CN111898313B - Fault detection method based on ICA and SVM integrated learning - Google Patents
Fault detection method based on ICA and SVM integrated learning Download PDFInfo
- Publication number
- CN111898313B CN111898313B CN202010612207.XA CN202010612207A CN111898313B CN 111898313 B CN111898313 B CN 111898313B CN 202010612207 A CN202010612207 A CN 202010612207A CN 111898313 B CN111898313 B CN 111898313B
- Authority
- CN
- China
- Prior art keywords
- sample
- model
- ica
- svm
- obtaining
- 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
Images
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
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2134—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on separation criteria, e.g. independent component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/04—Inference or reasoning models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N7/00—Computing arrangements based on specific mathematical models
- G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
-
- 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)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Software Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computing Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Evolutionary Biology (AREA)
- Computational Mathematics (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Analysis (AREA)
- Algebra (AREA)
- Probability & Statistics with Applications (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Computational Linguistics (AREA)
- Complex Calculations (AREA)
Abstract
The invention discloses a fault detection method based on ICA and SVM integrated learning, which integrates an ICA model with characteristic extraction capability and an SVM model with classification capability by adopting Bayesian inference and utilizes different advantages of different models, thereby improving detection accuracy and ensuring more stable detection effect.
Description
Technical Field
The invention relates to the field of fault detection in industrial processes, in particular to a fault detection method based on ICA and SVM integrated learning.
Background
Modern industrial production is more and more scaled and complicated, and if the production process breaks down, the product quality is influenced, and the life safety of people is threatened more easily. Fault detection techniques are therefore often employed to monitor industrial process conditions.
In the prior art, the measurement device is arranged to acquire the working data of the industrial production system to detect the fault in the industrial process, but the measurement device can only acquire the working data of the industrial production system and cannot directly draw a conclusion about whether the fault occurs, so how to perform fault detection according to the data acquired by the measurement device and improve the accuracy of fault detection become important research points in the field.
Disclosure of Invention
In order to overcome the defects in the prior art, the fault detection method based on ICA and SVM integrated learning provided by the invention can accurately and quickly detect faults through data acquired by measuring equipment.
In order to achieve the purpose of the invention, the invention adopts the technical scheme that:
a fault detection method based on ICA and SVM integrated learning is provided, which comprises the following steps:
s1, acquiring measurement data of the target industrial production system by the measurement equipment under different working conditions of the target industrial production system, listing the measurement data at different moments under the same working condition into a matrix, and taking each matrix as an initial sample set;
s2, standardizing the initial sample set under the normal working condition to obtain a standardized sample set under the normal working condition;
s3, constructing an ICA model based on the sample set under the standardized normal working condition;
s4, obtaining the number of SVM models to be constructed, randomly extracting samples from each initial sample set to form a sample subset of each working condition for each SVM model, and standardizing each sample subset to obtain a training subset corresponding to each SVM model;
s5, constructing a corresponding SVM model based on each training subset;
s6, acquiring measurement data of the current measurement equipment on the target industrial production system, and standardizing the measurement data to be used as a detection sample;
s7, respectively taking the detection samples as the input of all ICA models and SVM models, and correspondingly obtaining the output of each ICA model and the output of each SVM model;
s8, integrating the outputs of all ICA models and SVM models corresponding to the detection samples through Bayesian inference to calculate the integration probability value;
and S9, judging whether the integrated probability value is larger than or equal to a threshold value, if so, judging that the detection sample is a fault, otherwise, judging that the detection sample is normal, and finishing fault detection.
Further, the specific method of step S1 is:
the method comprises the steps that K measuring devices are adopted to simultaneously obtain measuring data of a target industrial production system under C working conditions, the measuring data corresponding to the K measuring devices at the same time are used as a sample, the measuring data at different times under the same working condition are listed into a matrix, and each matrix is used as an initial sample set to obtain C initial sample sets; i.e. each sample in each initial set of samples has K elements.
Further, the specific method of step S2 is:
for an initial sample set X under normal operating conditions0=[x0(1),x0(2),...,x0(r),...,x0(n0)]According to the formula:
obtaining the l-th sample of the r-th samplexNormalized value of individual elementFurther obtaining the values of all elements after standardization, completing the standardization of the initial sample set under the normal working condition, and obtaining the standardized sample set under the normal working condition; wherein x0(r)(lx) For the l-th sample in the r-th samplexA value of an element; mean (X)0(lx) Is the l-th sample of each sample in the initial sample set under normal conditionsxA mean of the individual elements; std (X)0(lx) Is the l-th sample of each sample in the initial sample set under normal conditionsxStandard deviation of individual elements; n is0The total number of samples in the initial sample set under the normal working condition.
