基于最大偏差相似性准则的交通流聚类算法

针对常用聚类算法对随机性强、波动频繁的交通流聚类效果不理想的问题，提出了一种新的交通流相似性度量准则——最大偏差相似性准则，并提出了一种基于最大偏差相似性准则的交通流聚类算法。最大偏差相似性准则能够有效刻画频繁波动交通流曲线的形态相似性，具有简明、合理、灵活等特点；聚类算法无需预先指定类别数，能够保证类间曲线的明显差异性和类内曲线的高度相似性。实验表明，所提出的算法聚类效果明显优于常用聚类算法，聚类结果能够较好地满足实际应用的需要。

Traffic flow clustering algorithm based on maximum deviation similarity criterion

Focusing on the problem that the common clustering algorithms are not ideal for traffic flow with strong randomicity and frequent fluctuation, this paper proposed a new traffic flow similarity measurement which is called maximum deviation similarity criterion (MDSC), and proposed a traffic flow clustering algorithm based on the MDSC. The MDSC can effectively describe the curve shape similarity of frequent fluctuating traffic flow, which has the characteristics of simple, reasonable, flexible and so on. The proposed clustering algorithm does not need to specify the number of classes in advance, which can ensure that the curves of different classes have obvious differences and the curves in the same class have high similarity. The experiments show that the clustering effect of the proposed algorithm is significantly better than that of the common clustering algorithms, and the clustering result of the proposed algorithm can better meet the needs of practical applications.