K-Shape clustering method based on time series data LSTM features

Abstract

The invention discloses a time series data LSTM feature-based K-Shape clustering method, and belongs to the technical field of data mining, and the method comprises the following steps: S1, collecting and preprocessing a time series data sample; s2, establishing and training a long and short-term memory model, and outputting time sequence dynamic characteristic data; s3, calculating an optimal clustering K value of clustering by using an elbow method and a contour method; and S4, establishing a K-Shape clustering model and outputting a result. According to the method, the dynamic characteristics of the time series data are obtained by using the LSTM model, and K-Shape clustering is performed, so that the problem that the clustering result of general clustering such as K-mean clustering on complex time series data is not clear can be better solved, and the change trend clustering result of the time series data is obtained; and meanwhile, the LSTM is utilized to extract the time sequence characteristics of the data before K-Shape clustering, so that the accuracy, the robustness and the universality of a clustering result are additionally improved.

Description

technical field [0001] The invention relates to the technical field of data mining, in particular to a K-Shape clustering method based on LSTM features of time series data. Background technique [0002] With the introduction and development of the concept of the Industrial Internet, a large amount of time series data will be generated in daily work and life. This time-series data is often unlabeled, making it difficult to directly utilize or extract information. Therefore, how to extract effective information from massive time series data has become a very important topic. [0003] Clustering is an important part of unsupervised learning, which aims to solve the classification problem of unlabeled data. Generally, the clustering algorithm divides the data set into several classes or clusters that are close to each other but different, so as to obtain the statistical information of the data. . Because time series data are mostly unlabeled data, the clustering method can be...

AI Technical Summary

Problems solved by technology

K-means clustering is a clustering method based on Euclidean Distance as a distance function, which can reflect the positional relationship between data corresponding to the same time point, but it is easily disturbed by numerical outliers and cannot be used Between time series data of different lengths, it is even more difficult to reflect the dynamic change characteristics of time series data such as peaks, troughs, and periodicity

[0005] Some time series data have strong trend similarities, but because time series data has the complexity of time and space, it is difficult to extract and describe the physical meaning of its dynamic characteristics from intuitive statistics, and traditional clustering cannot effectively deal with it. This similarity, and outliers in time series data without feature extraction are more likely to interfere with clustering results

Images