short-time traffic flow prediction. (3) The SA-SVR prediction model under the Spark platform is established. with the increase of the amount of traffic flow, in the process of processing mass traffic flow data, the prediction model in the stand-alone mode cannot satisfy the requirement of short-term traffic flow prediction to predict the real-time performance due to the limitation of physical factors, and in order to solve the problem of the prediction time, In this paper, in the background of the large data era, we study the Spark technology with distributed parallel processing ability to train the support vector regression machine in a large scale, The SA-SVR prediction model under the Spark platform is established by combining the advantages of the support vector regression machine to deal with the nonlinear small sample events and the short parallel processing time of Spark. Experimental results show that this model can shorten the forecast time on the premise of ensuring the prediction accuracy, and meet the requirement of short-time traffic flow prediction on the accuracy and real-time performance. In this paper, three groups of contrast experiments are carried out based on the prediction model, which are the RBF neural network and the support vector regression model, the mesh algorithm and the simulated annealing algorithm and the improved simulated annealing algorithm parameter optimization model, which is compared with the prediction model in the Spark parallel mode under the stand-alone mode. Compared with the traditional algorithm and the single-stand model, the model based on the improved simulated annealing algorithm is more competitive in the Spark environment than in the traditional algorithm and the stand-alone mode. The model not only solves the accuracy problem of short-time traffic flow prediction in the prediction process, but also solves the problem of prediction efficiency in short-time traffic flow prediction, and improves the capability of processing traffic flow data and the prediction accuracy and real-time performance in the short time traffic flow prediction. The main innovation point in this paper is to combine the sparse features of support vector regression machine with the parallel processing capability of distributed cluster Spark system, carry out large-scale SVR training under the Spark platform, and set up the SA-SVR short-time traffic flow prediction model under the Spark platform. The model well solved the accuracy and real-time problem of short-time traffic flow prediction.
【学位授予单位】：东华理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U491.14
【参考文献】
相关期刊论文 前10条
1 柳坐宫;钱勇生;曾俊伟;;不同道路特大交通事故特征分析及预防对策研究[J];中国公共安全(学术版);2015年03期
2 孙士平;吴建军;;直接搜索模拟退火算法的自适应改进[J];计算机工程与应用;2015年23期
3 金茂菁;;我国智能交通系统技术发展现状及展望[J];交通信息与安全;2012年05期
4 刘东;;智能交通控制网络的短时交通流预测研究[J];计算机仿真;2012年10期
5 向昌盛;周子英;夏艳军;;基于启发式广度优先搜索的SVR参数优化方法研究[J];生物信息学;2010年03期
6 杨玫;刘瑜;孔波;;SVR模型参数选择方法的研究[J];计算机时代;2009年11期
7 冀鹏浩;闫在在;;数量特征随机化回答模型中的比估计和回归估计方法[J];内蒙古工业大学学报(自然科学版);2009年03期
8 闫国华;朱永生;;支持向量机回归的参数选择方法[J];计算机工程;2009年14期
9 李一龙;;基于动态规划的城市交通流预测与分配[J];交通运输系统工程与信息;2009年03期
10 邢永忠;吴晓蓓;徐志良;;基于柯西分布加权的最小二乘支持向量机[J];控制与决策;2009年06期
相关博士学位论文 前4条
1 李国辉;粒子滤波及其在混沌信号降噪处理中的应用研究[D];西北工业大学;2015年
2 李s，
本文编号：2302780
本文链接：https://www.wllwen.com/kejilunwen/daoluqiaoliang/2302780.html
