CN109446629A - A kind of industrial process alarm root recognition methods based on probability graph model - Google Patents
A kind of industrial process alarm root recognition methods based on probability graph model Download PDFInfo
- Publication number
- CN109446629A CN109446629A CN201811233927.4A CN201811233927A CN109446629A CN 109446629 A CN109446629 A CN 109446629A CN 201811233927 A CN201811233927 A CN 201811233927A CN 109446629 A CN109446629 A CN 109446629A
- Authority
- CN
- China
- Prior art keywords
- alarm
- variable
- root
- probability
- conditional probability
- 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.)
- Withdrawn
Links
- 238000000034 method Methods 0.000 title claims abstract description 25
- 238000004519 manufacturing process Methods 0.000 title claims description 8
- 230000008859 change Effects 0.000 claims description 3
- 239000003245 coal Substances 0.000 description 5
- 230000002159 abnormal effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 230000001364 causal effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B21/00—Alarms responsive to a single specified undesired or abnormal condition and not otherwise provided for
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Emergency Management (AREA)
- Monitoring And Testing Of Exchanges (AREA)
Abstract
The invention belongs to industrial alarm system administrative skill field more particularly to a kind of industry alarm root recognition methods based on probability graph model, including obtain alarm variable and alarm root variable, calculate probability when each alarm root variable is 1 and 0；Probability when alarm variable is 0 and 1 under specified criteria is obtained, priori conditions probability parameter collection θ is formed；Using current data sample X (t) to the priori conditions probability θ (t) of θ, updated in line interation；Calculating alarm variable is xaWhen, alarm root variable derives from the posteriority conditional probability of specified criteriaPosterior probability vector when alarm variable is 1 is calculated, by the element sort in descending order of the vector, the first row element is root of alarming；Not the case where alarm root of judgement alarm variable is not known root.The present invention can solve and not account for failing to report alert and false alarm in alarm root identification at present, more alarm roots exist simultaneously and the factors such as the imperfection of known alarm root and the root that causes to alarm identifies inaccurate problem.
Description
Technical Field
The invention belongs to the technical field of industrial alarm system management, and particularly relates to an industrial alarm root cause identification method based on a probability map model.
Background
The practical problem of the current industrial alarm system is that a large number of alarms are generated in a short time, the judgment of operators is disturbed, and the alarm flooding becomes the most troublesome problem of the current industrial alarm system. The primary task of treating alarm flooding is to determine the alarm source, identify the source in time, and make a response measure in time according to the alarm source to eliminate the alarm, so that an operator does not spend a large amount of time on treating secondary alarms and neglects key alarms. At present, research work aiming at alarm root cause identification mainly aims at the analysis aspect of a single alarm root cause, and research aiming at the problems existing in an industrial process and the incompleteness aspect of a known root cause is blank.
The identification of the alarm source means that the system can identify the variable which is possibly alarmed for the appeared alarm, so that the operator can identify the abnormal source in time and eliminate the alarm. The main problems existing in the prior alarm source identification are as follows: 1) the negative effects of missed alarm and false alarm are not considered, all alarms are regarded as effective alarms for analysis and identification, and the accuracy of alarm identification is reduced; (2) the simultaneous problem of a plurality of sources without considering alarm variables is easy to occur, and the possibility of further improving the abnormal risk caused by neglecting some important alarms is easy to occur; (3) whether the known root cause is complete or not and whether the unknown root cause exists or not are not considered, so that the alarm root cause identification is invalid, and the judgment of an operator is influenced.
The above three problems cause obstacles for the identification of the alarm source, and if the problems are not solved, inaccurate alarm source identification results can be caused, the judgment of operators is influenced, and the safety and economic losses in the industrial production process are caused.
Disclosure of Invention
According to the defects of the prior art, the invention provides an industrial alarm root cause identification method based on a probability graph model, which can solve the problem that the alarm root cause identification is inaccurate because factors such as missing alarm and false alarm, simultaneous existence of multiple alarm root causes, incompleteness of known alarm root causes and the like are not considered in the conventional alarm root cause identification.
