CN111443686B - Industrial alarm design method based on multi-objective optimization and evidence iterative update - Google Patents
Industrial alarm design method based on multi-objective optimization and evidence iterative update Download PDFInfo
- Publication number
- CN111443686B CN111443686B CN202010207522.4A CN202010207522A CN111443686B CN 111443686 B CN111443686 B CN 111443686B CN 202010207522 A CN202010207522 A CN 202010207522A CN 111443686 B CN111443686 B CN 111443686B
- Authority
- CN
- China
- Prior art keywords
- alarm
- input
- evidence
- reference value
- output
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000005457 optimization Methods 0.000 title claims abstract description 21
- 238000013461 design Methods 0.000 title claims abstract description 12
- 230000004927 fusion Effects 0.000 claims abstract description 27
- 238000013507 mapping Methods 0.000 claims abstract description 27
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 6
- 235000012907 honey Nutrition 0.000 claims description 50
- 239000013598 vector Substances 0.000 claims description 45
- 238000009826 distribution Methods 0.000 claims description 22
- 239000011159 matrix material Substances 0.000 claims description 18
- 238000012544 monitoring process Methods 0.000 claims description 13
- 241000257303 Hymenoptera Species 0.000 claims description 12
- 230000002159 abnormal effect Effects 0.000 claims description 11
- 238000012549 training Methods 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 8
- 238000005259 measurement Methods 0.000 claims description 7
- MOVRNJGDXREIBM-UHFFFAOYSA-N aid-1 Chemical compound O=C1NC(=O)C(C)=CN1C1OC(COP(O)(=O)OC2C(OC(C2)N2C3=C(C(NC(N)=N3)=O)N=C2)COP(O)(=O)OC2C(OC(C2)N2C3=C(C(NC(N)=N3)=O)N=C2)COP(O)(=O)OC2C(OC(C2)N2C3=C(C(NC(N)=N3)=O)N=C2)COP(O)(=O)OC2C(OC(C2)N2C(NC(=O)C(C)=C2)=O)COP(O)(=O)OC2C(OC(C2)N2C3=C(C(NC(N)=N3)=O)N=C2)COP(O)(=O)OC2C(OC(C2)N2C3=C(C(NC(N)=N3)=O)N=C2)COP(O)(=O)OC2C(OC(C2)N2C3=C(C(NC(N)=N3)=O)N=C2)COP(O)(=O)OC2C(OC(C2)N2C(NC(=O)C(C)=C2)=O)COP(O)(=O)OC2C(OC(C2)N2C3=C(C(NC(N)=N3)=O)N=C2)COP(O)(=O)OC2C(OC(C2)N2C3=C(C(NC(N)=N3)=O)N=C2)COP(O)(=O)OC2C(OC(C2)N2C3=C(C(NC(N)=N3)=O)N=C2)COP(O)(=O)OC2C(OC(C2)N2C(NC(=O)C(C)=C2)=O)COP(O)(=O)OC2C(OC(C2)N2C3=C(C(NC(N)=N3)=O)N=C2)COP(O)(=O)OC2C(OC(C2)N2C3=C(C(NC(N)=N3)=O)N=C2)COP(O)(=O)OC2C(OC(C2)N2C3=C(C(NC(N)=N3)=O)N=C2)CO)C(O)C1 MOVRNJGDXREIBM-UHFFFAOYSA-N 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 2
- 238000012806 monitoring device Methods 0.000 claims 1
- 238000011426 transformation method Methods 0.000 abstract description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000005856 abnormality Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 239000003446 ligand Substances 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0259—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterized by the response to fault detection
- G05B23/0267—Fault communication, e.g. human machine interface [HMI]
- G05B23/027—Alarm generation, e.g. communication protocol; Forms of alarm
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B19/00—Programme-control systems
- G05B19/02—Programme-control systems electric
- G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]
- G05B19/4183—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM] characterised by data acquisition, e.g. workpiece identification
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/31—From computer integrated manufacturing till monitoring
- G05B2219/31439—Alarms can be warning, alert or fault
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Human Computer Interaction (AREA)
- General Engineering & Computer Science (AREA)
- Manufacturing & Machinery (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to an industrial alarm design method based on multi-objective optimization and evidence iterative update, and belongs to the technical field of industrial alarm design. The method provided by the invention comprises the steps of firstly converting a historical data set into alarm evidence by using a cast-on-point mapping transformation method, updating the reliability of the evidence in real time by using a heuristic criterion library, fusing the alarm evidence by using a fusion rule, performing off-line optimization on evidence weight and an input reference value by using a multi-dimensional swarm algorithm, and judging whether to give an alarm or not under a judgment criterion. The invention can effectively reduce the influence of uncertainty and improve the accuracy of the alarm.
Description
Technical Field
The invention relates to an industrial alarm design method based on multi-objective optimization and evidence iterative update, and belongs to the technical field of industrial alarm design.
Background
Nowadays, with the rapid development of high and new technologies, the informatization promotes the rapid development of industrialization, industrialization and modernization of enterprises, and all industries are accelerating the pace of informatization. The field of information communication exchange is also expanding continuously, covering all levels of the operation process of each workshop, factory and equipment, and meanwhile, due to the complex and complicated industrial production process, the information communication exchange as key equipment in the production process naturally becomes a key research object. How to monitor the running state of the equipment in real time and timely and quickly find the abnormality of the equipment by using an advanced information processing method so as to ensure the normal running of the production process is the primary problem to be solved by related scientific research and technical personnel.
Large industrial equipment consists of thousands of sensors, actuators, control loops and the like, and any component abnormality can cause failure, so that the equipment runs less efficiently, and irreparable failure or major accident occurs. Whether a traditional industrial alarm gives an alarm or not depends on whether a monitored alarm variable triggers a set alarm threshold or not, and the mechanism for triggering alarm generation through a single threshold often causes the situations of false alarm, missing alarm and the like in practical application, so that a worker cannot accurately judge the real condition of equipment, and the equipment fails. The basic indicators that typically measure the performance of an alarm are the False Alarm Rate (FAR), the false alarm rate (MAR), and the average delay time (AAD). The traditional alarm method adopts a mechanism that the process variable is immediately alarmed when exceeding the threshold value and is immediately relieved when being lower than the threshold value, but the traditional method cannot well process various uncertainties in consideration of other uncertain influence factors and actual complex running conditions, so that the performance of the alarm is deteriorated.
Disclosure of Invention
The invention aims to provide an industrial alarm design method based on multi-objective optimization and evidence iterative update, which is different from an absolute threshold alarm judgment mode adopted in a traditional alarm design method.
The invention comprises the following steps:
(1) setting an identification frame of the alarm as Θ ═ NA, a }, where NA ═ 0 denotes that the equipment is in a normal operation state, and a ═ 1 denotes that the equipment is in an abnormal operation state.
