CN103218521A - Equipment state dynamic self-adaptive alarm method based on hidden semi-Markov model (HSMM) - Google Patents
Equipment state dynamic self-adaptive alarm method based on hidden semi-Markov model (HSMM) Download PDFInfo
- Publication number
- CN103218521A CN103218521A CN2013100967265A CN201310096726A CN103218521A CN 103218521 A CN103218521 A CN 103218521A CN 2013100967265 A CN2013100967265 A CN 2013100967265A CN 201310096726 A CN201310096726 A CN 201310096726A CN 103218521 A CN103218521 A CN 103218521A
- Authority
- CN
- China
- Prior art keywords
- value
- equipment
- normal condition
- state
- threshold value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Abstract
The invention discloses an equipment state dynamic self-adaptive alarm method based on a hidden semi-Markov model (HSMM). The method is characterized by comprising the following steps of solving a degradation index (DI) curve by establishing the HSMM for history data and equipment operating data and defining a performance DI; dividing the DI into a plurality of stages by a limit error method; and solving the upper alarm threshold value and the lower alarm threshold value of the DI. In the way, the few history data are required for modeling the HSMM; the HSMM is modeled conveniently; and meanwhile, alarm threshold values of equipment operation can be acquired dynamically; and the alarm is flexible and accurate.
Description
Technical field
The present invention relates to a kind of equipment state dynamic self-adapting alarm method, relate in particular to a kind of equipment state dynamic self-adapting alarm method based on latent half Markov model.
Background technology
The various device grading standard that uses in the enterprise all is some absolute standards at present, shakes earthquake intensity standard etc. as ISO-2372, and the alarm threshold value of these prescribed by standard all is static, is difficult to the particular device under the specific work environments is done corresponding adjustment.This has just caused particular device before wear damage or keep in repair afterwards or change, and the former has caused unnecessary waste like this, and the latter then can impact production.
And the adaptive alarm technology is to change and to change along with actual conditions such as the condition of work of equipment, working time, power, speed, sets up the dynamic judge rule of warning index and equipment operation situation, the dynamic warning curve of a variation of formation.The open book CN1472674A of Chinese patent discloses a kind of equipment state dynamic self-adapting alarm method based on probability model, dynamic data by equipment operation, the probability model of probability of use Neural Network Self-learning component equipment state, the Changing Pattern of equipment state is dynamically described, the adaptive alarm curve of forming device operation.Such method needs a large amount of history datas come to make up by probabilistic neural network self study method the probability model of equipment state, and directly by the raw data modeling, causes to float greatlyyer, and warning message is not very accurate.
Summary of the invention
Technical matters to be solved by this invention provides a kind of equipment state dynamic self-adapting alarm method based on latent half Markov model, this alarm method need historical data less, report to the police accurately.
For solving the problems of the technologies described above, the technical solution adopted in the present invention is:
A kind of equipment state dynamic self-adapting alarm method based on latent half Markov model is characterized in that comprising following steps:
A. according to device history data, with identical time span data are divided into plurality of sections, as observed reading, as the observation sequence of latent half Markov training and state recognition, observation sequence group number is the K group to every section characteristic information extraction;
B. training conceals half Markov model based on the Baum-Welch algorithm, respectively latent half Markov model of apparatus for establishing normal condition and degenerate states at different levels;
C. define the departure degree of each degenerate state based on Ku Er-Lai Beier information distance, as the performance degradation index DI of equipment with respect to normal condition;
D. set the minimum length N of observation sequence, equipment operation a period of time, when the DI of equipment normal condition satisfies N, calculate mean value and the standard deviation of DI, thereby obtain the last lower threshold value of normal condition alarming line by limiting error method principle;
E. along with the passage of time of equipment operation, constantly obtain new DI value,, then new DI value is joined normal DI sequence, recomputate the mean value and the standard deviation of DI sequence if new DI value drops in the last lower threshold value of normal condition alarming line;
F. when continuous K DI value dropped on outside the last lower threshold value of normal condition alarming line, equipment departed from normal condition, enters catagen phase one, triggered one-level and reported to the police; When the DI sequence length of catagen phase one reaches N then, obtain new DI and calculate its mean value and standard deviation, determine the last lower threshold value of the alarming line of catagen phase one according to the method for step D and step e; When continuous K DI value dropped on outside the last lower threshold value of catagen phase one, equipment departed from degenerate state one, enters degenerate state two, triggered secondary and reported to the police;
G. along with the continuous adjustment of new DI value, equipment running status enters degenerate state three, degenerate state four until degenerate state M according to the rule of step F, the stage that degenerate state M needs repairing for the equipment operation that draws according to historical data;
H. when the DI value enters into catagen phase M, report to the police, remind the staff in the light of actual conditions to overhaul or processing such as shutdown.