Further, the specific method of step S3 includes the following sub-steps:
s3-1, sample set under normal working condition after standardizationWhitening to obtain a whitening transformation matrix QPACAnd according to the formula:
obtaining a whitening matrix Z;
s3-2, construction of QICAAn ICA model, for the ith ICA model, according to the formula:
Si=BiZ
different random number seeds are set, and a first demixing matrix B of the ith ICA model is obtained by adopting a FastICA algorithmiAnd independent vector matrix SiFurther, a first unmixing matrix and an independent vector matrix of each ICA model are obtained;
s3-3, according to the formula:
obtaining a second unmixing matrix W of the ith ICA modeliFurther obtaining a second unmixing matrix of each ICA model; wherein (·)TRepresents a transpose of a matrix;
s3-4, sequentially selecting d row vectors from the row vectors of the second unmixing matrix of the ith ICA model according to the sequence of vector norm from large to small to form a combined matrix W of the ith ICA modeld,iFurther obtaining a combination matrix of each ICA model;
s3-5, according to the formula:
obtaining a set of normalized samples under normal conditionsMiddle (r) th sampleBased on the ithStatistic value of ICA modelFurther obtaining a standardized sample set under normal working conditionsAll the samples in the statistical quantity combination based on the ith ICA modelAnd obtaining a standardized sample set under normal working conditionsAll samples in (a) are based on a combination of statistics for each ICA model; wherein n is0The total number of samples in the initial sample set under the normal working condition;
s3-6, obtaining statistic combination of ith ICA model by adopting kernel density estimation method KDEProbability density ofAnd according to the formula:
obtaining a solving interval [ start ] corresponding to the ith ICA modeli,endi]Further obtaining probability density functions and solving intervals of all ICA model statistic combinations; where min (-) denotes taking the minimum value and max (-) denotes taking the maximum value;
s3-7, solving interval [ start ] corresponding to ith ICA modeli,endi]Is equidistantly divided into numiIndividual subareaAnd according to the formula:
acquiring the number k of accumulated subintervals corresponding to the ith ICA modeliFurther obtaining the cumulative number of subintervals corresponding to each ICA model; wherein ΔiThe subinterval width corresponding to the ith ICA model;statistical combination representing ith ICA modelIn thatThe probability density of (d); α is the confidence of the control limit; xi is a constant;
s3-8, according to the formula:
UCLi=starti+kiΔi
obtaining the control limit UCL of the ith ICA modeliFurther obtaining the control limit of each ICA model;
and S3-9, regarding any ICA model, taking the ratio of the statistic value of the input sample to the control limit as the output of the ICA model, and completing the construction of the ICA model.
Further, the specific method of step S4 includes the following sub-steps:
s4-1, obtaining the quantity Q of SVM models to be constructedSVMFor the jth SVM model, m is randomly drawn from each initial sample set without replacementc,jEach sample constitutes a sample subset Y for each conditionc,j(ii) a Wherein m isc,j=int(nc,j×ratej) Int (·) denotes that only the integer part of the computation result is retained, nc,jIs the total number of samples in the initial sample set under the c-th working condition, when c is 0, the normal working condition is indicated, and ratejFor the extraction ratio corresponding to the jth SVM model, 0.0<ratej<1.0;
S4-2, according to the formula:
obtaining a sample subset Yc,jIth sample of (e)yNormalized value of individual elementFurther obtaining a sample subset Yc,jThe normalized values of all the elements in the sample subset Y are completedc,jTo obtain a training subset corresponding to the jth SVM modelFurther obtaining a training subset corresponding to each SVM model; whereinAs a subset of samples Yc,jNormalized results of the e sample; y isc,j(e)(ly) As a subset of samples Yc,jIth sample of (e)yAn element; mean (Y)0,j(ly) ) is sample subset Y under normal operating conditions0,jOf each sampleyA mean of the individual elements; std (Y)0,j(ly) ) is sample subset Y under normal operating conditions0,jOf each sampleyStandard deviation of individual elements.
Further, the specific method of step S5 includes the following sub-steps:
s5-1, for the jth SVM model, setting the sample labels under the normal working condition in the training subset corresponding to the jth SVM model as-1, and setting the sample labels under the fault working condition in the training subset corresponding to the jth SVM model as 1;
s5-2, obtaining the intercept coefficient tau of the jth SVM modeljThe u th training subset corresponding to the j th SVM modeljOne sample hj(uj) Corresponding specific gravity coefficient phij(uj);
S5-3, according to the formula:
K(hj(uj),h)=exp(-γ||hj(uj)-h||2)
acquiring a hyperplane equation of a jth SVM model; wherein hpj(h) Representing the value of the hyperplane equation of the jth SVM model when the detection sample is a sample h;mc,jthe number of samples in the sample subset of the c working condition in the training subset corresponding to the jth SVM model is determined; labelj(uj) For the u th training subset corresponding to the j th SVM modeljOne sample hj(uj) A tag value of (a); exp (·) is an exponential function; gamma is a hyperparameter, i.e., a constant; i | · | purple wind2Is the square of the vector two norm; k (-) is a radial basis function;
s5-4, according to the formula:
obtaining the output of the jth SVM model when the detection sample is the sample hFor each SVM model, twice of the value of the hyperplane equation corresponding to the detection sample after being activated by the sigmoid function is used as the output of the SVM model, and the construction of all SVM models is completed.
