A Traffic Speed Imputation Method Based on Self-adaption and Clustering

Traffic speed estimation plays an important role in Intelligent Transportation System (ITS) since current speed collections cannot cover all roads and may lose data. A tensor completion method called high accurate low rank tensor completion (HaLRTC), one of the research methods on speed imputation, has great performances in this field. In this paper, we propose a method based on HaLRTC, and analyze how the parameters affect the method. Then we can predetermine these parameters with incomplete data. To improve the efficiency, we use variances of one road segment in all days as its feature. We do k-means++ clustering on road segments with their variance feature, as for each cluster we apply HaLRTC on it independently. Then we merge these completion tensors into a whole tensor. The proposed method HaLRTC-CSP applies HaLRTC and is based on clustering and self-adaption parameters. Experiments is implemented on the speed dataset in Guangzhou, China, and the results validate HaLRTC-CSP outperforms other methods.