Clustering Multivariate Time Series from Large Sensor Networks

In recent years, as the sensor technology develops swiftly, a large number of sensors can be deployed to form a sensor network. Massive data are generated from large sensor networks. It becomes an interesting problem to cluster high-dimensional multivariate time series from large sensor networks to discover the hidden regularity. In this paper, we propose a new method to cluster the high-dimensional time series. In our scheme, the output of the sensors are treated as gray values of images. Then, new image features (BSF, Bipolar Sigmoid Feature) and the chain similarity are presented to measure the similarities of time series. Finally, the hierarchical clustering method is used to discover data patterns. The effectiveness of our method is evaluated on the Twin Cities traffic data. In the experimental results, it displays that there are three regular patterns in the weekday data and two patterns in the weekend day.