The technical scheme adopted by the invention for solving the technical problems is as follows:
the industrial process alarm root identification method based on the probability map model comprises the following steps:
step 1, obtaining historical alarm variable Xa(t) and alarm variable parent node setCalculating each alarm root variable as an alarm stateProbability of timeAnd is in a non-alarm stateProbability of timeWherein theta isi,0(t)+θi,1(t)＝1,i∈[1,n]；
Step 2, obtaining the jth group alarm variable Xa(t) set of possible values generated by parent nodeDefining an alarm variable Xa(t) ═ 0 indicates that the alarm variable is in a non-alarm state, and the alarm variable X isa(t) ═ 1 indicates that the alarm variable is in the alarm state;
calculating j-th group of possible values of alarm root variable of given alarm variable father node set RUnder the condition of (2), alarm variable XaParameter theta of prior conditional probability of (t) ═ 0a,0|j(t) and alarm variable XaParameter theta of prior conditional probability of (t) ═ 1a,1|j(t) wherein θa,0|j(t):＝P(Xa(t)＝0|R＝rj)，θa,1|j(t):＝P(Xa(t)＝1|R＝rj)，
Combining the alarm root variable obtained in the step 1Probability of being in alarm stateProbability of non-alarm stateObtaining each alarm root variable in the Bayesian networkAnd an alarm variable Xa(t) a set of a priori conditional probability parameters θ (t) ═ θi,0(t),θi,1(t),θa,0|j(t),θa,1|j(t)}i＝1,2,...,n，j＝1,2,...2n；
Step 3, theta (t) represents each alarm root variable in the Bayesian networkAnd an alarm variable XaThe prior conditional probability parameter set formed by (t) adopts each alarm root variable in the Bayesian network at the moment of (t-1) in an iterative modeAnd an alarm variable XaThe composed prior conditional probability parameter set theta (t-1) and the observable current numberAccording to the samples X (t), X (t) ═ X1(t),X2(t),X3(t),...,Xn(t),Xa(t)]TRespectively under the conditionsIn four cases, for a prior conditional probability parameter set θ (t) in a bayesian network, { θ ═i,0(t),θi,1(t),θa,0|j(t),θa,1|j(t)}，i＝1,2,...,n，j＝1,2,...2nPerforming iterative learning, and updating the probability parameter set on line;
step 4, according to Bayesian rule, calculating current alarm variable Xa(t) has the value xaUnder the condition of (2), alarm root variableFrom the jth parent set rjA posteriori conditional probability ofWherein,
step 5, calculating an alarm variable XaWhen t is 1, alarm root variableFrom the jth parent set rjA posteriori conditional probability ofj|1(t)，j＝1,2,...2nWhere the denominators are the same, its numerator is denotedThe posterior probability vector theta being composed of the molecules of the posterior conditional probabilityp(t) wherein,the posterior probability vector thetap(t) elements are arranged in descending order, the posterior probability vector after descending orderIs marked as thetap,s(t) then taking the posterior probability vector θpThe element of (t) where the probability is the maximum, i.e., (θ)p,s(t),η)＝max(θp(t)), the source of the alarm is the first element η (1) of η when the alarm variable X is presentaWhen the (t) is 0, namely in a non-alarm state, the alarm source does not need to be analyzed;
step 6, if the posterior conditional probability thetaj|1(t) for j ∈ [1,2 ]n-1]Is the maximum posterior conditional probability, then the alarm variable XaWhen t is 1, the alarm source is the corresponding jth father node set rjParent node X equal to 11,...,Xn(ii) a If the last element j of j is 2nCorresponding posterior conditional probabilityAt maximum, alarm variable XaThe alarm source of (t) ═ 1 is not X1,...,Xn。
Further, the specific implementation of step 1 includes,
alarm variable Xa(t) ═ 1 due to the state change of n pairs of independent alarm root variables, and the parent node set consisting of the alarm root variables is denoted as R: { X1(t),X2(t),...,Xn(t) }, each alarm root cause variableThe probability of the alarm state "1" isWherein theta isi,1(t):＝P(Xi(t)＝1)，i∈[1,n](ii) a Each alarm root variableThe probability of the non-alarm state "0" isWherein, thetai,0(t):＝P(Xi(t)＝0)，i∈[1,n]Where n represents the number of parent node sets that contain the alarm root variable.
Further, the specific implementation of step 3 includes,
when in useAndwhen the prior conditional probability parameter is updated on line, the prior conditional probability parameter is updated as shown in the following formula,
wherein the constant lambda is an update ratio,m is the number of samples;
alarm root variableAnd alarm root variableIs updated as shown in the following formula, namely when Xi(t) ═ 1, i ═ 1, 2., n, and Xi(t) 0, i 1,2, when n,
wherein the constant lambda is an update ratio,m is the number of samples.