(2) The input of the alarm is x, wherein x (t), t is 1,2,3 …, the input is an online sequence of sensor monitoring equipment, t is a sampling time, the number of samples is determined by the monitoring period of the alarm and monitoring computer equipment, and x isminIs the minimum value of the input, xmaxTo be transportedIf the maximum value is entered, the input reference value set is U ═ Un1,2, N, where xmin=U1<U2<…<UN=xmaxN is the number of reference values of the alarm input x (t); the output of the alarm is the running state of the equipment, and is recorded as y (t), and the reference value set is V ═ V { (t)m1,2}, where V is1=NA=0,V2=A=1。
(3) A historical data set of a measurement sequence of X is obtained by the sensor as a training sample, denoted as X ═ { X (T), T ═ 1,2,3, … T }, and known to be L thereof1One is the measured value of the equipment in the normal operation state, and the corresponding output y (t) is 0 and has L2The measured value is in the abnormal operation state of the equipment, and the corresponding output y (t) is 1 and satisfies L1+L2T sample vectors are represented as a set of sample vectors S ═ x (T), y (T)]Then respectively converting the forms into corresponding similarity forms of reference values, and the specific steps are as follows:
(3-1) sample vector of sample set [ x (t), y (t)]Input value x (t) and reference value U ofnIs distributed like the degree of similarity, i.e. the degree of similarity is
Ru(x(t))={(Un,αn)|n=1,...,N} (1a)
Wherein
αn'=0 n'=1,...,N,n'≠n,n+1 (1c)
αnIndicating that the input value x (t) is at the reference value UnSimilarity is as follows.
(3-2) calculating a sample vector [ x (t), y (t)]The output value y (t) and the reference value VmHas a similarity distribution of
Rv(y(t))={(Vm,βm)|m=1,2} (2a)
Wherein
βm'=0 m'=1,...,M,m'≠m,m+1 (2c)
βmIndicating that the output value y (t) is at the reference value VmSimilarity is as follows.
(3-3) sample set sample vector [ x (t), y (t)]Converting the input and output matching relation of the step (3-1) and the step (3-2) into a corresponding similarity distribution form (alpha)nβm,αn+1βm,αnβm+1,αn+1βm+1) Wherein α isnβmRepresenting a sample vector [ x (t), y (t)]In each case x (t) is a reference value U at the input valuenThen, y (t) is at the output reference value VmThe following overall similarity.
(4) According to step (3), all samples in the sample vector set S are converted into a form of comprehensive similarity, and then a projection mapping matrix between the output reference value and the input reference value is constructed, as shown in the following table 1, wherein epsilonm,nRepresents all sample vectors [ x (t), y (t)]The intermediate sample input value x (t) matches the reference value UnThe sample output value y (t) matches the reference value VmThe sum of the comprehensive similarity degrees of the above-mentioned components,representing all input values x (t) against a reference value UnThe sample vectors of (a) are integrated with the sum of similarity degrees,representing all output values y (t) versus the reference value VmIs a sample vector of the sum of similarity degrees, and has
TABLE 1 projection mapping matrix of samples [ x (t), y (t) ]
(5) According to the introduction in the step (4)The point mapping matrix is processed by likelihood normalization to obtain the corresponding reference value U when the input value x (t) corresponds tonThe output value y (t) corresponds to the reference value VmEvidence of (A) is
Thus, an alarm evidence matrix distribution table as shown in table 2 can be constructed to describe the relationship between the inputs x (t) and the outputs y (t);
TABLE 2 alarm evidence distribution Table of inputs x (t)
(6) The online acquisition of the input variable x (t), t 1,2,3, …, is bound to a certain interval [ U [ ]n-1,Un]And thus two corresponding alarm evidencesIs activated so that the alarm evidence of the alarm at that moment can be obtained in the form of a weighted sum of the alarm evidence
et={(Vm,pm,t),m=1,2} (5a)
pm,t=αn-1θm,n-1+αnθm,n (5b)
(7) T (t) can be obtained according to the step (6)>1) Evidence of alarm e at a time with respect to input x (t)tThen e is fused by using the fusion ruletWith the global alarm evidence obtained at the time of history tRow fusion to obtain global alarm evidence E at time tt={(Vm,pm,t→e(2)) Where m is 1,2, where p ism,t→e(2)Representing the probability of the two alarm evidences after fusion at the time t, and establishing a heuristic criterion library to update the reliability r of each newly input alarm evidence on line at the time t more than or equal to 3t,iTwo evidences weight w per time instantt,iFor a fixed value, i is 1,2, the specific steps are as follows:
(7-1) when t is 1, the moment can be obtained through the step (6) and is also the global alarm evidence E1=e1={(Vm,pm,1→e(2)),m=1,2}。
(7-2) when t is 2, the alarm evidence e at the moment can be obtained through the step (6)2={(Vm,pm,2) M is 1,2, and two evidential weights w are set2,i1, reliability r2,iAnd (2) fusing the alarm evidences at the two moments by using a fusion rule to obtain a global alarm evidence at the moment t-2
E2={(Vm,pm,2→e(2)),m=1,2} (6a)
Wherein
qm,i=w2,ipm,i,i=1,2 (6c)
(7-3) when t is more than or equal to 3, the alarm evidence e can be obtainedt={(Vm,pm,t) M 1,2, and a two-evidence weight wt,iInvariable, rt,1And (5) updating the reliability r of the newly-entered alarm evidence on line by using a heuristic criterion library as 0.9t,2Then, the global alarm evidence is obtained by fusing the formulas (6-a), (6-b) and (6-c), and the specific updating steps are as follows:
(7-3-1) calculation of evidence E Using Euclidean distancet-2、Et-1And etA distance between each, wherein Et-2={(Vm,pm,(t-2)→e(2)),m=1,2},Et-1={(Vm,pm,(t-1)→e(2)),m=1,2},et={(Vm,pm,t) And m is 1,2}, then
Then D is12,D13,D23Respectively using functionsMapping to an interval [0,1 ]]Is above d12,d13,d23Support degree of
(7-3-2) constructing a non-linear mapping between the input and output of the heuristic rule library, wherein the input is (Aid (E)t-2),Aid(Et-1),Aid(et) Output is r)t,2Input set of reference valuesi represents the ith input variable, JiFor the number of reference values of the ith input variable, the output reference value set B ═ BzAnd l Z is 1,2, the distance Z, and Z is the number of output reference values.
(7-3-3) constructing a heuristic criterion library, wherein the heuristic criterion library is composed of K criteria, and the L-th criterion in the heuristic criterion library can be described as follows:
(7-3-4) sample input (Aid (E) is acquired at time tt-2),Aid(Et-1),Aid(et) As input to the model built, is denoted as [ Aid1,Aid2,Aid3]And obtaining the estimated output r of the new alarm evidence at the time t through criterion reasoningt,2The method comprises the following specific steps:
calculating AidiSet of reference values relative to an inputEach input necessarily belongs to an interval of the reference value set, and the similarity is:
the similarity of the remaining reference values is 0.
According to input amount AidiAnd (9a), (9b) determining the activated criteria and calculating the activation weight of the L-th criteria
In the above formula, wL∈[0,1],L=1,2,...,K,ψL∈[0,1]K, is the weight of the lth criterion,is the ith input Aid in the L-th criterioniRelative to the similarity of the reference value, it is converted by inputting the matching information.
Obtaining the activation weight omega of each criterion according to the step IILThen, each criterion confidence m to be activatedz,LThe fusion was carried out and the fusion result obtained was recorded asThe fusion formula is as follows:
making a decision according to the step III, and correspondingly outputting at the moment t as follows:
(7-3-5) fusing the evidences according to (6a), (6b) and (6c) to obtain a global alarm evidence E at the t (t is more than or equal to 3) momentt={(Vm,pm,t→e(2)),m=1,2}。
(8) Obtaining the global alarm evidence E of each moment according to the step (7)t={(Vm,pm,t→e(2)) Where m is 1,2, or p1,t→e(2)≥p2,t→e(2)If so, indicating that the equipment is in a normal state at the moment t; if p is1,t→e(2)<p2,t→e(2)If the device is in an abnormal state at the moment t, the alarm is required.
