智能停车场泊位预测及诱导停车方法研究

发布时间：2018-02-21 04:47

本文关键词： 泊位预测诱导停车神经网络多属性决策路径寻优　出处：《哈尔滨商业大学》2015年硕士论文　论文类型：学位论文

【摘要】：随着城市居民私家车的拥有量急剧上升,停车难的问题日益突出。停车诱导系统主要解决停车场车位与驾驶员驾驶的车辆之间的供求关系平衡,但是现有的停车诱导技术主要注重停车场外部的区域引导,对于停车场内部的引导机制单一,一般仅仅随机给出一条从停车场入口处到有效空余泊位的最短路径,并没有从驾驶员所考虑的停车因素角度来进行诱导。此外对于停车场外部的引导,如果加入停车场的空余泊位预测技术,则会有效的帮助场外诱导系统。本文首先分析了目前主要的时间序列预测技术,在此基础之上,提出了利用BP神经网络算法来进行停车场空余泊位的预测。对标准的BP神经网络的原理进行了详细的研究,针对停车场空余泊位预测的问题进行BP算法的网络结构确定,并且针对BP神经网络训练过程中易震荡、收敛速度过慢和容易陷入局部最小的缺点,采用BP动量法与调节学习速率相结合的方法对其进行改进。在进行实验数据仿真时,通过数据预处理技术对停车场空余泊位数据进行了变换,防止了因数据过大而造成的网络瘫痪。通过仿真验证了BP神经网络对停车场空余泊位数预测的有效性,为停车场场外诱导提供了帮助。对于停车场场内部分,本文利用多属性决策的方法进行路径寻优来实现场内诱导。在分析驾驶员选择车位的主要考虑因素的基础上,利用Dijkstra算法确定行驶距离与路径、利用欧几里得距离确定步行距离以及利用三角模糊数期望值的方法确定停车位的环境信息值,来确定决策属性矩阵,最终利用基于灰熵关联度多属性决策的方法进行了停车场有效空余泊位的属性排序。其排序中最优的属性所对应的空余泊位即是最优泊位,所对应的行驶路径即为最优路径。
[Abstract]:With the rapid increase in the number of private cars owned by urban residents, the problem of parking difficulties is becoming increasingly prominent. The parking guidance system mainly solves the balance of supply and demand between parking spaces and vehicles driven by drivers. However, the existing parking guidance technology mainly pays attention to the regional guidance outside the parking lot. For the single guiding mechanism within the parking lot, the shortest path from the entrance of the parking lot to the effective free berth is generally given at random. In addition, for the outside guidance of the parking lot, if the free berth prediction technology is added, it will help the off-site guidance system effectively. Based on the analysis of the main time series prediction techniques, a BP neural network algorithm is proposed to predict the parking space. The principle of the standard BP neural network is studied in detail. The network structure of BP algorithm is determined for the prediction of parking space, and the shortcomings of BP neural network are that it is easy to concussion, converge too slowly and fall into local minimum easily in the training process of BP neural network. The BP momentum method and the method of adjusting the learning rate are adopted to improve it. In the simulation of the experimental data, the data of the parking lot free berth are transformed by the data preprocessing technology. The simulation results show that the BP neural network is effective in predicting the number of parking spaces, which is helpful for the off-site induction of parking lots. For the part of parking yard, this paper uses the method of multi-attribute decision to optimize the route to realize the induction in the field. On the basis of analyzing the main factors of driver's choice of parking space, the Dijkstra algorithm is used to determine the driving distance and path. The Euclidean distance is used to determine the walking distance and the triangular fuzzy number expectation value is used to determine the environmental information value of the parking space to determine the decision attribute matrix. Finally, the method of grey entropy correlation multi-attribute decision-making is used to sort the attributes of the parking lot's effective free berth. The optimal berth corresponding to the optimal attribute is the optimal berth, and the corresponding driving path is the optimal path.
【学位授予单位】：哈尔滨商业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U491.7;U495

【引证文献】