Deep clustering model for time-series data based on recurrence plot and variational auto-encoder

Time-series data are trendy and periodic, and have high data dimensionality, the current clustering methods cannot effectively target these characteristics. A recurrence plot variational auto-encoder deep clustering (RPVAEDC) model based on recurrence plot and variational auto-encoder is proposed to address the characteristics of time-series data. The time-series data are first transformed into recurrence plots to reveal their trends and periodicity; then the recurrence plots are fed into a deep clustering model for feature extraction and dimensionality reduction, and the distribution of the transformed data is normalized by variational auto-encoder; then the clustering results are obtained by adding a clustering layer to combine the auto-encoder reconstruction loss and clustering loss. It is experimentally demonstrated that the silhouette coefficient scores are achieved significantly better than other clustering algorithms on the public data sets.