Adopt such technical scheme,, and, determine the alarm threshold value in each stage of equipment, finally determine alarm threshold, realize the adaptive alarm of equipment running status by calculating the curve of performance degradation index DI by latent half Markov model of historical data.Compare with the type of alarm of traditional absolute index, alarm method of the present invention has the function of obtaining alarm threshold value automatically, and method is flexible, report to the police accurately, and based on the method for latent half Markov model, required historical data is less, is convenient to operation.
Embodiment
Latent half Markov model is called for short HSMM.Equipment state dynamic self-adapting alarm method based on latent half Markov model of the present invention is example mainly to comprise following steps with using of bearing:
A.Historical data according to bearing life, with identical time span data are divided into plurality of sections, to the observed reading of every section characteristic information extraction as the measured bearing running status, all observed readings are formed observation sequence, as the observation sequence of HSMM training and state recognition, observation sequence group number is the K group;
B.Based on Baum-Welch algorithm training HSMM model, set up the HSMM model of bearing normal condition and degenerate states at different levels respectively; The Baum-Welch algorithm is used to solve the HMM training problem, i.e. HMM parameter estimation problem.The Baum-Welch algorithm can be described as, a given sequence of observations O={ o
1o
2... o
T, this algorithm can be determined a model
, make P
Maximum.This is a general culvert extreme-value problem, thereby does not exist a preferred plan to estimate λ.In this case, the Baum-Welch algorithm utilizes the thought of recurrence, makes P (O| λ) for local maximum, obtains the parameter of model at last.
Definition
During for given training observation sequence O and parameter model λ, Markov chain is in state during moment t
And moment t+1 is in the probability of state j:
Can derive to the definition of variable according to front forward direction and back:
T moment Markov chain is in the probability of state i so:
Therefore,
The expectation value that expression is gone out from state transitions, and
The expectation value of expression from the state transitions to the state, derived revaluation formula famous in the Baum-Welch algorithm thus:
The parameter lambda of HMM=(process of asking for B) is for π, A: according to observation sequence O and the initial model λ that chooses so
0=(π, A B), try to achieve one group of new parameter by the revaluation formula and have obtained a new model simultaneously, can prove, have promptly that the revaluation formula obtains than good aspect the expression sequence of observations O.Repeat this process, progressively the improved model parameter, promptly no longer obviously increases the model of being asked exactly of this moment up to satisfying certain condition of convergence
;
C.Define the departure degree of each degenerate state based on Ku Er-Lai Beier information distance, as the performance degradation index DI of bearing with respect to normal condition;
Ku Er-Lai Beier information number is called for short K-L information number, if
Be the probability density function of reference model,
Be the probability density function of pattern to be checked, can calculate the similarity degree of comparison two class states according to K-L information number
D.Define the departure degree of each degenerate state based on Ku Er-Lai Beier information distance, as the performance degradation index DI of bearing with respect to normal condition;
Ku Er-Lai Beier information number is called for short K-L information number, if
Be the probability density function of reference model,
Be the probability density function of pattern to be checked, can calculate the similarity degree of comparison two class states according to K-L information number
The thought of forward direction algorithm is to be calculated by the state i of moment t to the t+1 probability of all approach of shifting of state j constantly by recursion method, and these probable value sums are P
At first calculate forward direction and the backward probability of corresponding every t and each i in the actual computation, apply mechanically formula then:
The back is to algorithm
With the forward direction class of algorithms seemingly, definition back to variable is:
Wherein
T(i)=1, the back is as follows to the computation process of algorithm:
Try to achieve according to HSMM under the normal condition
, obtain corresponding then with the HSMM of current degenerate state
. by
,
Calculate the KL distance, just can differentiate the degree that current state departs from normal state, be defined as degeneration index DI according to its value size
E. set the minimum length N of observation sequence, equipment operation a period of time, when the DI of equipment normal condition satisfies the minimum length N of observation sequence, calculate mean value and the standard deviation of DI, thereby obtain the last lower threshold value of normal condition alarming line by limiting error method principle;
F. along with the passage of time of equipment operation, constantly obtain new DI value,, then new DI value is joined normal DI sequence, recomputate the mean value and the standard deviation of DI sequence if new DI value drops in the last lower threshold value of normal condition alarming line;
G. when continuous K DI value dropped on outside the last lower threshold value of normal condition alarming line, equipment departed from normal condition, enters catagen phase one, triggered one-level and reported to the police; When the DI sequence length of catagen phase one reaches the minimum length N of observation sequence then, obtain new DI and calculate its mean value and standard deviation, determine the last lower threshold value of the alarming line of catagen phase one according to the method for step D and step e; When continuous K DI value dropped on outside the last lower threshold value of catagen phase one, equipment departed from degenerate state one, enters degenerate state two, triggered secondary and reported to the police;
H.Along with the continuous adjustment of new DI value, equipment running status enters degenerate state three, degenerate state four until degenerate state M according to the rule of step F, the stage that degenerate state M needs repairing for the equipment operation that draws according to historical data;
I.. when the DI value enters into catagen phase M, report to the police, remind the staff in the light of actual conditions to overhaul or processing such as shutdown.
