Method and apparatus for monitoring system

Patent metadata

Field	Value
Publication number	US-11276004-B2
Application number	US-201715428986-A
Country	US
Kind code	B2
Filing date	Feb 9, 2017
Priority date	Sep 9, 2016
Publication date	Mar 15, 2022
Grant date	Mar 15, 2022

Abstract

A method and apparatus for monitoring a system are provided. The method includes: acquiring a series of historical status index data of a monitored system during at least one data collection period; introducing the series of historical status index data into a pre-trained recommended feature extraction model to perform a matching operation, to obtain a feature extraction algorithm matching the series of historical status index data as a recommended feature extraction algorithm; determining a normal value range of feature values obtained by performing feature extraction on the series of status index data of the monitored system according to the recommended feature extraction algorithm; and monitoring the monitored system according to the recommended feature extraction algorithm and the normal value range.

First claim

What is claimed is:

1. A method for monitoring a system, comprising:
acquiring a series of status index data of a monitored system during at least one data collection period, as a series of historical status index data;
obtaining a feature extraction algorithm matching the series of historical status index data by: introducing the series of historical status index data into a pre-trained recommended feature extraction model to match the series of historical status index data to the feature extraction algorithm according to a pre-trained corresponding relationship between the series of historical status index data and the obtained feature extraction algorithm, the obtained feature extraction algorithm being a recommended feature extraction algorithm, wherein the corresponding relationship between the series of historical status index data and the obtained feature extraction algorithm is characterized by the recommended feature extraction model;
determining a maximum value and a minimum value of feature values obtained by performing feature extraction on the series of historical status index data according to the recommended feature extraction algorithm, wherein the recommended feature extraction algorithm comprises a period-over-period feature extraction algorithm, or a chain relative feature extraction algorithm, and different recommended feature extraction algorithms correspond to different feature values;
determining, based on the determined maximum value and minimum value, a normal value range of feature values obtained by performing feature extraction on the series of historical status index data of the monitored system according to the recommended feature extraction algorithm, wherein performing feature extraction on the series of historical status index data of the monitored system according to the recommended feature extraction algorithm comprises: calculating, from a second data collection period in the series of historical status index data, a ratio of historical status index data of each data collection period to historical status index data of a previous collection period of the collection period; and defining the calculated ratio as feature values obtained by performing feature extraction on the series of historical status index data according to the period-over-period feature extraction algorithm or the chain relative feature extraction algorithm; and
monitoring the monitored system according to the recommended feature extraction algorithm and the normal value range.

2. The method according to claim 1 , the method further comprising building the recommended feature extraction model, the building the recommended feature extraction model comprising:
obtaining the recommended feature extraction model through training by using a machine learning method, based on a manually labeled series of historical status index data of the monitored system and the feature extraction algorithm corresponding to the series of historical status index data of the monitored system.

3. The method according to claim 1 , wherein the obtaining a feature extraction algorithm matching the series of historical status index data by introducing the series of historical status index data into a pre-trained recommended feature extraction model to match the series of historical status index data to the feature extraction algorithm according to a pre-trained corresponding relationship between the series of historical status index data and the feature extraction algorithm, the feature extraction algorithm being a recommended feature extraction algorithm comprises:
introducing the historical status index data series into the recommended feature extraction model to obtain feature extraction algorithms with a matching degree used to represent an accuracy of determining feature extraction algorithms based on the series of historical status index data; and
defining a feature extraction algorithm of the obtained feature extraction algorithms with a highest matching degree as the recommended feature extraction algorithm.

4. The method according to claim 1 , wherein the determining a maximum value and a minimum value of feature values obtained by performing feature extraction on the series of historical status index data according to the recommended feature extraction algorithm comprises:
performing feature extraction on the series of historical status index data according to the recommended feature extraction algorithm to obtain feature values of the series of historical status index data;
acquiring a preset maximum abnormal point proportion and a preset minimum abnormal point proportion based on which the monitored system utilizes the recommended feature extraction algorithm to perform the feature extraction, the preset maximum abnormal point proportion and the preset minimum abnormal point proportion being numerical values greater than or equal to 0 and smaller than 1;
acquiring a number of the obtained feature values as a first number;
calculating a first product of the first number and the preset minimum abnormal point proportion and rounding the calculated first product into a second number, the second number being the rounded first product;
calculating and rounding a second product of the first number and the preset maximum abnormal point proportion to be a third number, the third number being the rounded second product;
calculating a difference between the first number and the third number as a fourth number;
ordering the obtained feature values in an ascending order;
selecting, from the obtained feature values, an i th feature value in the ascending order as the minimum value of eigenvalues obtained by performing feature extraction on the series of historical status index data according to the recommended feature extraction algorithm, i being a sum of the second number and 1; and
selecting, from the obtained feature values, a j th feature value in the ascending order as the maximum value of feature values obtained by performing feature extraction on the series of historical status index data according to the recommended feature extraction algorithm, j being a difference between the fourth number and 1.

5. The method according to claim 1 , wherein the determining, based on the determined maximum value and minimum value, a normal value range of feature values obtained by performing feature extraction on the series of historical status index data of the monitored system according to the recommended feature extraction algorithm comprises:
defining a product of the determined minimum value and a predetermined reduction factor as a minimum value of the value range, the predetermined reduction factor being a numerical value greater than 0 and smaller than 1;
defining a product of the determined maximum value and a predetermined magnification factor as a maximum value of the value range, the predetermined magnification factor being a numerical value greater than 1; and
determining that the normal value range of feature values obtained by performing feature extraction on the series of historical status index data of the monitored system according to the recommended feature extraction algorithm is greater than or equal to the minimum value of the value range, and is smaller than or equal to the maximum value of the value range.

6. The method according to claim 1 , wherein the monitoring the monitored system according to the recommended feature extraction algorithm and the normal value range comprises:
collecting periodically current status index data of the monitored system and generating a series of current status index data as a monitored series of current status index data;
performing feature extraction on the monitored series of current status index data according to the recommended feature extraction algorithm

Assignees

• Beijing Baidu Netcom Sci & Tech Co Ltd

Inventors

• Chen Yun
• Wang Bo
• Wang Dong
• Qu Xianping
• Guo Xuanyou
• Sun Yongqian
• Wan Yangsha

Classifications

• G06F11/3072

  where the reporting involves data filtering, e.g. pattern matching, time or event triggered, adaptive or policy-based reporting · CPC title

• H04L43/04

  Processing captured monitoring data, e.g. for logfile generation · CPC title

• G06N20/00 Primary

  Machine learning · CPC title

• G06F11/3055 Primary

  Monitoring arrangements for monitoring the status of the computing system or of the computing system component, e.g. monitoring if the computing system is on, off, available, not available (error or fault processing without redundancy G06F11/0703; error detection or correction by redundancy in data representation G06F11/08; error detection or correction of the data by redundancy in operations G06F11/14; error detection or correction by redundancy in hardware G06F11/16) · CPC title

• G06N5/04 Primary

  Inference or reasoning models · CPC title