Time series data clustering method based on noise reduction encoder and attention mechanism

Abstract

The invention provides a time series data clustering method based on a noise reduction encoder and an attention mechanism, which achieves clustering of label-free data in an unsupervised manner, and sequentially comprises the following steps of: 1, adding noise data into an original time sequence; step 2, adding an attention mechanism into the LSTM; 3, initializing a model by adopting a K-means method; 4, jointly optimizing the reconstruction loss and the KL divergence loss of the noise reduction auto-encoder by adopting an Adam optimization method; and iteratively optimizing the reconstruction loss and the clustering loss of the encoder to obtain a trained optimal model. Based on the deep noise reduction auto-encoder, the problem that time series data clustering is sensitive to noise is solved, and the clustering effect is improved by introducing an attention mechanism.

Description

technical field [0001] The invention belongs to the technical field of data mining, and relates to a time series clustering method, in particular to a time series data clustering technology based on a noise reduction encoder and an attention mechanism. Background technique [0002] Time series clustering research is an important research content in the field of data mining and machine learning. It is a research that digs out hidden patterns in time series data and divides time series into different categories. However, time series are usually high-dimensional, so the clustering research of time series is full of challenges. In recent years, deep learning has promoted the development of natural language processing, image processing and other fields. It can extract low-dimensional feature vectors from high-dimensional data, and this low-dimensional feature representation is needed in time series research. However, most of the existing time series clustering methods based on d...

AI Technical Summary

Problems solved by technology

However, most of the existing time series clustering methods based on deep learning are implemented based on the deep autoencoder framework, which is sensitive to noisy data

Moreover, the existing temporal clustering algorithms have the problem of distraction, because the weights of features at different moments are ignored in the feature extraction process.

In short, the existing time series clustering methods cannot extract robust and discriminative time series features well, which greatly affects the subsequent clustering accuracy

Images