摘要
【目的】通过改进停车泊位预测方法为交通出行提供有益帮助。【方法】利用李雅普指数对停车泊位序列进行分析,指出序列具有混沌特性,可进行多步预测。利用db32小波函数具有双正交性、紧支撑性以及消失矩阶数较大的特征,对归一化的停车泊位序列进行多尺度分解与重构,并作为小波神经网络(Wawelet neural network,WNN)的隐含层函数。为提高预测精度和降低预测时间,分别使用粒子群算法(Particle swarm optimization,PSO)和极限学习机(Extreme learning machine,ELM)来优化WNN。其中,使用PSO对WNN的权值进行调整,逐步迭代得到最优值;使用ELM将全局最优值作为单隐层前馈神经网络的输入,使得算法尽快收敛。优化后的WNN结合迭代多输出法对停车泊位进行预测。将上述预测方案称为极限学习机和粒子群算法双重优化的小波补缀网络多步预测(Multi-step prediction based on wavelet neural networkimproved by extreme learning machine and particale swarm optimization,MP-EPWNN)。【结果】仿真实验表明,相对于BP神经网络、遗传算法优化小波神经网络、极限学习机优化小波神经网络、粒子群优化小波神经网络4种算法,MPEPWNN算法的预测均方误差平均降低了96.6%,预测所需的时间平均降低了65.97%。【结论】MP-EPWNN算法预测停车泊位是有效的。
[Purposes]Parking forecasting can provide useful help for traffic trips.[Methods]The Li Yapu index is used to analyze the parking berth sequence,and it is pointed out that the sequence has chaotic characteristics and can be used for multi-step prediction.The"db32"wavelet function is characterized by orthogonality,compactness and large order of vanishing moments,the normalized parking berth sequence is decomposed and reconstructed with multi scales by the"db32"wavelet function,and it is used as the hidden layer function of the wavelet neural network(WNN).In order to improve prediction accuracy and reduce prediction time,particle swarm optimization(PSO)and extreme learning machine(ELM)are used to optimize WNN respectively.Where,through PSO,the weights of WNN are adjusted and the optimal values are obtained iteratively;the global optimum is used as the input of a single hidden layer feedforward neural network by using ELM,so that the algorithm could converge as soon as possible.Optimized WNN,combined with iterative multi-output method,is used to predict parking berth.The above prediction scheme can be abbreviated that multi-step prediction based on wavelet neural network improved by particle swarm optimization and extreme learning machine(MPEPWNN).[Findings]Simulation results show that,compared with the BP neural network,genetic algorithm optimization wavelet neural network,extreme learning machine optimization wavelet neural network,particle swarm optimization wavelet neural network,the mean square error of the MP-EPWNN algorithm is reduced by 96.6% on average,the average time required for prediction is reduced by 65.97%.[Conclusions]The MP-EPWNN algorithm is effective in predicting parking berth.
作者
李田田
杨有
余平
陈艳平
LI Tiantian;YANG You;YU Ping;CHEN Yanping(School of Computer and Information Science,Chongqing Normal University,Chongqing 401331;Department of Computer Engineering,Chongqing Aerospace Polytechnic College,Chongqing 400021,China)
出处
《重庆师范大学学报(自然科学版)》
CAS
CSCD
北大核心
2018年第5期107-114,共8页
Journal of Chongqing Normal University:Natural Science
基金
重庆师范大学研究生科研创新项目(No.YKC18027)
重庆市2015年高等学校教学改革研究(No.152017)
重庆市研究生教育教学改革研究项目(No.YJG20163009)
重庆市教委科学技术研究项目(No.KJ1400512
No.KJ1602801)
关键词
停车泊位
多步预测
小波神经网络
粒子群算法
极限学习机
parking berth
multi-step prediction
WNN(wavelet neural network)
PSO(particle swarm optimization)
ELM(extreme learning machine)