Design of Sequential Data Clustering Algorithm Based on Time Reconstruction and K-means

In data mining, effectively representing the temporal dynamics, multi-scale features, and clustering characteristics of sequences and improving the encoder's temporal serial clustering ability in the absence of labels is a significant challenge. This article adopts a time series clustering algorithm based on deep learning, which integrates time reconstruction and K-means into the seq2seq model to improve cluster structure and obtain cluster specific temporal representations. Meanwhile, the use of pseudo sample generation strategy and auxiliary classification tasks has improved the performance of the encoder. In order to verify the effectiveness of this method, it was compared with various clustering methods, and the experimental results showed that this method can achieve the expected goals and the effect is good.