Further, the specific implementation of step 4 includes,
when alarm variable Xa(t) has the value xaDefining alarm root variableJ-th parent node set rjIs taken as xi,jRoot cause of alarmIndependent of each other, and alarm root variables are calculated according to a chain ruleIs combined with the probability P (X)1＝x1,j,...,Xn＝xn,j) Wherein
in combination with what has been obtained in the preceding step 3 The updated probability parameter set theta (t) in four cases is ═ thetai,0(t),θi,1(t),θa,0|j(t),θa,1|j(t)}，i＝1,2,...,n，j＝1,2,...2nAccording to Bayes' rule, the alarm variable X is obtaineda(t) taking the value xaUnder the condition of (2), alarm root variableFrom the jth parent set rjA posteriori conditional probability ofWherein,is updated online to
Further, in step 5, η represents the classified element index, which is determined by the parent node set with the maximum posterior conditional probability of the alarm root, and the first element η (1) of η is taken as the alarm root, and the alarm root vector is recordedAnd finishing the alarm source.
The invention has the following beneficial effects: (1) the invention provides an alarm root cause identification method for establishing causal relationship between 1 alarm variable and n alarm root cause variables based on a probability graph model of a Bayesian network, which considers the negative influence of misinformation and missing report of a system, considers the influence of the number of samples on iteration parameters and avoids misleading a prediction result due to insufficient number of samples; (2) the invention provides a method for calculating a root cause determined by a father node set with the maximum posterior conditional probability, and simultaneously considers the condition that a plurality of sources exist simultaneously or the current alarm variable XaAnd when the (t) is 1, the condition that the variable meeting the requirement cannot be found in the alarm sources listed, so that the accuracy and comprehensiveness of the alarm source prediction are embodied.
Drawings
FIG. 1 is a flow chart of a method of alarm root cause identification for an alarm system of the present invention;
FIG. 2 is a view of BayesProbability of a priori conditions of nodes of the s-networkAndθa,0|j(t) and θa,1|j(t), j ═ 1,2,3, 4;
FIG. 3 is a molecular timing diagram of the posterior conditional probability parameters of Bayesian network nodes
FIG. 4 is a warning root variable X1(t)，X2(t), alarm variable Xa(t) alarm root vector of the proposed methodTiming diagram of (2).
Detailed Description
The invention is further described below with reference to the accompanying drawings.
The first embodiment is as follows:
as shown in fig. 1 to 4, the industrial alarm root cause identification method based on the probability map model includes the following steps:
step S1, obtaining historical alarm variable Xa(t) and alarm variable parent node setCalculating each alarm root variable as an alarm stateProbability of timeAnd is in a non-alarm stateProbability of timeWherein theta isi,0(t)+θi,1(t)＝1,i∈[1,n]；
Step S2, obtaining the jth group alarm variable Xa(t) set of possible values generated by parent nodeDefining an alarm variable Xa(t) ═ 0 indicates that the alarm variable is in a non-alarm state, and the alarm variable X isa(t) ═ 1 indicates that the alarm variable is in the alarm state;
calculating j-th group of possible values of alarm root variable of given alarm variable father node set RUnder the condition of (2), alarm variable XaParameter theta of prior conditional probability of (t) ═ 0a,0|j(t) and alarm variable XaParameter theta of prior conditional probability of (t) ═ 1a,1|j(t) wherein θa,0|j(t):＝P(Xa(t)＝0|R＝rj)，θa,1|j(t):＝P(Xa(t)＝1|R＝rj)，
Combining the alarm root variable obtained in the step 1Probability of being in alarm stateProbability of non-alarm stateObtaining each alarm root variable in the Bayesian networkAnd an alarm variable Xa(t) a set of a priori conditional probability parameters θ (t) ═ θi,0(t),θi,1(t),θa,0|j(t),θa,1|j(t)}i＝1,2,...,n，j＝1,2,...2n；
In step S3, theta (t) represents each alarm root variable in the Bayesian networkAnd an alarm variable XaThe prior conditional probability parameter set formed by (t) adopts each alarm root variable in the Bayesian network at the moment of (t-1) in an iterative modeAnd an alarm variable Xa(t-1) and an observable current data sample X (t), X (t) ([ X1(t),X2(t),X3(t),...,Xn(t),Xa(t)]TRespectively under the conditions In four cases, for a prior conditional probability parameter set θ (t) in a bayesian network, { θ ═i,0(t),θi,1(t),θa,0|j(t),θa,1|j(t)}，i＝1,2,...,n，j＝1,2,...2nCarrying out iterative learning and updating the probability parameter set on line;
