摘要
针对短时交通量的非线性和时变性,提出一种基于粒子群—小波神经网络的预测方法。该方法以前馈多层感知器的神经网络拓扑结构为基础,将预测误差反向传播,经粒子群优化算法对神经网络连接权值进行修正。隐含层神经元选择Morlet母小波基函数作为激活函数,利用小波分解分离短时交通量的高频部分和低频部分,防止高低频数据之间的过度影响,进一步提高预测的精度。根据最简化结构概念对神经网络结构进行泛化,确定最优网络结构,提高预测的速度。通过实例预测显示,该方法预测精度高,预测速度快,能够满足实际工程的要求。
For nonlinearity and time variance of short-term traffic volume, a kind of forecasting method named PSO（particle swam optimization）-wavelet neural network is proposed. The method is based on topological structure of multilayer feed-forward perceptrons （MLPs）, and the output error is back propagated to PSO module to revise connection weights of network. In the hide-layer, Morlet mother wavelet function is selected to be neuron active function which separate high-frequence and low-frequence of traffic volume data to avoid the effect between each other. The network generalizes by minimization feature structure（MFS） concept to a optimal structure to improve forecasting speed. The forecasting example shows, the accuracy and speed of the method proposed are suitable for practical engineering.
出处
《黑龙江工程学院学报》
CAS
2014年第2期42-45,共4页
Journal of Heilongjiang Institute of Technology
基金
国家自然科学基金资助项目(61304210)
福建工程学院科研发展基金资助项目(GY-211057)
福建省教育厅A类基金资助项目(JA11192)
关键词
短时交通量
泛化
小波
神经网络
粒子群优化算法
short-term traffic volume
generalize
wavelet
neural network
particle swam optimization
