基于马氏距离和Canopy改进K-means的交通聚类算法

Traffic Clustering Algorithm Based on Markov Distance and Canopy Improved K-means

在对交通数据的研究中经常会使用到聚类算法,且不同的聚类算法有不同的特性。K-means作为其中的一种聚类算法,具有较高的准确性和实用性,但其准确性易受主观选取K值和确定初始聚类中心的影响。为了优化聚类中心和K值的选取问题,提出MC-Kmeans算法。在所提方法中,首先通过Canopy算法选取K值,然后依据马氏距离的计算准则来确定初始聚类中心,最后将K值和聚类中心的值作为K-means的参数进行聚类。将MC-Kmeans算法应用到某时间段的纽约出租车交通数据中进行实际的验证。结果表明,与K-means算法比较,所提方法准确度更高,与实际交通情况更加相匹配,更能反映区域内的交通热点情况。

Clustering algorithms are often used in the research of traffic data,and different clustering algorithms have differ-ent characteristics.As one of the clustering algorithms,K-means has high accuracy and practicability,but its accuracy is easily af-fected by subjective selection of K value and determination of initial clustering center.In order to optimize the selection of clustering center and K value,MC-Kmeans algorithm is proposed In the proposed method,firstly,the K value is selected by canopy algo-rithm,and then the initial cluster center is determined according to the calculation criterion of Mahalanobis distance.Finally,the K value and the value of cluster center are clustered as the parameters of K-means MC-Kmeans algorithm is applied to New York taxi traffic data in a certain period of time for practical verification.The results show that compared with K-means algorithm,the pro-posed method has higher accuracy,better matches the actual traffic situation,and can better reflect the traffic hot spots in the re-gion.