Based on Multiple Time Series Affinity Propagation Algorithm

摘要

In multivariate time series due to its high dimension, high signal-to-noise ratio and different length sequence of factors leading to poor clustering results, so the article by improving a semi-supervised clustering affinity propagation algorithm tries to solve the problem. The improved algorithm AP_DTW is obtained by using dynamic time warp distance measurement and the affinity propagation clustering of component properties. Then, two experiments are carried out on this algorithm. Experiment 1 is to compare it with the traditional distance measurement method, and experiment 2 is to compare it with the traditional clustering algorithm through ten numerical experiments. A method to verify the validity of clustering was used to evaluate the clustering results in high dimensional time series and the final experimental results were obtained.