Further, the specific method of step S5-2 is:
for the jth SVM model, by solving an optimization problem:
0≤φj(uj)≤ξ',uj=1,2,…,Mj
obtaining the u-th training subset corresponding to the training subsetjOne sample hj(uj) Corresponding specific gravity coefficient phij(uj) (ii) a Wherein s.t. represents a constraint; h isj(ui) Representing the u th training subset corresponding to the j th SVM modeliA sample is obtained; labelj(ui) Is a sample hj(ui) A tag value of (a); xi 'is a penalty parameter, xi'>0;
Randomly selecting a sample h corresponding to the specific gravity coefficient which is more than 0 and less than xij(um) And sample hj(um) Label of (1)j(um) And according to the formula:
obtaining the intercept coefficient tau of the jth SVM modelj。
Further, the specific method of step S8 includes the following sub-steps:
s8-1, obtaining a detection sample xnewCorresponding all ICA model outputs and SVM model outputs to obtain the detection sample xnewCorresponding output matrixWhereinRepresenting the test sample xnewOutput on the ith ICA model;representing the test sample xnewIn the first placeOutputs on the j SVM models;
s8-2, according to the formula:
Pq(xnew|N)=exp(-vnew(q))
Pq(xnew|F)=exp(-1/vnew(q))
separately obtaining detection samples xnewConditional probability P under normal operating conditionsq(xnewN) and conditional probability P under fault conditionsq(xnew| F); wherein N refers to normal working conditions, and F refers to fault working conditions; exp (·) is an exponential function; v. ofnew(q) is the output matrix vnewThe qth value of (1);
s8-3, according to the formula:
Pq(xnew)=Pq(xnew|N)α+Pq(xnew|F)(1-α)
obtaining a test sample xnewThe total probability P corresponding to the q-th value in the sequenceq(xnew) (ii) a Wherein α is the confidence of the control limit in the ICA model;
s8-4, according to the formula:
separately obtaining detection samples xnewPosterior probability P of q-th value under normal working conditionq(N|xnew) And posterior probability P under fault conditionsq(F|xnew);
S8-5, according to the formula:
obtaining a test sample xnewIntegrated probability value P ofnew(ii) a Wherein QSIs a stand forWith the output of the ICA model and the number of outputs of the SVM model, i.e. the output matrix vnewTotal number of middle elements.
Further, the threshold in step S9 is 1- α, where α is the confidence of the control limit in the ICA model.
Further, the number of construction of the ICA model and the SVM model is 3.
The invention has the beneficial effects that: according to the method, an ICA model with a feature extraction capability and an SVM model with a classification capability are integrated by Bayesian inference, and different advantages of different models are utilized, so that the detection accuracy is improved, and the detection effect is more stable.
Drawings
FIG. 1 is a schematic flow chart of the present invention.
Detailed Description
The following description of the embodiments of the present invention is provided to facilitate the understanding of the present invention by those skilled in the art, but it should be understood that the present invention is not limited to the scope of the embodiments, and it will be apparent to those skilled in the art that various changes may be made without departing from the spirit and scope of the invention as defined and defined in the appended claims, and all matters produced by the invention using the inventive concept are protected.
As shown in fig. 1, the fault detection method based on the ICA and SVM ensemble learning includes the following steps:
s1, acquiring measurement data of the target industrial production system by the measurement equipment under different working conditions of the target industrial production system, listing the measurement data at different moments under the same working condition into a matrix, and taking each matrix as an initial sample set;
s2, standardizing the initial sample set under the normal working condition to obtain a standardized sample set under the normal working condition;
s3, constructing an ICA model based on the standardized sample set under the normal working condition;
s4, obtaining the number of SVM models to be constructed, randomly extracting samples from each initial sample set to form a sample subset of each working condition for each SVM model, and standardizing each sample subset to obtain a training subset corresponding to each SVM model;
s5, constructing a corresponding SVM model based on each training subset;
s6, acquiring measurement data of the current measurement equipment on the target industrial production system, and standardizing the measurement data to be used as a detection sample;
s7, respectively taking the detection samples as the input of all ICA models and SVM models, and correspondingly obtaining the output of each ICA model and the output of each SVM model;
s8, integrating the outputs of all ICA models and SVM models corresponding to the detection samples through Bayesian inference to calculate the integration probability value;
and S9, judging whether the integrated probability value is larger than or equal to a threshold value, if so, judging that the detection sample is a fault, otherwise, judging that the detection sample is normal, and finishing fault detection.
The specific method of step S1 is: the method comprises the steps that K measuring devices are adopted to simultaneously obtain measuring data of a target industrial production system under C working conditions, the measuring data corresponding to the K measuring devices at the same time are used as a sample, the measuring data at different times under the same working condition are listed into a matrix, and each matrix is used as an initial sample set to obtain C initial sample sets; i.e. each sample in each initial set of samples has K elements.
The specific method of step S2 is: for an initial sample set X under normal operating conditions0=[x0(1),x0(2),...,x0(r),...,x0(n0)]According to the formula:
obtaining the l-th sample of the r-th samplexNormalized value of each elementFurther obtaining the values of all elements after standardization, and completing the normal working conditionStandardizing the initial sample set to obtain a standardized sample set under a normal working condition; wherein x0(r)(lx) For the l-th sample in the r-th samplexA value of an element; mean (X)0(lx) Is the l-th sample of each sample in the initial sample set under normal conditionsxA mean of the individual elements; std (X)0(lx) Is the l-th sample of each sample in the initial sample set under normal conditionsxStandard deviation of individual elements; n is0The total number of samples in the initial sample set under the normal working condition.