(9) The method comprises the following steps of performing offline optimization on alarm parameters based on an improved artificial bee colony algorithm:
(9-1) determining an optimization parameter set P ═ { U ═ Un,wi|i=1,2;n=2,...,N-1},UnReference value, w, representing input variable x (t)iWeight of the alarm evidence is represented, and the constraint condition is
(9-2) initializing a bee colony, setting the total number Size of the bee colony, setting the leading bee and the observation bee to respectively account for half of the total number, setting the global search frequency as Iter, the maximum global search frequency as maxim and the maximum Limit frequency of honey source stay as Limit, and generating an initial solution space by using the following formula
Solvei,j=Solvemin,j+λ(Solvemax,j-Solvemin,j) (14)
Where i represents the ith set of solutions generated, j represents the number of parameters that need to be optimized, and λ is a random number between (0, 1).
Then the search bees perform global random search within the solution range, and the profitability is evaluated through a fitness function of the following formula
Wherein, the fixness (i) indicates that the ith honey source corresponds to the solutioniDegree of return of (f)(s)i) Represents the ith honey source corresponding to the solutioniThe objective function value of (1).
And (9-3) obtaining the income degree of the honey source according to the step (9-2), wherein the top 50% of the income degree rank becomes a leading bee, the back 50% of the income degree rank becomes an observation bee, and the observation bee waits for the honey source information of the leading bee to follow.
(9-4) leading the bees to perform multidimensional search around the honey source by using the following formula to generate a new honey source
new_solvei,j=solvei,j+μ(solvei,j-solvek,j) (16a)
In the formula, k belongs to {1,2,. multidot.,. Size }, k is not equal to i, j represents the number of parameters needing to be optimized, and mu is a random number between (-1, 1); if the newly generated honey source is in the solution space range, continuing searching, if the newly generated honey source is not in the solution space range, repeatedly using the formula (16a) to generate a new honey source until the generated new honey source is in the solution space range, evaluating the profitability of the new honey source, and using the following formula to achieve the purpose of selecting the optimal honey source;
(9-5) calculating the probability of the observation bees to turn into the leading bees to adopt the honey source, wherein the probability formula of each honey source to be selected is as follows:
in the formula, Select represents the probability of selecting a honey source, and fitness represents a fitness function;
(9-6) if an old honey source stays more than Limit times, updating is performed by the following formula
Solvei=Solvemin+λ(Solvemax-Solvemin) (18)
λ is a random number between (0, 1);
the iteration condition is met, and a global optimal parameter set P is recorded according to the profitability; an equipment monitoring sensor collects input characteristic signals, the input characteristic signals are preprocessed in the step (1) and the step (2), and then the steps (3) to (8) are repeatedly utilized to obtain a global alarm evidence, so that alarm decision is made.
The invention provides an industrial alarm design method based on multi-objective optimization and evidence iterative update, which comprises the steps of converting training samples obtained from historical data into alarm evidence by using a projection point mapping transformation method, updating the reliability of the evidence in real time by using a heuristic criterion library, fusing the alarm evidence at the current moment and the alarm evidence at the previous moment by using a fusion rule, performing offline optimization on weights and input reference values, and judging whether to send an alarm or not under a judgment criterion, thereby effectively reducing the influence of uncertainty factors and improving each performance index of the alarm. The program (compiling environment LabVIEW, C + + and the like) compiled according to the method can run on a monitoring alarm computer, and is combined with hardware such as a sensor, a data collector, a data storage and the like to form an online alarm system, so that the real-time alarm function of the running state of the monitoring equipment is realized.
Drawings
FIG. 1 is a block flow diagram of the method of the present invention;
FIG. 2 is a training sample sequence of x in an embodiment of the method of the present invention;
FIG. 3 is a sequence of test samples for x in an embodiment of the method of the present invention.
Detailed description of the invention
The invention adopts an evidence fusion technology different from the traditional alarm, updates the reliability of the alarm evidence containing uncertainty obtained at the current moment and the global alarm evidence at all previous moments in real time by using a heuristic criterion library, fuses the alarm evidence by using a fusion rule, and performs off-line optimization on other related parameters, thereby reducing the influence of uncertainty factors on the alarm performance in practice, wherein the flow chart is shown in figure 1 and comprises the following steps:
(1) setting an identification frame of the alarm as Θ ═ NA, a }, where NA ═ 0 denotes that the equipment is in a normal operation state, and a ═ 1 denotes that the equipment is in an abnormal operation state.
(2) The input of the alarm is x, wherein x (t), t is 1,2,3 …, the input is an online sequence of sensor monitoring equipment, t is a sampling time, the number of samples is determined by the monitoring period of the alarm and monitoring computer equipment, and x isminIs the minimum value of the input, xmaxThe maximum value of the input is set as the input reference value U ═ Un1,2, N, where xmin=U1<U2<…<UN=xmaxN is the number of reference values of the alarm input x (t); the output of the alarm is the running state of the equipment, and is recorded as y (t), and the reference value set is V ═ V { (t)m1,2}, where V is1=NA=0,V2=A=1。
For ease of understanding, the following is illustrated with the sample sequence in fig. 2. If a sample vector is obtained from historical data to form a sample set, after the data in the sample set is preprocessed through the steps, the variation range of the obtained input x (t) is [15,150], and the corresponding output is a discrete value 0 and 1, so that an output reference value set V is {1,2}, and M is 2; the reference value set U of input x (t) is {15,30,45,60,70,80,85,90,95,100,110,120,130,150}, and N is 14.
(3) Using sensorsA historical data set of a measurement sequence of X is obtained as a training sample, denoted X ═ X (T), T ═ 1,2,3, … T, and known to include L therein1One is the measured value of the equipment in the normal operation state, and the corresponding output y (t) is 0 and has L2The measured value is in the abnormal operation state of the equipment, and the corresponding output y (t) is 1 and satisfies L1+L2T sample vectors are represented as a set of sample vectors S ═ x (T), y (T)]Then respectively converting the forms into corresponding similarity forms of reference values, and the specific steps are as follows:
this is exemplified here for ease of understanding. The data T-4000 shown in fig. 2 is used as training samples, and arranged into a sequence X, where L is known13000 measurements are taken while the device is in normal operation, corresponding to an output y (k) of 0; l is21000 measurements are taken when the plant is in an abnormal operating state, corresponding to an output y (k) of 1; then there is L1+L2=T=4000。
The method comprises the following steps of representing T sample vectors in a sample vector set S ═ x (T), y (T), and then respectively converting the T sample vectors into corresponding reference value similarity forms:
(3-1) sample vector of sample set [ x (t), y (t)]Input value x (t) and reference value U ofnHas a similarity distribution of
Ru(x(t))={(Un,αn)|n=1,...,N} (1a)
Wherein
αn'=0 n'=1,...,N,n'≠n,n+1 (1c)
αnIndicating that the input value x (t) is at the reference value UnSimilarity of the following;
(3-2) calculating a sample vector [ x (t), y (t)]The output value y (t) and the reference value VmHas a similarity distribution of
Rv(y(t))={(Vm,βm)|m=1,2} (2a)
Wherein
βm'=0 m'=1,...,M,m'≠m,m+1 (2c)
βmIndicating that the output value y (t) is at the reference value VmSimilarity is as follows.
