Traffic Flow Time Series Clustering Based On Feature Analysis

As an important data mining technique,clustering algorithm can extract the data pattern distribution from massive data,including the division of the time series of traffic flow,mining the deep information in the time series,and discovering the knowledge.Under normal circumstances,the traffic flow time series has the characteristics of periodicity,large data volume,high noise,morning and evening peak,which makes the general clustering algorithm can not be uesd to the traffic flow time series clustering efficiently.Therefore,according to the traffic flow time series feature analysis,this paper proposes a neural network clustering algorithm based on combinatorial similarity measure,to solve the traffic flow time series clustering problem.Specific research includes:1.Traffic flow time series characteristics analysis.The EP coefficient(Euclidean metric-Pearson correlation coefficient)is proposed as the similarity measure of the traffic flow time series based on feature analysis,this coefficient consists of Euclidean distance and Pearson correlation coefficient.It is used to measure the time series of traffic flow Characteristics and morphological characteristics;2.Traffic flow time series clustering comparative study and model establishment.Based on the comparison of k-mean,PAM,hierarchical clustering and SOM neural network clustering,in this paper,the SOM neural network is used as the clustering algorithm of time series data of traffic flow combined with the characteristics of traffic flow time series data.3.Traffic flow time series clustering case analysis.The SOM neural network based on EP coefficient is used as a clustering algorithm to analyze the time series data of the real traffic flow.The different weight of Euclidean distance and Pearson correlation coefficient in the EP coefficient are compared to show how it is effect the clustering results.The clustering results show that the clustering algorithm proposed in this paper can effectively identify the statistical characteristics and morphological characteristics of traffic flow time series.The main innovation of this paper is proposed a clustering model suitable for traffic flow time series.This model can predict the trend of future traffic flow time series according to the current traffic flow.On the one hand,it can make the management department determine the management mode in advance and improve the utilization efficiency of management resources.On the other hand,the response can be faster because the unexpected situation on the road can be found by comparing the predicted results with the real-time monitoring results.