step S4, according to Bayes rule, calculating current alarm variable Xa(t) has the value xaUnder the condition of (2), alarm root variableFrom the jth parent set rjA posteriori conditional probability ofWherein,
step S5, calculating alarm variable XaWhen t is 1, alarm root variableFrom the jth parent set rjA posteriori conditional probability ofj|1(t)，j＝1,2,...2nWhere the denominators are the same, its numerator is denotedThe posterior probability vector theta being composed of the molecules of the posterior conditional probabilityp(t) wherein,the posterior probability vector thetap(t) arranging the elements in descending order, and recording the posterior probability vector after descending order as thetap,s(t) then taking the posterior probability vector θpThe element of (t) where the probability is the maximum, i.e., (θ)p,s(t),η)＝max(θp(t)), the source of the alarm is the first element η (1) of η when the alarm variable X is presentaWhen the (t) is 0, namely in a non-alarm state, the alarm source does not need to be analyzed;
step S6, if the conditional probability theta is posteriorj|1(t) for j ∈ [1,2 ]n-1]Is the maximum posterior conditional probability, then the alarm variable XaWhen t is 1, the alarm source is the corresponding jth father node set rjParent node X equal to 11(t),...,Xn(t); if the last of jElement j is 2nCorresponding posterior conditional probabilityThen alarm variable XaThe alarm source of (t) ═ 1 is not X1(t),...,Xn(t), i.e. when the alarm variable X is presentaThe alarm source of (t) ═ 1 is not a known source.
In step S1, an alarm variable Xa(t) ═ 1 due to the state change of n pairs of independent alarm root variables, and the parent node set consisting of the alarm root variables is denoted as R: { X1(t),X2(t),...,Xn(t) }, each alarm root cause variableThe probability of the alarm state "1" isWherein theta isi,1(t):＝P(Xi(t)＝1)，i∈[1,n](ii) a Each alarm root variableThe probability of the non-alarm state "0" isWherein, thetai,0(t):＝P(Xi(t)＝0)，i∈[1,n]Where n represents the number of parent node sets that contain the alarm root variable.
In step S3, whenAndwhen the prior conditional probability parameter is updated on line, the prior conditional probability parameter is updated as shown in the following formula,
wherein the constant lambda is an update ratio,m is the number of samples;
alarm root variableAnd alarm root variableIs updated as shown in the following formula, namely when Xi(t) ═ 1, i ═ 1, 2., n, and Xi(t) 0, i 1,2, when n,
wherein the constant lambda is an update ratio,m is the number of samples.
In step S4, when the alarm variable X is seta(t) has the value xaDefining alarm root variableJ-th parent node set rjIs taken as xi,jRoot cause of alarmIndependent of each other, and alarm root variables are calculated according to a chain ruleIs combined with the probability P (X)1(t)＝x1,j,...,Xn(t)＝xn,j) Wherein
in combination with what has been obtained in the preceding step S3 The updated probability parameter set theta (t) in four cases is ═ thetai,0(t),θi,1(t),θa,0|j(t),θa,1|j(t)}，i＝1,2,...,n，j＝1,2,...2nAccording to Bayes' rule, the alarm variable X is obtaineda(t) taking the value xaUnder the condition of (2), alarm root variableFrom the jth parent set rjA posteriori conditional probability ofWherein,is updated online to
In step S5, η represents the classified element index, which is determined by the parent node set with the alarm root taking the maximum posterior conditional probability, and the first element η (1) of η is taken as the alarm root, and the alarm root is recordedAnd finishing the alarm source.
In step S6, when the alarm variable X is setaWhen (t) is 1, assume that the number n of alarm root variables is 2, i.e. X1(t)，X2(t), the possible state combination of the two alarm root variables is { X1(t)＝1,X2(t)＝0}，{X1(t)＝0,X2(t)＝1}，{X1(t)＝1,X2(t)＝1}，{X1(t)＝0,X2(t) ═ 0}, and these four states correspond to j ═ 1, j ═ 2, j ═ 3, and j ═ 4, respectively, so if the conditional probability θ is posterior toj|1(t)，j∈[1,22-1]The maximum posterior conditional probability, then the alarm source is the corresponding jth father node set rjParent node X equal to 11(t) or X2(t) or X1(t)、X2(t); but if the last element j of j is 22The posterior conditional probability corresponding to 4 is maximum because of { X1(t)＝0,X2(t) ═ 0} hence alarm variable XaThe alarm source of (t) ═ 1 is not X1(t),X2(t), i.e. when the alarm variable X is presentaThe alarm source of (t) ═ 1 is not a known source.