(3-3) sample set sample vector [ x (t), y (t)]Converting the input and output matching relation of the step (3-1) and the step (3-2) into a corresponding similarity distribution form (alpha)nβm,αn+1βm,αnβm+1,αn+1βm+1) Wherein α isnβmRepresenting a sample vector [ x (t), y (t)]In each case x (t) is a reference value U at the input valuenThen, y (t) is at the output reference value VmThe following overall similarity.
To deepen the sample vector [ x (t), y (t)]Assuming that a sample vector (x (t), y (t)) is (86,0), the degree of similarity of the input value x (t) to the reference value obtained from equations (1a) - (1c) is α6=0.4,α70.6; the similarity of the output value y (t) of the formulas (2a) to (2c) to the reference value is beta1=1,β2Further, a comprehensive similarity distribution (α) of the sample vectors (x (t), y (t)) can be obtained as 0nβm,αn+1βm,αnβm+1,αn+1βm+1)=(0.4,0.6,0,0)。
(4) According to step (3), all samples in the sample vector set S are converted into a form of comprehensive similarity, and then a projection mapping matrix between the output reference value and the input reference value is constructed, as shown in the following table 1, wherein epsilonm,nRepresents all sample vectors [ x (t), y (t)]The intermediate sample input value x (t) matches the reference value UnThe sample output value y (t) matches the reference value VmThe sum of the comprehensive similarity degrees of the above-mentioned components,representing all input values x (t) against a reference value UnSample vector synthesis ofThe sum of the degrees of identity is calculated,representing all output values y (t) versus the reference value VmIs a sample vector of the sum of similarity degrees, and has
TABLE 1 projection mapping matrix of samples [ x (t), y (t) ]
In order to facilitate understanding of the projection point mapping relationship shown in the above table, the sample set and the reference value set in step (2) are used, and the comprehensive similarity distribution of all T-4000 sample vectors (x (T), y (T)) in the sample set is obtained according to step (3), so that the projection point mapping relationship in the above table can be constructed, as shown in table 3 below.
TABLE 3 projection mapping matrix for sample vectors (x (t), y (t))
(5) According to the projection point mapping matrix in the step (4), the corresponding reference value U of the input value x (t) can be obtained through the processing of likelihood normalizationnThe output value y (t) corresponds to the reference value VmEvidence of (A) is
Thus, an alarm evidence matrix distribution table as shown in table 2 can be constructed to describe the relationship between the inputs x (t) and the outputs y (t);
TABLE 2 alarm evidence distribution Table of inputs x (t)
Continuing to use the projection mapping matrix of the input value x (t) in step (4), the above table is further understood by way of example. According to Table 3, the reference value U is obtained from the input values x (t) obtained from equations (3) and (4)1The corresponding alarm evidence at 90 is
Similarly, alarm evidence corresponding to other reference values can be obtained, and then an alarm evidence distribution of the input values x (t) can be constructed, as shown in Table 4
TABLE 4 alarm evidence distribution Table of inputs x (t)
(6) The online acquisition of the input variable x (t), t 1,2,3, …, is bound to a certain interval [ U [ ]n-1,Un]And thus two corresponding alarm evidencesIs activated so that the alarm evidence of the alarm at that moment can be obtained in the form of a weighted sum of the alarm evidence
et={(Vm,pm,t),m=1,2} (5a)
pm,t=αn-1θm,n-1+αnθm,n (5b)
(7) T (t) can be obtained according to the step (6)>1) Evidence of alarm e at a time with respect to input x (t)tThen e is fused by using the fusion ruletFusing with the global alarm evidence obtained at the t historical moment to obtain a global alarm evidence E at the t momentt={(Vm,pm,t→e(2)) Where m is 1,2, where p ism,t→e(2)Representing the probability of the two alarm evidences after fusion at the time t, and establishing a heuristic criterion library to update the reliability r of each newly input alarm evidence on line at the time t more than or equal to 3t,iTwo evidences weight w per time instantt,iFor a fixed value, i is 1,2, the specific steps are as follows:
(7-1) when t is 1, the moment can be obtained through the step (6) and is also the global alarm evidence E1=e1={(Vm,pm,1→e(2)),m=1,2}。
(7-2) when t is 2, the alarm evidence e at the moment can be obtained through the step (6)2={(Vm,pm,2) M is 1,2, and two evidential weights w are set2,i1, reliability r2,iAnd (2) fusing the alarm evidences at the two moments by using a fusion rule to obtain a global alarm evidence at the moment t-2
E2={(Vm,pm,2→e(2)),m=1,2} (6a)
Wherein
qm,i=w2,ipm,i,i=1,2 (6c)
(7-3) when t is more than or equal to 3, the alarm evidence e can be obtainedt={(Vm,pm,t) M 1,2, and a two-evidence weight wt,iInvariable, rt,1And (5) updating the reliability r of the newly-entered alarm evidence on line by using a heuristic criterion library as 0.9t,2Then, the global alarm evidence is obtained by fusing the formulas (6-a), (6-b) and (6-c), and the specific updating steps are as follows:
(7-3-1) calculation of evidence E Using Euclidean distancet-2、Et-1And etA distance between each, wherein Et-2={(Vm,pm,(t-2)→e(2)),m=1,2},Et-1={(Vm,pm,(t-1)→e(2)),m=1,2},et={(Vm,pm,t) And m is 1,2}, then
Then D is12,D13,D23Respectively using functionsMapping to an interval [0,1 ]]Is above d12,d13,d23Support degree of
(7-3-2) constructing a non-linear mapping between the input and output of the heuristic rule library, wherein the input is (Aid (E)t-2),Aid(Et-1),Aid(et) Output is r)t,2Input set of reference valuesi represents the ith input variable, JiFor the number of reference values of the ith input variable, the output reference value set B ═ BzAnd l Z is 1,2, the distance Z, and Z is the number of output reference values.
(7-3-3) constructing a heuristic criterion library which is composed of K criteria, wherein the L-th criterion in the heuristic criterion library can be described as
(7-3-4) sample input (Aid (E) is acquired at time tt-2),Aid(Et-1),Aid(et) As input to the model built, is denoted as [ Aid1,Aid2,Aid3]And obtaining the estimated output r of the new alarm evidence at the time t through criterion reasoningt,2The method comprises the following specific steps:
calculating AidiSet of reference values relative to an inputEach input necessarily belongs to an interval of the reference value set, and the similarity is
The similarity of the remaining reference values is 0.
According to input amount AidiAnd (9a), (9b) determining the activated criteria and calculating the activation weight of the L-th criteria
In the above formula, wL∈[0,1],L=1,2,...,K,ψL∈[0,1]K, is the weight of the lth criterion,is the ith input Aid in the L-th criterioniSimilarity with respect to a reference value by inputting a matchThe ligand information is converted.
Obtaining the activation weight omega of each criterion according to the step IILThen, each criterion confidence m to be activatedz,LThe fusion was carried out and the fusion result obtained was recorded asThe fusion formula is as follows:
making decision according to the step (c), and correspondingly outputting as
To enhance the understanding of the heuristic criteria library inference fusion in step (7-3), this is exemplified herein. The heuristic rule base is a three-input one-output model and is provided for experts. For example, (1.2,1.2,1) is input and matching is performed according to the semantic reference value of table (5).
