【作者】 周正林；

【导师】 张昕；

【作者基本信息】 哈尔滨工程大学 ， 通信与信息系统， 2007， 硕士

【摘要】 随着经济的发展和城市规模的扩大，城市交通拥挤问题日益突出。目前，主要通过以下两方面的途径缓解交通拥挤：一是增加路网建设的密度，这是一种最为有效的办法。二是通过对交通流量的合理调控，优化路网的利用率。这就是世界发达国家正在研究、实施的智能运输交通系统(ITS)。而准确地预测交通流量是ITS的关键所在。本文以单十字路口交通流量预测为例，设计了三种预测方案。于2006年8月8日、10日6时～8时40分，实测了哈尔滨市红旗大街和先锋路交叉口各个方向的实际交通流量。并以所测得交通流量作为样本，对本文的BP网络预测模型进行训练和仿真，选出较好的方案作为实际预测模型。由于BP网络收敛速度慢且易陷入局部最小点，为提高模型的预测性能，又提出了通过调整学习率的改进BP算法，本算法可以根据预测精度来动态调整学习率，这是本文的一个创新之处。考虑到学习率和隐含层数目对网络预测精度和收敛速度有很大的影响。本文又利用遗传算法对改进的BP预测模型进行寻优，这也是本文最大的创新之处。本算法通过改变网络的学习率和隐含层单元数目，利用仿真结果来优化预测模型的总体性能。研究表明，利用遗传算法搜索到的预测模型具有预测精度高、收敛时间短的优点，可以作为实际的预测模型来使用。更多还原

【Abstract】 With the economical development and the expansion of the city scale, the problem of the transportation congestion is outstanding day by day. At present, two ways are used to effectively alleviate the traffic congestion. Increasing the road network’s density is one of the most effective means. The other is optimize the utility of the road network through the reasonable regulation of traffic flow, which is studying and applying intelligent transportation system (ITS) .the key of it is to forecast the traffic flow.This article has designed three kinds of forecast projects and take the single intersection road traffic flow forecast as an example, In the August 8th and 10th from 6:00 a.m. to 8 :40 2006, We measured the actual traffic flow of the Road intersection between Red flag street and XianFeng Road by all directions. The BP network model was trained and simulated. We have selected a better solution as a practical model by the swatch measured.Because the speed of BP network’s convergence is too slow and enters local minimum point easily, the article provides advanced BP algorithm through adjusting study rate to improve the performance of the forecast model. The algorithm may adjust the study rate according to the forecast precision. It is the innovation.Since the study rate and the conceal level numbers have great influence on the forecast precision of network. This article has researched the most superior BP neural network forecast model using the genetic algorithm. This is anther innovation.This algorithm changing the BP network the study rate and the concealment level unit number, optimizes overall performances of forecast model using the simulation result. The study indicates, the forecast model which using the genetic algorithm has the merits of high forecast precision and short restraining time , and can be applied as the actual forecast model.更多还原