Time series concept drift detection method and system, medium and equipment

Abstract

The invention provides a time sequence concept drift detection method and system, a medium and equipment, and the method comprises the steps: carrying out decomposition of an obtained original time sequence signal through employing an empirical mode decomposition method based on extreme value symmetric extension, and obtaining an intrinsic mode component containing the feature information of different time scales of the original signal; performing coarse graining processing on the obtained intrinsic mode component through fuzzy entropy to obtain an intrinsic mode component converted by using the fuzzy entropy; monitoring the intrinsic mode component after fuzzy entropy conversion by adopting a non-parametric statistical process control model based on generalized likelihood ratio test, calculating the maximum degree of freedom of the generalized likelihood ratio test, and comparing the maximum degree of freedom with a preset control threshold to determine whether a mean value, a variance or drifting of the mean value and the variance occurs or not; conceptual drift detection is realized from the perspective of time domain characteristics of different frequencies.

Description

technical field [0001] The present disclosure relates to the technical field of time series concept drift detection, in particular to a time series concept drift detection method, system, medium and equipment. Background technique [0002] The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art. [0003] The study of time series has strong theoretical significance and application value in real life. Due to its practical importance, the study of time series widely exists in the fields of finance, engineering, and medicine. When learning from time series, the target concept may change over time due to some external interference. For example, the patterns of ECG data of an epileptic patient before and after the onset are different, so the It will lead to differences between the previous time series model and the current model, and then the algorithm or model learned based on the previou...

AI Technical Summary

Problems solved by technology

[0005] The inventors of the present disclosure found in the research that although concept drift has received a lot of attention, most of the research is based on classification problems, and only a small part is dedicated to the concept drift of time series. The main reasons are as follows: (1) most of the concept drift detection Algorithms are based on the performance indicators of detection classifiers, and time series data are difficult to label in real environments, so ground truth (the accuracy of the training set for the classification of supervised learning techniques) is also an unavoidable problem;( 2) Time series data has time dependence, and many concept drift detection methods need to make assumptions about the distribution of data, or require data to be independent and identically distributed, so due to the particularity of time series data, they cannot be directly applied Other existing models; (3) In the real environment, due to the influence of noise or abnormality in the time series, the obtained data is difficult to be directly used for learning

Images