The invention provides an alarm root cause identification method for establishing causal relationship between 1 alarm variable and n alarm root cause variables based on a probability graph model of a Bayesian network, which considers the negative influence of misinformation and missing report of a system, considers the influence of the number of samples on iteration parameters, avoids misleading on a prediction result due to insufficient number of samples, determines a calculation method for taking a father node set with the maximum posterior conditional probability as an alarm root cause, and considers the simultaneous existence of a plurality of sources or the existence of an alarm variable X at presentaAnd (t) the condition that the variable meeting the requirement cannot be found in the alarm sources listed in the (1), and the accuracy and comprehensiveness of the alarm source prediction are embodied.
The following is the application of the method of the invention in a specific application scenario.
Using the operating state of the coal feeder as a reportAnd the alarm variable identifies the alarm source. Coal supply Xc(t) is a low alarm threshold of 10t/h, and according to knowledge of the process, the alarm state X isa(t) may be due to multiple sources, considering two major sources for ease of illustration, reducing the amount of coal input by X1(t), Motor failure X2(t)。
Step 1: obtaining historical alarm variable Xa(t) and alarm root variable X1(t),X2(t) calculating the root variable of each alarm as 1 and 0, namely X1(t)＝0，X1(t)＝1，X2(t)＝0，X2The probability when (t) is 1, is represented by θ1,0(t),θ1,1(t),θ2,0(t),θ2,1(t) is represented by, as shown in FIG. 2, wherein (a) θ1,0(t) (dotted line), θ1,1(t) (solid line), (b) θ2,0(t) (dotted line), θ2,1(t) (solid line);
step 2: obtaining jth group of alarm variables Xa(t) set of possible values generated by parent nodeWhen n is 2, the subscript j is 1,2,3 and 4, and the value of the given alarm variable father node set R is calculatedTime, stand for alarm variable Xa(t) 0 and an alarm variable XaParameter theta of prior conditional probability when (t) is 1a,0|j＝1(t)，θa,1|j＝1(t),θa,0|j＝2(t)，θa,1|j＝2(t)，θa,0|j＝3(t)，θa,1|j＝3(t)，θa,0|j＝4(t)，θa,1|j＝4(t) as shown in FIG. 2, wherein (c) θa,0|j＝1(t) (dotted line), θa,1|j＝1(t) (solid line), (d) θa,0|j＝2(t) (dotted line), θa,1|j＝2(t) (solid line), (e) θa,0|j＝3(t) (dotted line), θa,1|j＝3(t) (solid line), (f) θa,0|j＝4(t) (dotted line), θa,1|j＝4(t) (solid line).
Combining stepProbability θ obtained in 11,0(t)，θ1,1(t)，θ2,0(t)，θ2,1(t) obtaining a prior conditional probability parameter set θ (t) in the bayesian network, { θ ═ θ }i,0(t),θi,1(t),θa,0|j(t),θa,1|j(t)}，i＝1,2，j＝1,2,3,4。
And step 3: iteratively from observable data samplesFor a prior conditional probability parameter set θ (t) ═ θ in a Bayesian networki,0(t),θi,1(t),θa,0|j(t),θa,1|j(t), i is 1,2, j is 1,2,3,4, performing iterative learning, and updating the prior conditional probability parameter set theta (t) online;
and 4, step 4: according to Bayes' rule, calculating alarm variable Xa(t) taking the value xaAlarm root variable under conditionFrom the jth parent set rjA posteriori conditional probability of
And 5: obtaining an alarm variable XaWhen t is 1, alarm root variableFrom the jth parent set rjA posteriori conditional probability ofMolecular set of (2)Are respectively as As shown in fig. 3, in which,calculating the alarm variable XaPosterior probability vector theta when (t) is 1p(t) wherein,the posterior probability vector thetap(t) arranging the elements in descending order, and recording the posterior probability vector after descending order as thetap,s(t) then taking the posterior probability vector θpThe element of (t) where the probability is the maximum, i.e., (θ)p,s(t),η)＝max(θp(t)), the alarm root cause is η, the first element η (1) is the alarm root cause vectorNamely, it isAnd finishing the alarm source.