By such technical scheme, to set up the adaptive alarm curve of DI and obtain alarm threshold value up and down by HSMM by the limiting error method, needed historical data is less, and can dynamically obtain to determine alarm threshold value, and it is flexible, accurate to report to the police.
Claims (1)
1. equipment state dynamic self-adapting alarm method based on latent half Markov model is characterized in that comprising following steps:
A. according to device history data, with identical time span data are divided into plurality of sections, as observed reading, as the observation sequence of latent half Markov model training and state recognition, observation sequence group number is the K group to every section characteristic information extraction;
B. training conceals half Markov model based on the Baum-Welch algorithm, respectively latent half Markov model of apparatus for establishing normal condition and degenerate states at different levels;
C. define the departure degree of each degenerate state based on Ku Er-Lai Beier information distance, as the performance degradation index DI of equipment with respect to normal condition;
D. set the minimum length N of observation sequence, equipment operation a period of time, when the DI of equipment normal condition satisfies N, calculate mean value and the standard deviation of DI, thereby obtain the last lower threshold value of normal condition alarming line by limiting error method principle;
E. along with the passage of time of equipment operation, constantly obtain new DI value,, then new DI value is joined normal DI sequence, recomputate the mean value and the standard deviation of DI sequence if new DI value drops in the last lower threshold value of normal condition alarming line;
F. when continuous K DI value dropped on outside the last lower threshold value of normal condition alarming line, equipment departed from normal condition, enters catagen phase one, triggered one-level and reported to the police; When the DI sequence length of catagen phase one reaches N then, obtain new DI and calculate its mean value and standard deviation, determine the last lower threshold value of the alarming line of catagen phase one according to the method for step D and step e; When continuous K DI value dropped on outside the last lower threshold value of catagen phase one, equipment departed from degenerate state one, enters degenerate state two, triggered secondary and reported to the police;
G. along with the continuous adjustment of new DI value, equipment running status enters degenerate state three, degenerate state four until degenerate state M according to the rule of step F, the stage that degenerate state M needs repairing for the equipment operation that draws according to historical data;
H. when the DI value enters into catagen phase M, report to the police, remind the staff in the light of actual conditions to overhaul or processing such as shutdown.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2013100967265A CN103218521A (en) | 2013-03-25 | 2013-03-25 | Equipment state dynamic self-adaptive alarm method based on hidden semi-Markov model (HSMM) |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2013100967265A CN103218521A (en) | 2013-03-25 | 2013-03-25 | Equipment state dynamic self-adaptive alarm method based on hidden semi-Markov model (HSMM) |
Publications (1)
Publication Number | Publication Date |
---|---|
CN103218521A true CN103218521A (en) | 2013-07-24 |
Family
ID=48816300
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2013100967265A Pending CN103218521A (en) | 2013-03-25 | 2013-03-25 | Equipment state dynamic self-adaptive alarm method based on hidden semi-Markov model (HSMM) |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103218521A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105893208A (en) * | 2016-03-31 | 2016-08-24 | 城云科技(杭州)有限公司 | Cloud computing platform system fault prediction method based on hidden semi-Markov models |
CN106682503A (en) * | 2017-01-06 | 2017-05-17 | 浙江中都信息技术有限公司 | Application of genetic algorithm based hidden Markov model to mainframe risk assessment |
CN106885697A (en) * | 2017-03-17 | 2017-06-23 | 华东交通大学 | The performance degradation assessment method of the rolling bearing based on FCM HMM |
CN111080977A (en) * | 2019-12-27 | 2020-04-28 | 安徽芯核防务装备技术股份有限公司 | Self-adaptive threshold dynamic setting method and device based on internal environment change of bus |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1472674A (en) * | 2003-08-04 | 2004-02-04 | 西安交通大学 | Self-adapt dynamic apparatus status alarming method based on probability model |