TABLE 5 semantic and reference values for inputs and outputs
Thus, the 21 st and 22 nd criteria can be activated
TABLE 6 activation of corresponding criteria
21 | PS^PS^VS | {(VS,0.35)(PS,0.65)(PM,0)(L,0)} |
22 | PS^PS^PS | {(VS,0)(PS,0.72)(PM,0.28)(L,0)} |
The activation weights of the two criteria obtained from equations (9) and (10) are ω21=0.25,ω220.75, and then the fusion result is { (VS, 0.0875) (PS, 0.7025) (PM, 0.21) (L, 0) } according to equation (11), then the corresponding output is r ═ 0.612 according to equation (12);
(7-3-5) fusing the evidences according to (6a), (6b) and (6c) to obtain a global alarm evidence E at the t (t is more than or equal to 3) momentt={(Vm,pm,t→e(2)),m=1,2};
This is exemplified here in order to enhance the understanding of step (7). First, suppose that the 3 measured values x (t) of the 3 time instants, where t is 1,2,3, x (1) is 105.55, x (2) is 77.37, and x (3) is 90.21, the alarm evidences about x (t) are calculated sequentially in step (6), as shown in the table:
TABLE 7 Warning evidence of inputs x (t)
Time t | t=1 | t=2 | t=3 |
Evidence of alarm | e1=(0.226,0.774) | e2=(0.825,0.1752) | e3=(0.597,0.403) |
According to the step (7), the global alarm evidences at 3 moments can be given as follows:
when t is 1, as obtained according to step (8-1), E1=(0.226,0.774)。
When t is 2, according to step (8-2), take w1=w2=1,r1=r2Fusing E according to formula (6) as 11(0.226,0.774) and e2Get global alarm evidence E at time t 2 (0.825,0.1752)2=(0.578,0.422)。
When t is 3, take w1=w2=1,r1When r is 0.9, according to step (8-3), r can be obtained2Fuse E according to equation (6) at 0.7452(0.578,0.422) and e3Get global alarm evidence E at time t-3 (0.597,0.403)3=(0.617,0.383)。
(8) Obtaining the global alarm evidence E of each moment according to the step (7)t={(Vm,pm,t→e(2)) Where m is 1,2, or p1,t→e(2)≥p2,t→e(2)If so, indicating that the equipment is in a normal state at the moment t; if p is1,t→e(2)<p2,t→e(2)If the device is in an abnormal state at the moment t, the alarm is required.
In the above example, based on the global alarm evidence output at 3 times, the alarm output can be given according to step (7), as shown in table 8:
TABLE 8 alarm output
Time t | Global alert evidence Et | Alarm result |
t=1 | E1=(0.226,0.774) | Alarm (A) |
t=2 | E2=(0.578,0.422) | Alarm-free (NA) |
t=3 | E3=(0.617,0.383) | Alarm-free (NA) |
(9) The method comprises the following steps of performing offline optimization on alarm parameters based on an improved artificial bee colony algorithm:
(9-1) determining an optimization parameter set P ═ { U ═ Un,wi|i=1,2;n=2,...,N-1},UnReference value, w, representing input variable x (t)iWeight of the alarm evidence is represented, and the constraint condition is
(9-2) initializing a bee colony, setting the total number Size of the bee colony, setting the leading bee and the observation bee to respectively account for half of the total number, setting the global search frequency as Iter, the maximum global search frequency as maxim and the maximum Limit frequency of honey source stay as Limit, and generating an initial solution space by using the following formula
Solvei,j=Solvemin,j+λ(Solvemax,j-Solvemin,j) (14)
Where i represents the ith set of solutions generated, j represents the number of parameters to be optimized, λ is a random number between (0,1),
then the searching bees perform global random search within the solution range, and the profitability is evaluated through a fitness function of the following formula:
wherein, the fixness (i) indicates that the ith honey source corresponds to the solutioniDegree of return of (f)(s)i) Represents the ith honey source corresponding to the solutioniThe objective function value of (1).
To enhance the understanding of step (9-2), how to solve the fitness function is illustrated. The false alarm rate (MAR) and False Alarm Rate (FAR) are important indicators of the performance of the alarm. For example, if MAR is 0.4 and FAR is 0.2, then the profitability of formula (15) is 0.625, which means that the profitability approaches 1, indicating that the parameter is better.
And (9-3) obtaining the income degree of the honey source according to the step (9-2), wherein the top 50% of the income degree rank becomes a leading bee, the back 50% of the income degree rank becomes an observation bee, and the observation bee waits for the honey source information of the leading bee to follow.
(9-4) leading the bees to perform multidimensional search around the honey source by using the following formula to generate a new honey source
new_solvei,j=solvei,j+μ(solvei,j-solvek,j) (16a)
In the formula, k belongs to {1,2,. multidot.,. Size }, k is not equal to i, j represents the number of parameters needing to be optimized, and mu is a random number between (-1, 1); if the newly generated honey source is in the solution space range, the search is continued, if the newly generated honey source is not in the solution space range, the formula (16a) is repeatedly used to generate a new honey source until the generated new honey source is in the solution space range, the profitability of the new honey source is evaluated, and the purpose of selecting the best honey is achieved by using the following formula.
In order to enhance the understanding of step (9-4), an example will now be described. For example, the current parameter set P ═ 30,45,60,70,80,85,90,95,100,110,120,130,0.8,0.6}, searching for a new solution around the current set of parameters, randomly selecting two parameters around which to generate new parameter values using equation (16a), assuming a new set of parameters P1If the constraint condition is not met, the new parameter P needs to be generated by reusing (16a)2Assume a new parameter set P2The profitability of the parameter set is calculated and compared with the parameter set P by using equation (16b), thereby determining whether or not to substitute the parameter set P {30,45,60,70,80,85,90,95,100,110,120,130,0.9,0.5 }.
(9-5) calculating the probability of the observation bees to turn into the leading bees to adopt the honey source, wherein the probability formula of each honey source to be selected is as follows:
in the formula, Select represents the probability of selecting a honey source, and fitness represents a fitness function;
(9-6) if an old honey source stays more than Limit times, updating is performed by the following formula
Solvei=Solvemin+λ(Solvemax-Solvemin) (18)
λ is a random number between (0, 1);
the iteration condition is met, and a global optimal parameter set P is recorded according to the profitability; an equipment monitoring sensor collects input characteristic signals, the input characteristic signals are preprocessed in the step (1) and the step (2), and then the steps (3) to (8) are repeatedly utilized to obtain a global alarm evidence, so that alarm decision is made.
Embodiments of the method of the present invention are described in detail below with reference to the accompanying drawings:
the flow chart of the method of the invention is shown in figure 1, and the core part is as follows: after a sample data sequence needing to be monitored is determined, a training sample obtained from a historical data set is converted into an alarm evidence by using a cast-on-point mapping transformation method, the reliability of the evidence is updated in real time by using a heuristic criterion library, then the alarm evidence at the current moment and the alarm evidence at the previous moment are fused by using a fusion rule, the weight and an input reference value are optimized off-line, and whether an alarm is sent or not is judged under a judgment criterion, so that the influence of uncertainty factors is effectively reduced, and each performance index of the alarm is improved.
The following description will be given of the preferred embodiments of the present invention in detail with reference to the x (t) sample data shown in fig. 2.
1. Experimental data acquisition and preprocessing
The sample data sequence x (T) is shown in fig. 2, where T is 4000, and the sample data indicates that x varies within a range of [15,150 ]. The output corresponding to each time is y (t), and a sample vector set S ═ x (t), y (t) can be obtained.