Reduction of coal input by X1(t), Motor failure X2(t), alarm variable Xa(t) and alarm root vectorAs shown in FIG. 4, wherein X1(t) (solid line), X2(t) (dotted line), Xa(t) (dotted line),(dot-dash line), X is removed1(t)，X2(t) negative effects of false positives and false negatives at intervals of less than 20 samples. Here, X1(t) false alarm (X)1(t) ═ 1), at time interval t e [2303,2316 ∈]And t e [2348,2351]False alarm (X)1(t) ═ 0), at time interval t e [1354,1355 ∈]，t∈[1362,1364]，t∈[1522,1525]，t＝1698，t∈[1844,1849]，t∈[2429,2436]，t∈[2453,2457]；X2(t) false alarm (X)2(t) ═ 1), at time interval t e [1854,1856 ∈]；Xa(t) false alarm (X)a(t) ═ 1), at time interval t e [2305,2317 ∈]，t∈[2328,2330]，t∈[2349,2355]t∈[2363,2364]Alarm against leakage (X)a(t) ═ 0), at time interval t e [1367,1368 ∈]，t∈[1524,1527]，t∈[2425,2432]，t∈[2436,2439]In contrast, alarm root vectorAnd is not affected by the false reports and the false missing reports.
Meanwhile, the problem of the combination of multiple sources can be solved. At time interval t e 1250,1600]Decrease of coal feed amount X1(t) is the first to result in Xa(t) root of alarm State, Motor Fault X2(t) is the second to result in Xa(t) the root cause of the alarm condition. The method takes correct result to alarm the root vector under the joint existence of two roots
In addition, the proposed method can find imperfections of known origin. Time interval t e 2530,3000]，Xa(t) is in an alarm state, but X1(t) and X2(t) are all in "0" state, thus the cause of the alarm is not X1(t) and X2(t) the proposed method also successfully detects the alarm root cause vector
If the usual method of observing measurable data samples is used here, the alarm variable XaThe root cause of (t) ═ 1 is Xi(t)＝1，i∈[1,n]The alarm variable of (2). Root cause of fruit useIs shown, i.e.
Here, theSerious quilt X1(t) and X2The false alarm of (t) affects and can not process the imperfection problem of the known root cause, so the alarm root cause vectorRatio ofHas better performance.
The above description is an exemplary embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications, equivalents, and flow charts produced by using the contents of the present invention and the accompanying drawings, which are directly or indirectly applied to other related technical fields, are included in the scope of the present invention.
Claims (5)
1. An industrial process alarm root cause identification method based on a probability map model is characterized in that: the method comprises the following steps:
step 1, obtaining historical alarm variable Xa(t) and alarm variable parent node setCalculating each alarm root variable as an alarm stateProbability of timeAnd is in a non-alarm stateProbability of timeWherein theta isi,0(t)+θi,1(t)＝1,i∈[1,n]；
Step 2, obtaining the jth group alarm variable Xa(t) set of possible values generated by parent nodeDefining an alarm variable Xa(t) ═ 0 indicates that the alarm variable is in a non-alarm state, and the alarm variable X isa(t) ═ 1 indicates that the alarm variable is in the alarm state;
calculating j-th group of possible values of alarm root variable of given alarm variable father node set RUnder the condition of (2), alarm variable XaParameter theta of prior conditional probability of (t) ═ 0a,0|j(t) and alarm variable XaParameter theta of prior conditional probability of (t) ═ 1a,1|j(t) wherein θa,0|j(t):＝P(Xa(t)＝0|R＝rj)，θa,1|j(t):＝P(Xa(t)＝1|R＝rj)，θa,0|j(t)+θa,1|j(t)＝1,
Combining the alarm root variable obtained in the step 1Probability of being in alarm stateProbability of non-alarm stateObtaining each alarm root variable in the Bayesian networkAnd an alarm variable Xa(t) a set of a priori conditional probability parameters θ (t) ═ θi,0(t),θi,1(t),θa,0|j(t),θa,1|j(t)}i＝1,2,...,n，j＝1,2,...2n；
Step 3, theta (t) represents each alarm root variable in the Bayesian networkAnd an alarm variable XaThe prior conditional probability parameter set formed by (t) adopts each alarm root variable in the Bayesian network at the moment of (t-1) in an iterative modeAnd an alarm variable Xa(t) a set of a priori conditional probability parameters θ (t-1) and an observable current data sample X (t), X (t): X ═ X1(t),X2(t),X3(t),...,Xn(t),Xa(t)]TUnder the condition X respectivelya(t)＝1,R(t)＝rj，Xa(t)＝0,R(t)＝rj，In four cases, for a prior conditional probability parameter set θ (t) in a bayesian network, { θ ═i,0(t),θi,1(t),θa,0|j(t),θa,1|j(t)}，i＝1,2,...,n，j＝1,2,...2nPerforming an iterationLearning, updating the probability parameter set on line;
step 4, according to Bayesian rule, calculating current alarm variable Xa(t) has the value xaUnder the condition of (2), alarm root variableFrom the jth parent set rjA posteriori conditional probability ofWherein,
