基于Hadoop平台的智能交通流预测及路径诱导算法研究

发布时间：2018-04-10 23:34

本文选题：Hadoop + BP神经网络算法　；参考：《兰州交通大学》2017年硕士论文

【摘要】：随着交通系统复杂性程度日益提高,尽管相关部门在各种交通设施建设方面投入了大量的资金和资源,但其仍然不能满足人们的出行要求。面对如此复杂的交通系统,为了提高智能交通系统的搜索效率,缩减其搜索范围,在更短的时间内反馈路网信息,缩短用户在出行过程中“无谓的等待时间”,对短时交通流预测和路径诱导算法的研究是很有必要的。然而要解决上述问题,最重要的是提高短时交通流预测以及路径诱导算法的效率。从短时交通流预测算法、路径诱导算法的角度来看,交通管理的成效,和预测、诱导精度以及算法效率直接相关。但是,一般来说,精度和效率之间呈负相关关系,算法的精度越高,代表其逻辑太过复杂,或者是计算工作量大,因此计算耗时更长,效率更低,严重时彻底失去实用性。在短时交通流预测方面,论文中对常用短时交通流预测算法进行了分析对比,并指出各种算法的优缺点及使用范围,因BP神经网络算法构建的数学模型具有十分严谨的特点,同时具有自主学习能力、良好的容错能力以及良好的泛化性,所以选取BP神经网络算法对短时交通流进行预测研究。但是BP神经网络算法因采用静态梯度下降法来优化网络权值和阈值,使其BP神经网络算法存在一定的局限性,如稳定较差,收敛速率缓慢,容易达到局部极小值等缺陷。为了克服上述缺陷,论文在短时交通流预测中采用改进后的遗传算法来优BP神经网络预测模型。遗传算法作为一种全局范围的搜索算法,通过模拟遗传过程中遗传因子复制、交叉和变异的特性,对个体不断进行择优,将最终得到的最优解作为神经网络算法的初始值。但是交通流数据的复杂多样性使得遗传算法在搜索的过程中可能存在最优解丢失的情况,从而导致算法过早收敛,反而降低了短时交通流预测的准确性。为了克服以上缺陷,在遗传算法中引入跟短时交通流运动极其匹配的混沌现象,组成混沌遗传算法(CGA)。其核心思想主要是在待优化变量中引入混沌状态,并把混沌运动的遍历范围“扩展”至待优化变量的取值范围中,进行全局细化搜索,这样就能避免过早陷入局部最优解,最终通过不断优化得到最优解。然后用得到的最优解初始化BP神经网络的初始权值和阈值,从而提高对短时交通流预测的实效性和准确度。论文并验证了改进后的算法其性能比之前有明显提高。在路径诱导方面,论文中也是对常用路径诱导算法进行了研究对比,并分析出各自算法的优缺点和使用范围,由于蚁群算法具有智能化搜索,能够达到全局优化的目的,在鲁棒性、自组织性、并行性方面表现十分突出,并且适合复杂的非线性交通系统中,所以采用蚁群算法对路径诱导进行研究。当然,任何算法都会有自身的局限性和不足,论文针对蚁群算法上存在的缺陷分别对蚁群算法的状态转移规则和信息素更新规则进行了改进。从而减少出行用户对无效路径的搜索,并且能从综合因素中选择最优路径。在本课题的研究中,在满足短时交通流预测和路径诱导算法实用性要求的前提下,充分发挥云计算平台在数据保存和并行处理方面的优势作用,结合Hadoop平台,对改进后的BP神经网络算法和蚁群算法进行了MapReduce的设计和实现,成功地设计出新的短时交通流预测以及路径诱导方法,在预测、诱导的精度和效率之间找到良好的平衡点,大大强化了两种算法在实用性方面的表现,并且在实验中验证了算法的性能和实用性。
[Abstract]:With the increasing complexity of traffic system, although the relevant departments invested a lot of money and resources in various transportation facilities, but it still can not meet people's travel requirements. Facing such complicated traffic system, in order to improve the searching efficiency of the intelligent transportation system, reduce the search scope and network feedback information in a shorter period of time in the course of travel, reduce the user in "waiting time", research on short term traffic flow forecasting and route guidance algorithm is very necessary. However, in order to solve the above problems, the most important is to improve the short-term traffic flow prediction and route guidance algorithm. From the short-time traffic flow prediction algorithm, routing algorithm the point of view, the effectiveness of traffic management, and prediction, accuracy and efficiency of the algorithm by directly related. However, in general, showed a negative correlation between accuracy and efficiency The relationship, the higher accuracy of the algorithms, on behalf of the logic is too complex, or the calculation workload, so the computation time is longer, the efficiency is lower, which completely lose the practicability. In prediction of short-term traffic flow, the paper used the short-time traffic flow prediction algorithm are analyzed, and points out the advantages and disadvantages of various algorithms and the scope of use, due to the construction of mathematical model of BP neural network algorithm has the characteristics of very strict, and has self-learning ability, good tolerance and good generalization, so the selection of the BP neural network algorithm to forecast the short-term traffic flow. But the BP neural network algorithm by using static gradient to optimize the network weights and threshold descent, the BP neural network algorithm has some limitations, such as poor stability, slow convergence rate, easy to reach a local minimum in order to overcome the above defects. The defects of short-term traffic flow prediction using improved genetic algorithm to optimize BP neural network prediction model. The genetic algorithm is a global search algorithm, through the simulation of genetic factor in genetic process, characteristics of crossover and mutation, the ongoing individual merit, finally obtained the optimal solution as the initial nerve the value of the network. But the complexity and diversity of traffic flow data make the genetic algorithm possible loss of optimal solution in the search process, which leads to premature convergence of the algorithm, it reduces the accuracy of short-term traffic flow forecasting. In order to overcome the above defects, introducing chaos phenomena with short-term traffic flow, genetic algorithm in the extreme in the form of chaos genetic algorithm (CGA). The core idea is to be optimized in chaotic state variables, and the chaotic motion of the traverse range expanding The range of variables to be optimized to show "in the global refining search, so you can avoid falling into local optimal solution, finally through continuous optimization to obtain the optimal solution. And then get the optimal solution to initialize BP neural network initial weights and thresholds, so as to improve the effectiveness of short time traffic flow prediction and accuracy. The results show that the improved algorithm has significantly improved its performance than before. In the aspect of the induced path, but also on the common path guidance algorithm is studied, and the analysis of the advantages and disadvantages of each algorithm and the range of use, due to the ant colony algorithm with intelligent search, can achieve the purpose of global optimization, robustness, self-organization, parallel performance is very prominent, and is suitable for nonlinear complex traffic system, so the ant colony algorithm of route guidance. Of course, any algorithm All have their own limitations and shortcomings, aiming at the shortcomings of ant colony algorithm on ant colony algorithm respectively to the state transition rule and pheromone updating rule is improved. So as to reduce travel users of invalid path search, and can choose the optimal path from the comprehensive factors. In this study, in order to meet the short term traffic flow prediction and route guidance algorithm practical requirements under the premise, give full play to the cloud computing platform in data storage and parallel processing advantage role, with the platform of Hadoop, the improved BP neural network algorithm and ant colony algorithm for the design and implementation of MapReduce, successfully designed a new short-term traffic flow forecasting and route guidance method, in the forecast, find good balance between accuracy and efficiency of induction, greatly enhanced the performance of the two algorithms in the practical aspects, and in reality In the experiment to verify the performance and practicability of the algorithm.

【学位授予单位】：兰州交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U495

【参考文献】