The specific method of step S3 includes the following substeps:
s3-1, sample set under normal working condition after standardizationWhitening processing is carried out to obtain a whitening transformation matrix QPACAnd according to the formula:
obtaining a whitening matrix Z;
s3-2, construction of QICAAn ICA model, for the ith ICA model, according to the formula:
Si=BiZ
different random number seeds are set, and a FastICA algorithm is adopted to solve a first unmixing matrix B of an ith ICA modeliAnd independent vector matrix SiFurther, a first unmixing matrix and an independent vector matrix of each ICA model are obtained;
s3-3, according to the formula:
obtaining a second unmixing matrix W of the ith ICA modeliFurther obtaining a second unmixing matrix of each ICA model; wherein (·)TRepresents a transpose of a matrix;
s3-4, sequentially selecting d row vectors from the row vectors of the second unmixing matrix of the ith ICA model according to the sequence of vector norm from large to small to form a combined matrix W of the ith ICA modeld,iFurther obtaining a combination matrix of each ICA model;
s3-5, according to the formula:
obtaining a set of normalized samples under normal conditionsMiddle (r) th sampleStatistic value based on ith ICA modelFurther obtaining a standardized sample set under normal working conditionsAll the samples in the statistical quantity combination based on the ith ICA modelAnd obtaining a standardized sample set under normal working conditionsBased on the statistic combination of each ICA model; wherein n is0The total number of samples in the initial sample set under the normal working condition;
s3-6, obtaining statistic combination of ith ICA model by adopting kernel density estimation method KDEProbability density ofAnd according to the formula:
obtaining a solving interval [ start ] corresponding to the ith ICA modeli,endi]Further obtaining probability density functions and solving intervals of all ICA model statistic combinations; where min (-) denotes taking the minimum value and max (-) denotes taking the maximum value;
s3-7, solving interval [ start ] corresponding to ith ICA modeli,endi]Is equidistantly divided into numiSub-intervals and according to the formula:
acquiring the number k of accumulated subintervals corresponding to the ith ICA modeliFurther obtaining the cumulative number of subintervals corresponding to each ICA model; wherein ΔiThe subinterval width corresponding to the ith ICA model;statistical combination representing ith ICA modelAt starti+ξΔiThe probability density of (d); α is the confidence of the control limit; xi is a constant;
s3-8, according to the formula:
UCLi=starti+kiΔi
obtaining the control limit UCL of the ith ICA modeliFurther obtaining the control limit of each ICA model;
and S3-9, regarding any ICA model, taking the ratio of the statistic value of the input sample to the control limit as the output of the ICA model, and completing the construction of the ICA model.
The specific method of step S4 includes the following sub-steps:
s4-1, obtaining the quantity Q of SVM models to be constructedSVMFor the jth SVM model, m is randomly drawn from each initial sample set without replacementc,jEach sample constitutes a sample subset Y for each conditionc,j(ii) a Wherein m isc,j=int(nc,j×ratej) Int (·) denotes that only the integer part of the computation result is retained, nc,jIs the total number of samples in the initial sample set under the c-th working condition, when c is 0, the normal working condition is indicated, and ratejFor the extraction ratio corresponding to the jth SVM model, 0.0<ratej<1.0;
S4-2, according to the formula:
obtaining a sample subset Yc,jIth sample of (e)yNormalized value of individual elementFurther obtaining a sample subset Yc,jThe normalized values of all the elements in the sample subset Y are completedc,jTo obtain a training subset corresponding to the jth SVM modelFurther obtaining a training subset corresponding to each SVM model; whereinAs a subset of samples Yc,jNormalized results of the e sample; y isc,j(e)(ly) As a subset of samples Yc,jIth sample of (e)yAn element; mean (Y)0,j(ly) ) is sample subset Y under normal operating conditions0,jOf each sampleyA mean of the individual elements; std (Y)0,j(ly) ) is sample subset Y under normal operating conditions0,jOf each sampleyStandard deviation of individual elements.
The specific method of step S5 includes the following substeps:
s5-1, for the jth SVM model, setting the sample labels under the normal working condition in the training subset corresponding to the jth SVM model as-1, and setting the sample labels under the fault working condition in the training subset corresponding to the jth SVM model as 1;
s5-2, obtaining the intercept coefficient tau of the jth SVM modeljThe u th training subset corresponding to the j th SVM modeljOne sample hj(uj) Corresponding specific gravity coefficient phij(uj);
S5-3, according to the formula:
K(hj(uj),h)=exp(-γ||hj(uj)-h||2)
acquiring a hyperplane equation of a jth SVM model; wherein hpj(h) Representing the value of the hyperplane equation of the jth SVM model when the detection sample is a sample h;mc,jthe number of samples in the sample subset of the c working condition in the training subset corresponding to the jth SVM model is determined; labelj(uj) For the u th training subset corresponding to the j th SVM modeljOne sample hj(uj) A tag value of (a); exp (·) is an exponential function; gamma is a hyperparameter, i.e., a constant; i | · | purple wind2Is the square of the vector two norm; k (-) is a radial basis function;
s5-4, according to the formula:
obtaining the output of the jth SVM model when the detection sample is the sample hFor each SVM model, twice of the value of the hyperplane equation corresponding to the detection sample after being activated by the sigmoid function is used as the output of the SVM model, and the construction of all SVM models is completed.
The specific method of step S5-2 is:
for the jth SVM model, by solving an optimization problem:
0≤φj(uj)≤ξ',uj=1,2,…,Mj
obtaining the u-th training subset corresponding to the u-th training subsetjOne sample hj(uj) Corresponding specific gravity coefficient phij(uj) (ii) a Wherein s.t. represents a constraint; h isj(ui) Representing the u th training subset corresponding to the j th SVM modeliA sample is obtained; labelj(ui) Is a sample hj(ui) A tag value of (a); xi 'is a penalty parameter, xi'>0;
Randomly selecting a sample h corresponding to the specific gravity coefficient larger than 0 and smaller than xij(um) And sample hj(um) Label of (1)j(um) And according to the formula:
obtaining an intercept coefficient tau of a jth SVM modelj。
The specific method of step S8 includes the following substeps:
s8-1, obtaining a detection sample xnewCorresponding all ICA model outputs and SVM model outputs to obtain the detection sample xnewCorresponding output matrixWhereinRepresenting the test sample xnewOutput on the ith ICA model;representing the test sample xnewAn output on the jth SVM model;
s8-2, according to the formula:
Pq(xnew|N)=exp(-vnew(q))
Pq(xnew|F)=exp(-1/vnew(q))
separately obtaining detection samples xnewConditional probability P under normal operating conditionsq(xnewN) and conditional probability P under fault conditionsq(xnew| F); wherein N refers to normal working conditions, and F refers to fault working conditions; exp (·) is an exponential function; v. ofnew(q) is the output matrix vnewThe qth value of (1);
s8-3, according to the formula:
Pq(xnew)=Pq(xnew|N)α+Pq(xnew|F)(1-α)
obtaining a test sample xnewThe total probability P corresponding to the q-th value in the sequenceq(xnew) (ii) a Wherein α is the confidence of the control limit in the ICA model;
s8-4, according to the formula:
separately obtaining detection samples xnewPosterior probability P of q-th value under normal working conditionq(N|xnew) And posterior probability P under fault conditionsq(F|xnew);
S8-5, according to the formula:
obtaining a test sample xnewIntegrated probability value P ofnew(ii) a Wherein QSFor all ICA model outputs and the number of SVM model outputs, i.e. the output matrix vnewTotal number of middle elements.