step 5, calculating an alarm variable XaWhen t is 1, alarm root variableFrom the jth parent set rjA posteriori conditional probability ofj|1(t)，j＝1,2,...2nWhere the denominators are the same, its numerator is denotedThe posterior probability vector theta being composed of the molecules of the posterior conditional probabilityp(t) wherein,the posterior probability vector thetap(t) arranging the elements in descending order, and recording the posterior probability vector after descending order as thetap,s(t) then taking the posterior probability vector θpThe element of (t) where the probability is the maximum, i.e., (θ)p,s(t),η)＝max(θp(t)), the source of the alarm is the first element η (1) of η when the alarm variable X is presentaWhen the (t) is 0, namely in a non-alarm state, the alarm source does not need to be analyzed;
step 6, if the posterior conditional probability thetaj|1(t) for j ∈ [1,2 ]n-1]Is the maximum posterior conditional probability, then the alarm variable XaWhen t is 1, the alarm source is the corresponding jth father node set rjParent node X equal to 11,...,Xn(ii) a If the last element j of j is 2nCorresponding posterior conditional probabilityAt maximum, alarm variable XaThe alarm source of (t) ═ 1 is not X1,...,Xn。
2. The industrial process alarm root cause identification method based on the probability map model as claimed in claim 1, characterized in that: the specific implementation of the step 1 comprises that,
alarm variable Xa(t) ═ 1 due to the state change of n pairs of independent alarm root variables, and the parent node set consisting of the alarm root variables is denoted as R: { X1(t),X2(t),...,Xn(t) }, each alarm root cause variableThe probability of the alarm state "1" isWherein theta isi,1(t):＝P(Xi(t)＝1)，i∈[1,n](ii) a Each alarm root variableThe probability of the non-alarm state "0" isWherein, thetai,0(t):＝P(Xi(t)＝0)，i∈[1,n]Where n represents the number of parent node sets that contain the alarm root variable.
3. The industrial process alarm root cause identification method based on the probability map model as claimed in claim 1, characterized in that: the specific implementation of the step 3 includes that,
when X is presenta(t)＝1,R(t)＝rj，And Xa(t)＝0,R(t)＝rj，When the prior conditional probability parameter is updated on line, the prior conditional probability parameter is updated as shown in the following formula,
wherein the constant lambda is an update ratio,m is the number of samples;
alarm root variableAnd alarm root variableIs updated as shown in the following formula, namely when Xi(t) ═ 1, i ═ 1, 2., n, and Xi(t) 0, i 1,2, when n,
wherein the constant lambda is an update ratio,m is the number of samples.
4. The industrial process alarm root cause identification method based on the probability map model as claimed in claim 1, characterized in that: the specific implementation of the step 4 includes that,
when alarm variable Xa(t) has the value xaDefining alarm root variableJ-th parent node set rjIs taken as xi,jRoot cause of alarmIndependent of each other, and alarm root variables are calculated according to a chain ruleIs combined with the probability P (X)1(t)＝x1,j,...,Xn(t)＝xn,j) Wherein
combining X obtained in the previous step 3a(t)＝1,R(t)＝rj，Xa(t)＝0,R(t)＝rj，The updated prior conditional probability parameter set theta (t) in four cases is ═ thetai,0(t),θi,1(t),θa,0|j(t),θa,1|j(t)}，i＝1,2,...,n，j＝1,2,...2nAccording to Bayes' rule, the alarm variable X is obtaineda(t) taking the value xaUnder the condition of (2), alarm root variableFrom the jth parent set rjA posteriori conditional probability ofWherein,is updated online to
5. The method as claimed in claim 1, wherein in step 5, η represents the classified element index determined by the parent node set with the maximum posterior conditional probability of the alarm root, the first element η (1) of η is taken as the alarm root, and the alarm root vector is recordedNamely, it isAnd finishing the alarm source.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811233927.4A CN109446629A (en) | 2018-10-23 | 2018-10-23 | A kind of industrial process alarm root recognition methods based on probability graph model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811233927.4A CN109446629A (en) | 2018-10-23 | 2018-10-23 | A kind of industrial process alarm root recognition methods based on probability graph model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109446629A true CN109446629A (en) | 2019-03-08 |
Family
ID=65547907
