基于遗传小波神经网络的停车泊位预测方法  

Parking space prediction method based on genetic wavelet neural network

作　　者：韩锟[1] 李斯宇 HAN Kun;LI Siyu(School of Traffic and Transportation Engineering,Central South University,Changsha 410075,China)

机构地区：[1]中南大学交通运输工程学院,湖南长沙410075

出　　处：《铁道科学与工程学报》2020年第9期2216-2224,共9页Journal of Railway Science and Engineering

基　　金：国家级大学生创新创业训练计划资助项目(201610533302)。

摘　　要：在车位共享模式下,对社区停车位进行短时、准确地预测,既有利于停车需求方选择更合适的车位,也有利于车位资源的合理分配。提出一种社区停车位的短时预测方法,首先利用C-C方法对空余停车泊位时间序列进行相空间重构,采用李雅普诺夫指数法证明该时间序列的可预测性;然后将重构后的时间序列输入小波神经网络(wavelet neural network,WNN)进行训练。采用遗传算法(genetic algorithm,GA)优化小波神经网络初始参数,经过多次迭代获取最优参数,得到优化后的预测模型;最后,通过matlab对该算法进行编程,并调研长沙市某社区空余停车泊位数据进行实验分析。研究结果表明:基于相空间重构的遗传小波神经网络(CC-GA-WNN)模型具有较好的预测精度和优化效果。For parking space sharing mode,short-term and accurate prediction of the community parking space is beneficial for drivers to select a more suitable parking space,and is also conducive to the reasonable allocation of parking space resources.In this paper,a short-term prediction method for community parking spaces was proposed.Firstly,the phase space reconstruction of the free parking space time series is carried out by CC method and the predictability of the time series was proved by Lyapunov exponent method.Then the reconstructed time series was input into the wavelet neural network for training.The genetic algorithm was used to optimize the initial parameters of the wavelet neural network,and the optimum parameters were obtained after several iterations to establish the optimized prediction model.Finally,the algorithm is programmed by matlab,and the data of free parking spaces in a community in Changsha was investigated for experimental analysis.The results show that the CC-GA-WNN model has better prediction accuracy and optimization effect.

关 键 词：空余停车泊位 相空间重构 遗传算法 小波神经网络 

分 类 号：U491.7[交通运输工程—交通运输规划与管理]