In a specific implementation, the threshold in step S9 is 1- α, where α is the confidence in the control limits in the ICA model. The number of constructed ICA models and SVM models is 3.
In one embodiment of the present invention, a model of the U.S. Tennessee Eastman (TE) chemical process is used, which is taken from a real chemical process. The TE chemical process comprises five main units: the reactor, the condenser, the compressor, the separator and the stripping tower are widely applied to the research of various fault detection and diagnosis methods due to the fact that internal mechanisms of the reactor, the condenser, the compressor, the separator and the stripping tower are complex. The whole TE chemical process mainly comprises 22 continuous process measurement variables, 19 composition measurement variables and 12 operation variables, and can simulate normal working conditions and 21 fault working conditions. The data set in this embodiment is a data set that is disclosed and simulated by the american Institute of Technology (MIT), and is divided into a training set and a test set, where the two portions each include a sample set for each operating condition, and each sample is characterized by 52 dimensions. In the training set, there are 500 samples in the normal case sample set and 480 samples in the fault case sample set. In the test set, there are 960 samples for each case of the sample set, but because the fault was introduced after 160 normal conditions, the first 160 samples in the sample set of fault conditions belong to normal condition samples. In the embodiment, the training set part in the data set is used as the training set of the embodiment to train the model. The test set part in the data set is selected as the test set of the embodiment to test the effect of the fault detection method based on the ICA and SVM integration method.
In this embodiment, 3 ICA models and 3 SVM models are constructed, and thus, their individual detection results will be used for comparison with the detection results of the ICA and SVM integration method. Table 1 is a failure detection result recall rate statistical table of each model in this embodiment. Wherein ICA1、ICA2、ICA3Respectively representing No. 1, 2 and 3 ICA models, and judging in such a way that the statistic value of the sample exceeds the control limit, the model is regarded as a fault, and the SVM model1、SVM2、 SVM3Respectively representing the 1 st SVM model, the 2 nd SVM model and the 3 rd SVM model, and judging the mode that the hyperplane equation value of the sample exceeds 0 to be regarded as a fault.
Table 1: statistical table of fault detection result recall rate of each model
As the higher the recall value of the fault indicates the better the detection effect on the fault, it can be seen from table 1 that the detection effect of the method on the fault detection of the industrial production system is better than that of the prior art. Table 2 is a statistical table of the comparison result of the false alarm rate between the method of the present embodiment and the prior art.
Table 2: false alarm rate comparison result statistical table of method
Method | ICA1 | ICA2 | ICA3 | SVM1 | SVM2 | SVM3 | Method for producing a composite material |
False alarm rate | 0.37% | 0.67% | 0.39% | 2.31% | 2.34% | 2.13% | 1.01% |
The lower the false alarm rate is, the less possibility of false alarm to the normal working condition is. As can be seen from Table 2, the method better integrates the false alarm conditions of the ICA and SVM models, thereby obtaining a smaller and moderate false alarm value. The method has good identification effect on the normal working condition of the industrial production system, and the normal production process is not easily influenced. Therefore, the method achieves excellent effects in terms of both the false alarm rate and the fault detection rate.
In conclusion, the ICA model with the feature extraction capability and the SVM model with the classification capability are integrated by Bayesian inference, and different advantages of different models are utilized, so that the detection accuracy is improved, and the detection effect is more stable.
Claims (10)
1. A fault detection method based on ICA and SVM integrated learning is characterized by comprising the following steps:
s1, acquiring measurement data of the target industrial production system by the measurement equipment under different working conditions of the target industrial production system, listing the measurement data at different moments under the same working condition into a matrix, and taking each matrix as an initial sample set;
s2, standardizing the initial sample set under the normal working condition to obtain a standardized sample set under the normal working condition;
s3, constructing an ICA model based on the standardized sample set under the normal working condition;
s4, obtaining the number of SVM models to be constructed, randomly extracting samples from each initial sample set to form a sample subset of each working condition for each SVM model, and standardizing each sample subset to obtain a training subset corresponding to each SVM model;
s5, constructing a corresponding SVM model based on each training subset;
s6, acquiring measurement data of the current measurement equipment on the target industrial production system, and standardizing the measurement data to be used as a detection sample;
s7, respectively taking the detection samples as the input of all ICA models and SVM models, and correspondingly obtaining the output of each ICA model and the output of each SVM model;
s8, integrating the outputs of all ICA models and SVM models corresponding to the detection samples through Bayesian inference to calculate the integration probability value;
and S9, judging whether the integrated probability value is larger than or equal to a threshold value, if so, judging that the detection sample is a fault, otherwise, judging that the detection sample is normal, and finishing fault detection.