2. Input x (t) and output y (t) selection of reference values.
After the data in the sample is preprocessed through the above steps, the range of variation of the input x (t) is [15,150], and the corresponding output is discrete values 0 and 1, so the output reference value set V ═ 1,2, and M ═ 2; the reference value set U of input x (t) is {15,30,45,60,70,80,85,90,95,100,110,120,130,150}, and N is 14.
3. And acquiring similarity forms of the sample vectors [ x (t), y (t) ] relative to the reference values, and constructing sample projection point mapping matrixes of the sample vectors [ x (t), y (t) ].
Selecting T-4000 group data from the historical data set as training samples, arranging the training samples into a sequence X, and confirming that L is included in the sequence X13000 measurements are taken while the device is in normal operation, corresponding to an output y (k) of 0; l is21000 measurements are taken when the plant is in an abnormal operating state, corresponding to an output y (k) of 1; then there is L1+L2T4000. The comprehensive similarity distribution of all T ═ 4000 sample vectors (x (T), y (T)) in the sample set is obtained, and the sample projection point mapping matrix shown in table 1 in step (4) of the method of the present invention can be constructed as shown in table 3 below. Input sample vector [ x (t), y (t)]The projection point mapping matrix of (a) is shown in the following table:
TABLE 9 projection mapping matrix for sample vector [ x (t), y (t) ]
4. According to the method, step (5) of the invention, alarm evidences corresponding to the input x (t) reference values are obtained, and an alarm evidence distribution table is constructed.
After the projection point mapping matrix is obtained according to the step (4), the evidences corresponding to the reference values are obtained according to the step (5), and an evidence distribution table is further constructed, as shown in the following table:
TABLE 10 alarm evidence distribution Table of inputs x (t)
5. According to the method, the alarm evidence at each moment is obtained in the step (6), and the evidence is fused in the step (7).
The 3 measured values x (t) of the new entry at the 3 moments given t ═ 1,2 and 3, x (1) ═ 105.55, x (2) ═ 77.37, and x (3) ═ 90.21 for x (t), are calculated sequentially in step (6), as shown in the table:
TABLE 11 Warning evidence of inputs x (t)
Time t | t=1 | t=2 | t=3 |
Evidence of alarm | e1=(0.226,0.774) | e2=(0.825,0.1752) | e3=(0.597,0.403) |
According to the method, the step (7) can give out the global alarm evidence at 3 moments as follows:
when t is 1, as obtained according to step (7-1), E1=(0.226,0.774);
When t is 2, according to step (7-2), take w1=w2=1,r1=r2Fusing E according to formula (6) as 11(0.226,0.774) and e2Get global alarm evidence E at time t 2 (0.825,0.1752)2=(0.578,0.422);
When t is 3, take w1=w2=1,r10.9, according to step (7-3) of the present invention, r can be obtained2Fuse E according to equation (6) at 0.7452(0.578,0.422) and e3Get global alarm evidence E at time t-3 (0.597,0.403)3=(0.617,0.383)。
6. Alarm decision
According to inventive step (8) the alarm output can be given as shown in table 12:
watch 12 alarm output
Time t | Global alert evidence Et | Alarm output |
t=1 | E1=(0.226,0.774) | Alarm (A) |
t=2 | E2=(0.578,0.422) | Alarm-free (NA) |
t=3 | E3=(0.617,0.383) | Alarm-free (NA) |
7. Training optimization
According to the step (9), a multi-dimensional search bee colony algorithm is adopted as an optimization algorithm, and an optimization parameter set is determined, wherein the initial parameter weight is w1=1,w2And (3) repeating the steps (3) to (8) to obtain the optimized parameters, wherein the optimized FAR is 2.5%, the optimized MAR is 2.1%, and the alarm performance is greatly improved.
8. Comparison with conventional alarm method
The test is performed according to the test sample sequence of fig. 3, and the method is compared with the conventional methods such as the time delay method and the digital sliding filtering method for the false alarm rate and the false negative rate under multiple random experiments, and the comparison of various alarm methods is shown in the following table 13
Method | False alarm rate | Rate of missing reports |
Digital sliding filter method (%) | 19.75 | 9.62 |
Time delay method (%) | 8.65 | 16.33 |
Alarm reliability fusion method (%) | 2.9 | 2.7 |
Claims (3)
1. The industrial alarm design method based on multi-objective optimization and evidence iterative update is characterized by comprising the following steps:
(1) setting an identification frame of the alarm as Θ ═ NA, a }, wherein NA ═ 0 represents that the equipment is in a normal operation state, and a ═ 1 represents that the equipment is in an abnormal operation state;
(2) let x be the input of the alarm, where x (t) is the on-line sequence of the sensor monitoring device, t is the sampling time, the number of samples is determined by the monitoring period of the alarm and the monitoring computer device, xminIs the minimum value of the input, xmaxThe maximum value of the input is set as the input reference value U ═ Un1,2, N, where xmin=U1<U2<…<UN=xmaxN is the number of reference values of the alarm input x (t); the output of the alarm is the running state of the equipment, and is recorded as y (t), and the reference value set is V ═ V { (t)m1,2}, where V is1=NA=0,V2=A=1;
(3) A historical data set of a measurement sequence of X is obtained by the sensor as a training sample, denoted as X ═ { X (T), T ═ 1,2,3, … T }, and known to be L thereof1One is the measured value of the equipment in the normal operation state, and the corresponding output y (t) is 0 and has L2Is measured when the equipment is in abnormal operation stateMagnitude, corresponding to output y (t) of 1, satisfying L1+L2T sample vectors are represented as a set of sample vectors S ═ x (T), y (T)]Then respectively converting into corresponding comprehensive similarity forms;
(4) all samples in the sample vector set S are converted into a form of comprehensive similarity, and then a projection mapping matrix between an output reference value and an input reference value is constructed, as shown in the following table 1, wherein epsilonm,nRepresents all sample vectors [ x (t), y (t)]The intermediate sample input value x (t) matches the reference value UnThe sample output value y (t) matches the reference value VmThe sum of the comprehensive similarity degrees of the above-mentioned components,representing all input values x (t) against a reference value UnThe sample vectors of (a) are integrated with the sum of similarity degrees,representing all output values y (t) versus the reference value VmIs a sample vector of the sum of similarity degrees, and has
TABLE 1 projection mapping matrix of samples [ x (t), y (t) ]
(5) According to the projection point mapping matrix in the step (4), a corresponding reference value U of the input value x (t) is obtained through likelihood normalization processingnThe output value y (t) corresponds to the reference value VmEvidence of (A) is
Constructing an alarm evidence matrix distribution table as shown in table 2 to describe the relationship between the input x (t) and the output y (t);
TABLE 2 alarm evidence distribution Table of inputs x (t)
(6) On-line acquisition of an input variable x (t), which must belong to a certain interval [ Un-1,Un]And thus two corresponding alarm evidencesActivated, the alarm evidence of the alarm at the moment can be obtained in the form of weighted sum of the alarm evidence
et={(Vm,pm,t),m=1,2}
pm,t=αn-1θm,n-1+αnθm,n
Wherein alpha isnIndicating that the input variable x (t) is at the reference value UnDegree of similarity of degree of expression, αn-1Indicating that the input variable x (t) is at the reference value Un-1Similarity of the following;
