Method and device for detecting events on time series

Abstract

The invention discloses a method and device for detecting events on a time series, and relates to the field of intelligent pipelines. The search region L of the event length is segmented, for each segmented section (p, q), sp, sq and s (p...q) are calculated, if the s (p...q) of one section is smaller than the sp or the sq of any section, the section is cut out, and the residual sections are further segmented, each subsection obtained through continuous segmentation is cut out through the above segmentation method, and till the residual search region of the event length is small enough to meet the preset value, the segmentation and cutting out are stopped. Traversal is conducted on the points of the residual search region of the event length, important events corresponding to all the points and the degrees of importance of the important events are recorded, and the event, with the largest degree of importance, in all the important events serves as a global important event. No priori knowledge is needed in the whole event detecting process, automatic adaptation of event window width is achieved, and wrong recognizing or loss events can be avoided; several subsections are cut out, the search region of the events is shortened, and complexity of a detecting algorithm is lowered.

Description

technical field [0001] The invention relates to the field of intelligent pipelines, in particular to a time series event detection method and device. Background technique [0002] A time series is a sequence formed by continuously sampling an attribute of an object at uniform time intervals. For example: the instantaneous traffic of a network port at noon every day constitutes a time series. [0003] Informally speaking, event detection on time series refers to the detection of "peaks" in time series, "peaks" are called "events", and "peaks" indicate that the sequence value in this time period is significantly higher than normal, usually caused by certain special events. For example: an event on a sequence of network traffic may originate from a flood attack. [0004] In order to eliminate the influence of noise (such as outliers) when detecting events, currently, segmental summation or fixed-width window smoothing techniques are mainly used. These two techniques have th...

AI Technical Summary

Problems solved by technology

[0005] A technical problem to be solved by the embodiment of the present invention is to solve the problem that the traditional event detection technology requires prior knowledge, and the problem of misrecognized events or lost events caused by the deviation of prior knowledge

[0006] Another technical problem to be solved by the embodiments of the present invention is to solve the problem of lack of a unified comparison standard between events of different lengths

Images