2. The fault detection method based on the ICA and SVM ensemble learning of claim 1, wherein the specific method of step S1 is:
the method comprises the steps that K measuring devices are adopted to simultaneously obtain measuring data of a target industrial production system under C working conditions, the measuring data corresponding to the K measuring devices at the same time are used as a sample, the measuring data at different times under the same working condition are listed into a matrix, and each matrix is used as an initial sample set to obtain C initial sample sets; i.e. each sample in each initial set of samples has K elements.
3. The fault detection method based on the ICA and SVM integrated learning of claim 1, wherein the specific method of the step S2 is:
for an initial sample set X under normal operating conditions0=[x0(1),x0(2),...,x0(r),...,x0(n0)]According to the formula:
obtaining the l-th sample of the r-th samplexNormalized value of individual elementFurther obtaining the values of all elements after standardization, completing the standardization of the initial sample set under the normal working condition, and obtaining the standardized sample set under the normal working condition; wherein x0(r)(lx) For the l-th sample in the r-th samplexA value of an element; mean (X)0(lx) Is the l-th sample of each sample in the initial sample set under normal conditionsxA mean of the individual elements; std (X)0(lx) Is the l-th sample of each sample in the initial sample set under normal conditionsxStandard deviation of individual elements; n is0The total number of samples in the initial sample set under the normal working condition.
4. The fault detection method based on the ICA and SVM ensemble learning of claim 1, wherein the specific method of the step S3 comprises the following sub-steps:
s3-1, sample set under normal working condition after standardizationWhitening to obtain a whitening transformation matrix QPACAnd according to the formula:
obtaining a whitening matrix Z;
s3-2, construction of QICAAn ICA model, for the ith ICA model, according to the formula:
Si=BiZ
different random number seeds are set, and a FastICA algorithm is adopted to solve a first unmixing matrix B of an ith ICA modeliAnd independent vector matrix SiFurther, a first unmixing matrix and an independent vector matrix of each ICA model are obtained;
s3-3, according to the formula:
obtaining a second unmixing matrix W of the ith ICA modeliFurther obtaining a second unmixing matrix of each ICA model; wherein (·)TRepresents a transpose of a matrix;
s3-4, sequentially selecting d row vectors from the row vectors of the second unmixing matrix of the ith ICA model according to the sequence of vector norm from large to small to form a combined matrix W of the ith ICA modeld,iFurther obtaining a combination matrix of each ICA model;
s3-5, according to the formula:
obtaining a set of normalized samples under normal conditionsMiddle (r) th sampleStatistic value based on ith ICA modelFurther obtaining a standardized sample set under normal working conditionsAll samples in the statistical quantity combination based on ith ICA modelAnd obtaining a standardized sample set under normal working conditionsAll samples in (a) are based on a combination of statistics for each ICA model; wherein n is0The total number of samples in the initial sample set under the normal working condition;
s3-6, obtaining statistic combination of ith ICA model by adopting kernel density estimation method KDEProbability density ofAnd according to the formula:
obtaining a solving interval [ start ] corresponding to the ith ICA modeli,endi]Further obtaining probability density functions and solving intervals of all ICA model statistic combinations; where min (-) denotes taking the minimum value and max (-) denotes taking the maximum value;
s3-7, solving interval [ start ] corresponding to ith ICA modeli,endi]Is equidistantly divided into numiSub-intervals and according to the formula:
acquiring the number k of accumulated subintervals corresponding to the ith ICA modeliFurther obtaining the cumulative number of subintervals corresponding to each ICA model; wherein ΔiThe subinterval width corresponding to the ith ICA model;statistical combination representing ith ICA modelAt starti+ξΔiThe probability density of (d); α is the confidence of the control limit; xi is a constant;
s3-8, according to the formula:
UCLi=starti+kiΔi
obtaining the control limit UCL of the ith ICA modeliFurther obtaining the control limit of each ICA model;
and S3-9, regarding any ICA model, taking the ratio of the statistic value of the input sample to the control limit as the output of the ICA model, and completing the construction of the ICA model.
5. The fault detection method based on the ICA and SVM ensemble learning of claim 1, wherein the specific method of the step S4 comprises the following sub-steps:
s4-1, obtaining the quantity Q of SVM models to be constructedSVMFor the jth SVM model, m is randomly drawn from each initial sample set without replacementc,jEach sample constitutes a sample subset Y for each conditionc,j(ii) a Wherein m isc,j=int(nc,j×ratej) Int (·) denotes that only the integer part of the computation result is retained, nc,jIs the total number of samples in the initial sample set under the c-th working condition, when c is 0, the normal working condition is indicated, and ratejFor the extraction ratio corresponding to the jth SVM model, 0.0<ratej<1.0;
S4-2, according to the formula:
obtaining a sample subset Yc,jIth sample of (e)yNormalized value of individual elementFurther obtaining a sample subset Yc,jThe normalized values of all the elements in the sample subset Y are completedc,jTo obtain a training subset corresponding to the jth SVM modelFurther obtaining a training subset corresponding to each SVM model; whereinAs a subset of samples Yc,jNormalized results of the e sample; y isc,j(e)(ly) As a subset of samples Yc,jIth sample of (e)yAn element; mean (Y)0,j(ly) ) is sample subset Y under normal operating conditions0,jOf each sampleyOf a single elementMean value; std (Y)0,j(ly) ) is sample subset Y under normal operating conditions0,jOf each sampleyStandard deviation of individual elements.
