基于K-均值聚类算法RBF神经网络交通流预测

针对目前道路拥堵等交通问题,本文采用K-均值聚类算法对径向基函数(radial basis function,RBF)网络进行优化,通过K-均值聚类算法把所有的输入样本进行统一聚类,求得所有隐含层节点的RBF中心值Ci,并用最小二乘法(LMS)进行RBF网络的权值调整,同时在一定的时间和路段内对车流量进行数据采集,通过建立RBF神经网络模型,运用Matlab软件把采集的数据、图像进行计算机仿真,仿真结果表明,未加入K-均值聚类的RBF神经网络,其预测输出曲线大致可以和实际输出曲线拟合,但在数据波动较大的时刻,预测曲线的收敛速度偏慢且效率偏低;而采用K-均值聚类算法的RBF神经网络,在实际输出波动较大时,预测输出的曲线收敛速度和准确度都较高,因此,本研究相对于普通的BP神经网络,有更高的预测精度和较好的收敛性。该研究适用于市区内的交通流预测。

Traffic Flow Prediction Based on K-Means Clustering Algorithm and RBF Neural Network

In view of the current traffic congestion and other road traffic problems, this paper optimizes the RBF neural network by using the K-means clustering algorithm. It uses K-means clustering algorithm to unifyall of the input sample clustering, it obtains the RBF center value Ci of all hidden layer nodes and ad- justs the weights of RBF network with the least squares method. At the same time within a certain amount of time and road for traffic data collection, through the establishment of RBF neural network model, it simulates the data and images collected by MATLAB. Simulation results show that without the K-means clustering, the prediction of the output curve can roughly fit the actual output, but at the moment of vola- tile data, the prediction curve is of slow convergence speed and the efficiency is low. With the K-means clustering algorithm of RBF neural network, when the actual outputs are volatile, the forecast of the curve of the outputs has higher convergence speed and accuracy. Therefore, this paper puts forward the RBF net- work based on K-means clustering algorithm. Compared with the normal neural network it has a higher prediction precision and better convergence. The study is applicable for traffic flow prediction in urban area.