(7) obtaining alarm evidence e about input x (t) at time t according to step (6)tThen e is fused by using the fusion ruletFusing with the global alarm evidence obtained at the t historical moment to obtain a global alarm evidence E at the t momentt={(Vm,pm,t→e(2)) Where m is 1,2, where p ism,t→e(2)Representing the probability of the two alarm evidences after fusion at the time t, and establishing a heuristic criterion library to update the reliability r of each newly input alarm evidence on line at the time t more than or equal to 3t,iTwo evidences weight w per time instantt,iIs a constant value;
(8) obtaining the global alarm evidence E of each moment according to the step (7)t={(Vm,pm,t→e(2)) Where m is 1,2, or p1,t→e(2)≥p2,t→e(2)If so, indicating that the equipment is in a normal state at the moment t; if p is1,t→e(2)<p2,t→e(2)If the device is in an abnormal state at the moment t, the alarm is required;
wherein the step (3) is specifically as follows:
(3-1) sample vector of sample set [ x (t), y (t)]Input value x (t) and reference value U ofnSimilarity distribution of (d):
Ru(x(t))={(Un,αn)|n=1,...,N}
wherein
αn'=0 n'=1,...,N,n'≠n,n+1
αnIndicating that the input value x (t) is at the reference value UnSimilarity of the following;
(3-2) calculating a sample vector [ x (t), y (t)]The output value y (t) and the reference value VmHas a similarity distribution of
Rv(y(t))={(Vm,βm)|m=1,2}
Wherein
βm'=0 m'=1,...,M,m'≠m,m+1
βmIndicating that the output value y (t) is at the reference value VmSimilarity of the following;
(3-3) sample set sample vector [ x (t), y (t)]Converting the input and output matching relation of the step (3-1) and the step (3-2) into a corresponding similarity distribution form (alpha)nβm,αn+1βm,αnβm+1,αn+1βm+1) Wherein α isnβmRepresenting a sample vector [ x (t), y (t)]In each case x (t) is a reference value U at the input valuenThen, y (t) is at the output reference value VmComprehensive similarity of the following;
wherein the step (7) is specifically as follows:
(7-1) when t is 1, obtaining the global alarm evidence E at the moment through the step (6)1=e1={(Vm,pm,1→e(2)),m=1,2};
(7-2) when t is 2, obtaining the alarm evidence e at the moment through the step (6)2={(Vm,pm,2) M is 1,2, and two evidential weights w are set2,i1, reliability r2,iAnd (2) fusing the alarm evidences at the two moments by using a fusion rule to obtain a global alarm evidence at the moment t-2
E2={(Vm,pm,2→e(2)),m=1,2}
Wherein
qm,i=w2,ipm,i,i=1,2
(7-3) when t is more than or equal to 3, the alarm evidence e can be obtainedt={(Vm,pm,t) M 1,2, and a two-evidence weight wt,iInvariable, rt,1And (5) updating the reliability r of the newly-entered alarm evidence on line by using a heuristic criterion library as 0.9t,2Then, fusion is carried out to obtain global alarm evidenceThe method specifically comprises the following steps:
(7-3-1) calculation of evidence E Using Euclidean distancet-2、Et-1And etA distance between each, wherein Et-2={(Vm,pm,(t-2)→e(2)),m=1,2},Et-1={(Vm,pm,(t-1)→e(2)),m=1,2},et={(Vm,pm,t) And m is 1,2}, then
Then D is12,D13,D23Respectively using functionsMapping to an interval [0,1 ]]Is above d12,d13,d23Support degree of
Aid(Et-2)=1-d12+1-d13=2-d12-d13
Aid(Et-1)=1-d12+1-d23=2-d12-d23
Aid(et)=1-d13+1-d23=2-d13-d23
(7-3-2) constructing a non-linear mapping between the input and output of the heuristic rule library, wherein the input is (Aid (E)t-2),Aid(Et-1),Aid(et) Output is r)t,2Input set of reference valuesi represents the ith input variable, JiFor the number of reference values of the ith input variable, the output reference value set B ═ Bz1,2, a, Z, wherein Z is the number of output reference values;
(7-3-3) constructing a heuristic criterion library which is composed of K criteria, wherein the L-th criterion in the heuristic criterion library can be described as
Then{(B1,m1,L),(B2,m2,L),...,(BZ,mZ,L)}
(7-3-4) sample input (Aid (E) is acquired at time tt-2),Aid(Et-1),Aid(et) As input to the model built, is denoted as [ Aid1,Aid2,Aid3]And obtaining the estimated output r of the new alarm evidence at the time t through criterion reasoningt,2;
(7-3-5) fusing the evidence according to the formula in (7-2) to obtain the global alarm evidence E at the moment t is more than or equal to 3t={(Vm,pm,t→e(2)),m=1,2}。
2. The industrial alarm design method based on multi-objective optimization and evidence iterative update as claimed in claim 1, characterized in that: the step (7-3-4) is specifically:
calculating AidiSet of reference values relative to an inputEach input necessarily belongs to an interval of the reference value set, and the similarity is
Similarity of the other reference values is 0;
according to input amount AidiThe formula in (1) determines the activated criterion and calculates the activation weight of the L-th criterion
In the above formula, wL∈[0,1],L=1,2,...,K,ψL∈[0,1]K, is the weight of the lth criterion,is the ith input Aid in the L-th criterioniSimilarity relative to the reference value, which is converted by inputting matching information;
obtaining the activation weight w of each criterionLThen, each criterion confidence m to be activatedz,LThe fusion was carried out and the fusion result obtained was recorded asThe fusion formula is as follows:
making decision according to the third step, and correspondingly outputting the result at the time t
3. The industrial alarm design method based on multi-objective optimization and evidence iterative update as claimed in claim 1, characterized in that: the method further comprises the step of carrying out off-line optimization on the alarm parameters based on the improved artificial bee colony algorithm, and the method comprises the following specific steps:
(9-1) determining an optimization parameter set P ═ { U ═ Un,wi|i=1,2;n=2,...,N-1},UnReference value, w, representing input variable x (t)iWeight of the alarm evidence is represented, and the constraint condition is
U1<U2<...<UN-1<UN
0<w1<1,0<w2<1
(9-2) initializing the bee colony, setting the total number Size of the bee colony, setting the leading bee and the observation bee to respectively account for half of the total number, setting the global search frequency as Iter, the maximum global search frequency as maxim and the maximum Limit frequency of honey source stay as Limit, and generating an initial solution space by using the following formula
Solvei,j=Solvemin,j+λ(Solvemax,j-Solvemin,j)
Wherein i represents the generated ith group of solutions, j represents the number of parameters needing to be optimized, and lambda is a random number between (0, 1);
the search bees perform global random search within the solution range, and the profitability is evaluated through a fitness function of the following formula:
wherein, the fixness (i) indicates that the ith honey source corresponds to the solutioniDegree of return of (f)(s)i) Represents the ith honey source corresponding to the solutioniThe objective function value of (1);
(9-3) obtaining the income degree of the honey source according to the step (9-2), wherein the income degree is ranked 50% and becomes a leading bee, the income degree is ranked 50% and becomes an observation bee, and the observation bee waits for honey source information of the leading bee to follow;
(9-4) leading the bees to perform multidimensional search around the honey source by using the following formula to generate a new honey source
new_solvei,j=solvei,j+μ(solvei,j-solvek,j)
In the formula, k belongs to {1,2,. multidot.,. Size }, k is not equal to i, j represents the number of parameters needing to be optimized, and mu is a random number between (-1, 1); if the newly generated honey source is in the solution space range, continuing searching, if the newly generated honey source is not in the solution space range, repeatedly generating a new honey source until the generated new honey source is in the solution space range, evaluating the profitability of the new honey source, and achieving the purpose of preference by using the following formula;