6. The fault detection method based on the ICA and SVM ensemble learning of claim 1, wherein the specific method of the step S5 comprises the following sub-steps:
s5-1, for the jth SVM model, setting the sample labels under the normal working condition in the training subset corresponding to the jth SVM model as-1, and setting the sample labels under the fault working condition in the training subset corresponding to the jth SVM model as 1;
s5-2, obtaining the intercept coefficient tau of the jth SVM modeljThe u th training subset corresponding to the j th SVM modeljOne sample hj(uj) Corresponding specific gravity coefficient phij(uj);
S5-3, according to the formula:
K(hj(uj),h)=exp(-γ||hj(uj)-h||2)
acquiring a hyperplane equation of a jth SVM model; wherein hpj(h) Representing the value of the hyperplane equation of the jth SVM model when the detection sample is a sample h;mc,jthe number of samples in the sample subset of the c working condition in the training subset corresponding to the jth SVM model is determined; labelj(uj) For the u th training subset corresponding to the j th SVM modeljOne sample hj(uj) A tag value of (a); exp (·) is an exponential function; gamma is a hyperparameter, i.e., a constant; i | · | purple wind2Is the square of the vector two norm; k (-) is a radial basis function;
s5-4, according to the formula:
obtaining the output of the jth SVM model when the detection sample is the sample hFor each SVM model, twice of the value of the hyperplane equation corresponding to the detection sample after being activated by the sigmoid function is used as the output of the SVM model, and the construction of all SVM models is completed.
7. The fault detection method based on ICA and SVM ensemble learning of claim 6, wherein the specific method of step S5-2 is:
for the jth SVM model, by solving an optimization problem:
0≤φj(uj)≤ξ',uj=1,2,…,Mj
obtaining the u-th training subset corresponding to the u-th training subsetjOne sample hj(uj) Corresponding specific gravity coefficient phij(uj) (ii) a Wherein s.t. represents a constraint; h isj(ui) Representing the u th training subset corresponding to the j th SVM modeliA sample is obtained; labelj(ui) Is a sample hj(ui) The tag value of (a); xi 'is a penalty parameter, xi'>0;
Randomly selecting a sample h corresponding to the specific gravity coefficient larger than 0 and smaller than xij(um) And sample hj(um) Label of (1)j(um) And according to the formula:
obtaining the intercept coefficient tau of the jth SVM modelj。
8. The fault detection method based on the ICA and SVM ensemble learning of claim 1, wherein the specific method of the step S8 comprises the following sub-steps:
s8-1, obtaining a detection sample xnewCorresponding all ICA model outputs and SVM model outputs to obtain and detect sample xnewCorresponding output matrixWhereinRepresents the detection sample xnewOutput on the ith ICA model;represents the detection sample xnewAn output on the jth SVM model;
s8-2, according to the formula:
Pq(xnew|N)=exp(-vnew(q))
Pq(xnew|F)=exp(-1/vnew(q))
separately obtaining detection samples xnewConditional probability P under normal operating conditionsq(xnewN) and conditional probability P under fault conditionsq(xnew| F); wherein N refers to normal working conditions, and F refers to fault working conditions; exp (·) is an exponential function; v. ofnew(q) is the output matrix vnewThe qth value of (1);
s8-3, according to the formula:
Pq(xnew)=Pq(xnew|N)α+Pq(xnew|F)(1-α)
obtaining a test sample xnewThe total probability P corresponding to the q-th value in the sequenceq(xnew) (ii) a Wherein α is the confidence of the control limit in the ICA model;
s8-4, according to the formula:
respectively obtaining detection samples xnewPosterior probability P of q-th value under normal working conditionq(N|xnew) And posterior probability P under fault conditionsq(F|xnew);
S8-5, according to the formula:
obtaining a test sample xnewIntegrated probability value P ofnew(ii) a Wherein QSFor all ICA model outputs and the number of SVM model outputs, i.e. the output matrix vnewTotal number of middle elements.
9. The fault detection method based on ICA and SVM integrated learning of claim 1, wherein the threshold in the step S9 is 1- α, where α is the confidence of the control limit in the ICA model.