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811233927.4A Withdrawn CN109446629A (en) | 2018-10-23 | 2018-10-23 | A kind of industrial process alarm root recognition methods based on probability graph model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109446629A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110176132A (en) * | 2019-05-27 | 2019-08-27 | 山东科技大学 | Logic alarm root-cause analysis method and system based on Bayesian network |
| CN111915855A (en) * | 2020-07-20 | 2020-11-10 | 浙江浙能台州第二发电有限责任公司 | Alarm analysis method for alarm system of million-kilowatt ultra-supercritical unit |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102724199A (en) * | 2012-06-26 | 2012-10-10 | 北京航空航天大学 | Attack intention recognition method based on Bayesian network inference |
| CN103713976A (en) * | 2013-12-30 | 2014-04-09 | 卡斯柯信号有限公司 | Signal appliance fault source searching method for signal central monitoring system |
-
2018
- 2018-10-23 CN CN201811233927.4A patent/CN109446629A/en not_active Withdrawn
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102724199A (en) * | 2012-06-26 | 2012-10-10 | 北京航空航天大学 | Attack intention recognition method based on Bayesian network inference |
| CN103713976A (en) * | 2013-12-30 | 2014-04-09 | 卡斯柯信号有限公司 | Signal appliance fault source searching method for signal central monitoring system |
Non-Patent Citations (1)
| Title |
|---|
| JIANDONG WANG, ET AL: ""Root-cause analysis of occurring alarms in thermal power plants based on Bayesian networks"", 《ELECTRICAL POWER AND ENERGY SYSTEMS》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110176132A (en) * | 2019-05-27 | 2019-08-27 | 山东科技大学 | Logic alarm root-cause analysis method and system based on Bayesian network |
| CN111915855A (en) * | 2020-07-20 | 2020-11-10 | 浙江浙能台州第二发电有限责任公司 | Alarm analysis method for alarm system of million-kilowatt ultra-supercritical unit |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108197845B (en) | Transaction index abnormity monitoring method based on deep learning model LSTM | |
| US11507049B2 (en) | Method for detecting abnormity in unsupervised industrial system based on deep transfer learning | |
| CN107561997B (en) | A kind of power equipment state monitoring method based on big data decision tree | |
| EP2410312A1 (en) | A method for computer-assisted analyzing of a technical system | |
| CN111949480B (en) | Log anomaly detection method based on component perception | |
| Parvez et al. | Real-time pattern matching and ranking for early prediction of industrial alarm floods | |
| CN112906738B (en) | Water quality detection and treatment method | |
| CN115237717A (en) | Micro-service abnormity detection method and system | |
| CN109446629A (en) | A kind of industrial process alarm root recognition methods based on probability graph model | |
| Lyu et al. | A data-driven approach for identifying possible manufacturing processes and production parameters that cause product defects: A thin-film filter company case study | |
| CN111126477A (en) | Learning and reasoning method of hybrid Bayesian network | |
| CN110781206A (en) | Method for predicting whether electric energy meter in operation fails or not by learning meter-dismantling and returning failure characteristic rule | |
| CN116882292B (en) | Lost circulation overflow early warning method based on LightGBM and anomaly detection algorithm | |
| Parvez et al. | An association rule mining approach to predict alarm events in industrial alarm floods | |
| Molinié et al. | BSOM: a two-level clustering method based on the efficient self-organizing maps | |
| Smirnov | Intelligent decision support system for the control of complex technical systems | |
| CN115169426B (en) | Anomaly detection method and system based on similarity learning fusion model | |
| CN110176132A (en) | Logic alarm root-cause analysis method and system based on Bayesian network | |
| König et al. | Towards algorithm-agnostic uncertainty estimation: Predicting classification error in an automated machine learning setting | |
| CN113918416A (en) | Software system abnormity positioning method based on event map | |
| Valero et al. | AI FOR DETECTING VARIATIONS IN THE OEE DATA RECEPTION RATE IN THE MANUFACTURING INDUSTRY | |
| Mori et al. | Dynamic Bayesian network based networked process monitoring for fault propagation identification and root cause diagnosis of complex dynamic processes | |
| CN117290801A (en) | Sequential monitoring index anomaly detection method based on Gaussian process regression | |
| Liu et al. | Research on interval triangular fuzzy multi-attribute fault diagnosis methods based on the grey relation grade | |
| CN117648588B (en) | Meteorological radar parameter anomaly identification method based on correlation network graph cluster analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190308 |
|
| WW01 | Invention patent application withdrawn after publication |