CN102129397A (en) * | 2010-12-29 | 2011-07-20 | 深圳市永达电子股份有限公司 | Method and system for predicating self-adaptive disk array failure |
-
2013
- 2013-03-25 CN CN2013100967265A patent/CN103218521A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1472674A (en) * | 2003-08-04 | 2004-02-04 | 西安交通大学 | Self-adapt dynamic apparatus status alarming method based on probability model |
CN102129397A (en) * | 2010-12-29 | 2011-07-20 | 深圳市永达电子股份有限公司 | Method and system for predicating self-adaptive disk array failure |
Non-Patent Citations (1)
Title |
---|
彭颖: "基于退化隐式半马尔科夫模型的设备健康预测及系统性维护策略研究", 《上海交通大学博士学位论文》, 31 December 2011 (2011-12-31) * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105893208A (en) * | 2016-03-31 | 2016-08-24 | 城云科技(杭州)有限公司 | Cloud computing platform system fault prediction method based on hidden semi-Markov models |
CN106682503A (en) * | 2017-01-06 | 2017-05-17 | 浙江中都信息技术有限公司 | Application of genetic algorithm based hidden Markov model to mainframe risk assessment |
CN106682503B (en) * | 2017-01-06 | 2018-12-21 | 浙江中都信息技术有限公司 | Application based on the Hidden Markov Model of genetic algorithm in host risk assessment |
CN106885697A (en) * | 2017-03-17 | 2017-06-23 | 华东交通大学 | The performance degradation assessment method of the rolling bearing based on FCM HMM |
CN106885697B (en) * | 2017-03-17 | 2019-09-20 | 华东交通大学 | The performance degradation assessment method of rolling bearing based on FCM-HMM |
CN111080977A (en) * | 2019-12-27 | 2020-04-28 | 安徽芯核防务装备技术股份有限公司 | Self-adaptive threshold dynamic setting method and device based on internal environment change of bus |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110322731B (en) | Ship meeting navigation aid early warning method | |
CN107941537B (en) | A kind of mechanical equipment health state evaluation method | |
CN109377750B (en) | Traffic system elastic critical point determining method based on seepage analysis | |
CN104200288B (en) | A kind of equipment fault Forecasting Methodology based on dependency relation identification between factor and event | |
CN108870090B (en) | Pipeline leakage detection method based on least square support vector machine information fusion | |
CN103218521A (en) | Equipment state dynamic self-adaptive alarm method based on hidden semi-Markov model (HSMM) | |
AU2012201444A1 (en) | Wind power prediction method of single wind turbine generator | |
CN104537891B (en) | A kind of boats and ships track real-time predicting method | |
Huang et al. | Prognostics with variational autoencoder by generative adversarial learning | |
CN108418822A (en) | The automatic Compilation Method of TAF messages, system and the terminal of aeronautical meteorology | |
CN105678409A (en) | Adaptive and distribution-free time series abnormal point detection method | |
CN113607170B (en) | Real-time detection method for deviation behavior of navigation path of air-sea target | |
US20180046917A1 (en) | Identification of process anomalies in a technical facility | |
CN104504277A (en) | Early-warning method for collision of ship | |
CN113446157B (en) | Fault early warning method and equipment for offshore wind generating set | |
CN110532613B (en) | Ship power system operation condition identification method and device | |
Li et al. | Fuzzy Markov chains based on the fuzzy transition probability | |
CN111814908B (en) | Abnormal data detection model updating method and device based on data flow | |
Aizpurua et al. | Probabilistic power transformer condition monitoring in smart grids | |
KR20190131329A (en) | System for generating learning data | |
Paul et al. | Forecasting of reservoir inflow by the combination of deep learning and conventional machine learning | |
CN113933708B (en) | Self-correction method for intelligent fault cause diagnosis probability of generator set | |
Hasan et al. | Fault Occurrence Detection and Classification of Fault Type in Electrical Power Transmission Line with Machine Learning Algorithms. | |
Demir et al. | Predicting The Power of a Wind Turbine with Machine Learning-Based Approaches from Wind Direction and Speed Data | |
Hewa Nadungodage et al. | Online multi-dimensional regression analysis on concept-drifting data streams |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C12 | Rejection of a patent application after its publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20130724 |