(9-5) calculating the probability of the observation bees to turn into the leading bees to adopt the honey source, wherein the probability formula of each honey source to be selected is as follows:
in the formula, Select represents the probability of selecting a honey source, and fitness represents a fitness function;
(9-6) if an old honey source stays more than Limit times, updating is performed by the following formula
Solvei=Solvemin+λ(Solvemax-Solvemin)
λ is a random number between (0, 1);
the iteration condition is met, and a global optimal parameter set P is recorded according to the profitability; an equipment monitoring sensor collects input characteristic signals, the input characteristic signals are preprocessed in the step (1) and the step (2), and then the steps (3) to (8) are repeatedly utilized to obtain a global alarm evidence, so that alarm decision is made.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010207522.4A CN111443686B (en) | 2020-03-23 | 2020-03-23 | Industrial alarm design method based on multi-objective optimization and evidence iterative update |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010207522.4A CN111443686B (en) | 2020-03-23 | 2020-03-23 | Industrial alarm design method based on multi-objective optimization and evidence iterative update |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111443686A CN111443686A (en) | 2020-07-24 |
CN111443686B true CN111443686B (en) | 2021-08-10 |
Family
ID=71648999
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010207522.4A Active CN111443686B (en) | 2020-03-23 | 2020-03-23 | Industrial alarm design method based on multi-objective optimization and evidence iterative update |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111443686B (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101630161A (en) * | 2009-08-05 | 2010-01-20 | 北方工业大学 | Intelligent system for complex industrial production and construction method thereof |
CN102789676A (en) * | 2012-08-10 | 2012-11-21 | 杭州电子科技大学 | Method for designing industrial alarm on basis of alarm evidence fusion |
CN108398934A (en) * | 2018-02-05 | 2018-08-14 | 常州高清信息技术有限公司 | The system that a kind of equipment fault for rail traffic monitors |
CN109145972A (en) * | 2018-08-09 | 2019-01-04 | 杭州电子科技大学 | A kind of watercraft electric propulsion system frequency converter alarm design method |
US10176706B2 (en) * | 2014-08-15 | 2019-01-08 | The Adt Security Corporation | Using degree of confidence to prevent false security system alarms |
CN110597232A (en) * | 2019-09-26 | 2019-12-20 | 杭州电子科技大学 | Frequency converter cooling water pump fault alarm method based on dynamic confidence rule base |
Family Cites Families (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102270271B (en) * | 2011-05-03 | 2014-03-19 | 北京中瑞泰科技有限公司 | Equipment failure early warning and optimizing method and system based on similarity curve |
CN105988459B (en) * | 2015-02-11 | 2019-01-18 | 中芯国际集成电路制造(上海)有限公司 | Method based on the small drift forecasting board failure of mean value |
CN104699077B (en) * | 2015-02-12 | 2017-06-06 | 浙江大学 | A kind of failure variable partition method based on nested iterations Fei Sheer discriminant analyses |
CN106199421B (en) * | 2016-06-27 | 2018-03-02 | 北京协同创新研究院 | A kind of method for early warning and system based on industrial big data |
JP2018132916A (en) * | 2017-02-15 | 2018-08-23 | 三菱電機株式会社 | Water treatment plant operation support system |
US10941980B2 (en) * | 2017-09-06 | 2021-03-09 | International Business Machines Corporation | Predictive maintenance of refrigeration cases |
CN108257365B (en) * | 2018-01-29 | 2020-04-24 | 杭州电子科技大学 | Industrial alarm design method based on global uncertainty evidence dynamic fusion |
CN109856488A (en) * | 2019-03-15 | 2019-06-07 | 长沙理工大学 | A kind of Transformer State Assessment and fault detection method based on multisource data fusion |
-
2020
- 2020-03-23 CN CN202010207522.4A patent/CN111443686B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101630161A (en) * | 2009-08-05 | 2010-01-20 | 北方工业大学 | Intelligent system for complex industrial production and construction method thereof |
CN102789676A (en) * | 2012-08-10 | 2012-11-21 | 杭州电子科技大学 | Method for designing industrial alarm on basis of alarm evidence fusion |
US10176706B2 (en) * | 2014-08-15 | 2019-01-08 | The Adt Security Corporation | Using degree of confidence to prevent false security system alarms |
CN108398934A (en) * | 2018-02-05 | 2018-08-14 | 常州高清信息技术有限公司 | The system that a kind of equipment fault for rail traffic monitors |
CN109145972A (en) * | 2018-08-09 | 2019-01-04 | 杭州电子科技大学 | A kind of watercraft electric propulsion system frequency converter alarm design method |
CN110597232A (en) * | 2019-09-26 | 2019-12-20 | 杭州电子科技大学 | Frequency converter cooling water pump fault alarm method based on dynamic confidence rule base |
Also Published As
Publication number | Publication date |
---|---|
CN111443686A (en) | 2020-07-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105467975B (en) | A kind of equipment fault diagnosis method | |
CN109814527B (en) | Industrial equipment fault prediction method and device based on LSTM recurrent neural network | |
CN109146246B (en) | Fault detection method based on automatic encoder and Bayesian network | |
CN116757534B (en) | Intelligent refrigerator reliability analysis method based on neural training network | |
Lindemann et al. | Anomaly detection and prediction in discrete manufacturing based on cooperative LSTM networks | |
CN110135630A (en) | The short term needing forecasting method with multi-step optimization is returned based on random forest | |
CN113505935B (en) | Electric power abnormal fluctuation detection and prediction calculation method based on integrated algorithm | |
CN110570012B (en) | Storm-based power plant production equipment fault early warning method and system | |
CN110457184B (en) | Chemical engineering abnormal cause and effect analysis and graph display method based on time sequence fluctuation correlation | |
CN105631596A (en) | Equipment fault diagnosis method based on multidimensional segmentation fitting | |
CN113107626B (en) | Load prediction method of combined cycle generator set based on multivariable LSTM | |
CN111525587B (en) | Reactive load situation-based power grid reactive voltage control method and system | |
CN109634233A (en) | Industrial big data intellectual analysis decision-making technique, readable storage medium storing program for executing and terminal | |
CN114881101B (en) | Bionic search-based power system typical scene association feature selection method | |
CN113325721A (en) | Model-free adaptive control method and system for industrial system | |
CN109614451A (en) | Industrial big data intellectual analysis decision making device | |
CN115128978A (en) | Internet of things environment big data detection and intelligent monitoring system | |
Zhang et al. | Remaining useful life prediction for rolling bearings with a novel entropy-based health indicator and improved particle filter algorithm | |
CN118114118A (en) | CNDT-based fault diagnosis method for typical weapon equipment | |
Wang et al. | An artificial immune and incremental learning inspired novel framework for performance pattern identification of complex electromechanical systems | |
CN113515891A (en) | Method for predicting and optimizing quality of emulsion explosive | |
CN111443686B (en) | Industrial alarm design method based on multi-objective optimization and evidence iterative update | |
Huang et al. | Distributed dictionary learning for industrial process monitoring with big data | |
CN116401545A (en) | Multimode fusion type turbine runout analysis method | |
CN116467592A (en) | Production equipment fault intelligent monitoring method and system based on deep learning |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |