Method for time sequence analysis based on tensor time-domain correlation decomposition model

Abstract

A method for time sequence analysis based on a tensor time-domain correlation decomposition model comprises the steps that time sequences are encoded into tensors; the initial tensor is converted into a dimension decrease form; a self-regression model is applied to the obtained tensors with the dimension decrease form, so that time-domain continuity can be maintained and course update results are learned dynamically till algorithm convergence is achieved, and result optimization is achieved. The method for the time sequence analysis based on the tensor time-domain correlation decomposition model provided by the invention is characterized in that the time-domain correlation model is introduced to a tensor decomposition framework, so a space-domain dimension and a time-domain dimension of the time sequences can be processed at the same time in an anisotropic manner; noise and redundant information in the time sequence space domain can be removed; internal relations among time sequence time-domain information can be controlled; and the time-domain continuity of the time sequences is maintained. According to the invention, accuracy for time sequence prediction is enhanced especially for a high-dimensional time sequence prediction issue.

Description

Technical field [0001] The invention relates to a time series analysis method. In particular, it relates to a time series analysis method based on a tensor time-domain correlation decomposition model that combines tensor decomposition technology and a time-domain correlation model to predict time series. Background technique [0002] Accessible data has been increasing exponentially in quantity, speed and variety in recent years. This growth trend is facing various challenges in the scientific community. Among the accessible data is an important one. The data representation is a time series. A time series is a series of data points composed of a set of continuous measurement values ​​over a period of time. In the past ten years, mining time series has become a very promising research field. The analysis tasks of time series mainly include the following: prediction, monitoring, feedback control, anomaly detection, clustering, classification and segmentation; covering environmen...

AI Technical Summary

Problems solved by technology

However, the most common method of migrating tensor time series data into vectors or matrices is not only prone to the disaster of dimensionality, but also destroys the data structure and causes the loss of relationship information between adjacent data.

Images