10. The fault detection method based on ICA and SVM ensemble learning of claim 1, wherein the number of constructions of ICA models and SVM models is 3.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010612207.XA CN111898313B (en) | 2020-06-30 | 2020-06-30 | Fault detection method based on ICA and SVM integrated learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010612207.XA CN111898313B (en) | 2020-06-30 | 2020-06-30 | Fault detection method based on ICA and SVM integrated learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111898313A CN111898313A (en) | 2020-11-06 |
CN111898313B true CN111898313B (en) | 2022-05-20 |
Family
ID=73207315
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010612207.XA Active CN111898313B (en) | 2020-06-30 | 2020-06-30 | Fault detection method based on ICA and SVM integrated learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111898313B (en) |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101021723A (en) * | 2006-12-22 | 2007-08-22 | 浙江大学 | Melt index detection fault diagnozing system and method in propylene polymerization production |
CN102339389A (en) * | 2011-09-14 | 2012-02-01 | 清华大学 | Fault detection method for one-class support vector machine based on density parameter optimization |
CN103377316A (en) * | 2013-07-15 | 2013-10-30 | 浙江大学 | Penicillin production process monitoring method based on statistical analysis and Bayesian ensemble |
CN105739489A (en) * | 2016-05-12 | 2016-07-06 | 电子科技大学 | Batch process fault detecting method based on ICA-KNN |
CA2995864A1 (en) * | 2017-03-09 | 2018-09-09 | General Electric Company | Multi-modal, multi-disciplinary feature discovery to detect cyber threats in electric power grid |
CN108804740A (en) * | 2018-04-02 | 2018-11-13 | 西安建筑科技大学 | Long distance pipeline pressure monitoring method based on integrated improvement ICA-KRR algorithms |
CN109948673A (en) * | 2019-03-05 | 2019-06-28 | 中国人民解放军海军勤务学院 | Structural Damage Identification based on isolated component and support vector machines |
CN110738259A (en) * | 2019-10-16 | 2020-01-31 | 电子科技大学 | fault detection method based on Deep DPCA-SVM |
CN111474920A (en) * | 2020-04-29 | 2020-07-31 | 电子科技大学 | ICA and ACVA-based fault detection method |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080040152A1 (en) * | 2006-08-10 | 2008-02-14 | The Boeing Company | Systems and Methods for Health Management of Single or Multi-Platform Systems |
-
2020
- 2020-06-30 CN CN202010612207.XA patent/CN111898313B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101021723A (en) * | 2006-12-22 | 2007-08-22 | 浙江大学 | Melt index detection fault diagnozing system and method in propylene polymerization production |
CN102339389A (en) * | 2011-09-14 | 2012-02-01 | 清华大学 | Fault detection method for one-class support vector machine based on density parameter optimization |
CN103377316A (en) * | 2013-07-15 | 2013-10-30 | 浙江大学 | Penicillin production process monitoring method based on statistical analysis and Bayesian ensemble |
CN105739489A (en) * | 2016-05-12 | 2016-07-06 | 电子科技大学 | Batch process fault detecting method based on ICA-KNN |
CA2995864A1 (en) * | 2017-03-09 | 2018-09-09 | General Electric Company | Multi-modal, multi-disciplinary feature discovery to detect cyber threats in electric power grid |
CN108804740A (en) * | 2018-04-02 | 2018-11-13 | 西安建筑科技大学 | Long distance pipeline pressure monitoring method based on integrated improvement ICA-KRR algorithms |
CN109948673A (en) * | 2019-03-05 | 2019-06-28 | 中国人民解放军海军勤务学院 | Structural Damage Identification based on isolated component and support vector machines |
CN110738259A (en) * | 2019-10-16 | 2020-01-31 | 电子科技大学 | fault detection method based on Deep DPCA-SVM |
CN111474920A (en) * | 2020-04-29 | 2020-07-31 | 电子科技大学 | ICA and ACVA-based fault detection method |
Non-Patent Citations (4)
Title |
---|
A Novel Deep DPCA-SVM Method for Fault Detection in Industrial Processes;Zhang Jian et al.;《2019 IEEE 58th Conference on Decision and Control (CDC)》;20200312;2916–2921 * |
Application of Independent Component Analysis to the aero-engine fault diagnosis;Li Zhonghai,et al.;《2009 Chinese Control and Decision Conference》;20090807;5330-5333 * |
三重化冗余多机系统心跳检测机制研究;邹见效等;《电子科技大学学报》;20100530;379-383 * |
基于ICA的鲁棒故障检测方法研究及在TE过程的应用;张建;《中国优秀博硕士学位论文全文数据库(硕士)工程科技Ⅰ辑》;20200715;B016-63 * |
Also Published As
Publication number | Publication date |
---|---|
CN111898313A (en) | 2020-11-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Ko et al. | Fault classification in high-dimensional complex processes using semi-supervised deep convolutional generative models | |
Yan et al. | Chiller faults detection and diagnosis with sensor network and adaptive 1D CNN | |
Monroy et al. | A semi-supervised approach to fault diagnosis for chemical processes | |
CN113642754A (en) | Complex industrial process fault prediction method based on RF noise reduction self-coding information reconstruction and time convolution network | |
Tian et al. | Identification of abnormal conditions in high-dimensional chemical process based on feature selection and deep learning | |
CN111397902A (en) | Rolling bearing fault diagnosis method based on feature alignment convolutional neural network | |
Zhang et al. | Rotating machinery remaining useful life prediction scheme using deep-learning-based health indicator and a new RVM | |
CN111583592B (en) | Experimental environment safety early warning method based on multidimensional convolution neural network | |
CN111563468A (en) | Driver abnormal behavior detection method based on attention of neural network | |
CN116167010A (en) | Rapid identification method for abnormal events of power system with intelligent transfer learning capability | |
CN117056678B (en) | Machine pump equipment operation fault diagnosis method and device based on small sample | |
CN107491058A (en) | A kind of industrial control system sequence attack detection method and equipment | |
CN110738259B (en) | Fault detection method based on Deep DPCA-SVM | |
CN111898313B (en) | Fault detection method based on ICA and SVM integrated learning | |
Xu et al. | Deep spiking residual shrinkage network for bearing fault diagnosis | |
CN114186639A (en) | Electrical accident classification method based on dual-weighted naive Bayes | |
Danesh et al. | Hybridization of model-specific and model-agnostic methods for interpretability of Neural network predictions: Application to a power plant | |
CN113135480A (en) | Elevator fault early warning method based on local and overall characteristics | |
CN111898249A (en) | Landslide displacement nonparametric probability density prediction method, equipment and storage medium | |
CN116400168A (en) | Power grid fault diagnosis method and system based on depth feature clustering | |
CN112203311A (en) | Network element abnormity diagnosis method, device, equipment and computer storage medium | |
Zhang et al. | A Flexible Monitoring Framework via Dynamic-Multilayer Graph Convolution Network | |
CN115858606A (en) | Method, device and equipment for detecting abnormity of time series data and storage medium | |
CN113526283A (en) | Method, medium and equipment for diagnosing short-term sudden faults of elevator | |
Shah et al. | Analyzing data complexity using metafeatures for classification algorithm selection |
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 |