Time sequence feature extraction method and system based on confidence interval

Abstract

The invention discloses a time sequence feature extraction method and system based on a confidence interval. The method comprises the steps that the value range of the number of segments of historicaltime sequence data is determined; a data segment segmentation step: determining the number K of segments, and segmenting the historical time sequence data into K continuous non-overlapping data segments; a weight mean value calculation step: calculating the area of the intersection of the parallelogram confidence space of each divided data segment and the discrete signal convex hull, and calculating the weight of the area of each intersection in the area of the parallelogram and the mean value of the weights; adding 1 to the value of the piecewise number K, and repeating the data segment segmentation step and the weight mean value calculation step to obtain the mean value of the weights under different piecewise numbers; performing ending until the number K of segments is greater than themaximum number of segments; selecting the number of segments corresponding to the maximum value of the weight mean value as the optimal number of segments; and segmenting the historical time sequencedata by using the optimal segmentation number to obtain a historical time sequence data feature extraction result.

Description

technical field [0001] The present disclosure relates to a time series feature extraction method and system based on a confidence interval. Background technique [0002] The statements in this section merely mention background art related to the present disclosure and do not necessarily constitute prior art. [0003] In the process of realizing the present disclosure, the inventors found that the following technical problems existed in the prior art: [0004] The safe and efficient operation of modern industrial processes usually requires the monitoring, analysis and control of the time series of production process variables. These time series data are large in volume, high in dimension, and complex in structure, so it is very difficult to perform data mining directly on the original data. Therefore, extracting the main features of time series through piecewise linear representation (PLR: piece-wise linear representation), transforming time series from high-dimensional to ...

AI Technical Summary

Problems solved by technology

[0013] In order to solve the deficiencies of the prior art, the disclosure provides a time series feature extraction method and system based on a confidence interval. The disclosure does not need to calculate the corner difference and is not based on a loss function. The number of segments is determined through two balance effects, so that the confidence interval Include as many data points as possible while minimizing the approximation error, overcoming the impact of noise on the number of segments and the inability to accurately capture changes in the data

Images