CN112598022A - Improved FDA process industrial fault diagnosis method based on ensemble learning method - Google Patents

Improved FDA process industrial fault diagnosis method based on ensemble learning method Download PDF

Info

Publication number
CN112598022A
CN112598022A CN202011374338.5A CN202011374338A CN112598022A CN 112598022 A CN112598022 A CN 112598022A CN 202011374338 A CN202011374338 A CN 202011374338A CN 112598022 A CN112598022 A CN 112598022A
Authority
CN
China
Prior art keywords
expression
fda
follows
fault diagnosis
data
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202011374338.5A
Other languages
Chinese (zh)
Inventor
贺彦林
赵阳
朱群雄
徐圆
张洋
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing University of Chemical Technology
Original Assignee
Beijing University of Chemical Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beijing University of Chemical Technology filed Critical Beijing University of Chemical Technology
Priority to CN202011374338.5A priority Critical patent/CN112598022A/en
Publication of CN112598022A publication Critical patent/CN112598022A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • 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
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/245—Classification techniques relating to the decision surface
    • G06F18/2451—Classification techniques relating to the decision surface linear, e.g. hyperplane
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Probability & Statistics with Applications (AREA)
  • Medical Informatics (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Complex Calculations (AREA)

Abstract

The invention discloses an improved FDA process industrial fault diagnosis method based on an ensemble learning method, which fully retains the time sequence relevance and more useful information of an industrial process among data after the data are subjected to dynamic processing, and is convenient for feature extraction of an FDA method of L2 norm normalization. According to the technical scheme provided by the invention, under the action of the integrated learning method, the time for establishing the comprehensive fault diagnosis model is shortened, and the fault diagnosis efficiency is improved. Compared with a fault diagnosis model which is not subjected to dynamic processing and is based on a Bayesian method, the technical scheme provided by the invention has obvious advantages through experimental simulation results.

Description

Improved FDA process industrial fault diagnosis method based on ensemble learning method
Technical Field
The invention relates to the technical field of process industrial fault diagnosis, in particular to an improved FDA (food and drug administration) process industrial fault diagnosis method based on an integrated learning method.
Background
The process industry related to the industries of petroleum, chemical industry, metallurgy, paper making and the like is taken as an important prop industry of national economy in China, and the safe production of the process industry is an important concern. The development of the process industry gradually shows the characteristics of large scale, nonlinearity and strong coupling of production equipment so far due to the continuous development and updating of discipline fusion and computer technology, and the production process also has high complexity and high risk correspondingly, so that various faults frequently occur, and the economic loss and even the loss of life of people are caused. Therefore, fault diagnosis in the process industry is always an important research subject of the production process, is a key technology of industrial safety, and is important for the safe operation of an industrial control system. Finding and solving faults in time is an important means for improving production safety and economic benefit. Therefore, the method has important practical significance for researching a fault diagnosis model and improving the fault diagnosis accuracy.
In recent years, data-driven fault diagnosis techniques based on multivariate statistical techniques have been widely studied and applied, such as principal component analysis methods, local preserving projection methods, canonical variable analysis methods, and the like. Principal Component Analysis (PCA) was once the most widely used data-driven technique in industrial monitoring systems, the classical method for data dimensionality reduction. Extracting the structural features of the data using PCA techniques can eliminate the correlation of the variables while ensuring that the sample data has the greatest variance. Due to the advantages of the PCA method in data structure feature extraction, further studies on PCA have not been interrupted. For example, the scholars propose a novel pattern recognition framework based on multi-scale pca (mspca) which incorporates an adaptive neuro-fuzzy inference system to creatively divide the state space into a score space and a residual space after extracting data features, which allows fault diagnosis to be performed by multiple classifiers. However, in the fault diagnosis process, when the dimension of the data is reduced, the data information among all the mode types is also an object needing important attention, and based on the object, the dimension of the data is reduced by adopting Fisher discriminant analysis. Fisher Discriminant Analysis (FDA) is a data dimension reduction method widely used in pattern classification, and this technique can fully consider data information between pattern classes. Particularly in the case of a relatively small number of process fault observations affecting production and a relatively large number of variables contained in the fault mode class, the FDA is more suitable for use in fault diagnosis when determining a representation of lower dimensionality due to its ability to focus on inter-class information.
As a supervised learning method, the FDA has been extensively studied with functions of feature extraction and pattern classification. As one method of data feature selection, FDA outperforms PCA in fault diagnosis. The FDA may also perform pre-classification at the same time as feature extraction, which can reduce the performance requirements on the classifier when performing failure mode recognition compared to other data-driven methods.
When the invention is used for fault mode classification, an algorithm which is widely applied in the field of machine learning and data mining is selected: the Adaboost algorithm. The Adaboost algorithm is an integrated algorithm and is applied to the fields of two classes, namely the Adaboost M1 variant, and is applied to the fields of multiple classes, namely the Adaboost M2 variant. Considering that the flow industry has more fault types and the fault diagnosis belongs to a multi-classification task, the Adaboost M2 variant is selected by the invention. In the invention, the algorithm is aimed at multi-classification tasks, and a special combination strategy is applied to collect a plurality of homogeneous weak classifiers (basic classifiers) with poor classification performance, so that the homogeneous weak classifiers are promoted to be strong classifiers and obtain excellent classification performance. The working mechanism of Adaboost M2 is: firstly, training a first basic classifier 1 by using initial weight from a training set, then classifying the patterns, further updating the weight of samples in the training set according to the learning error rate of the basic classifier 1 to the fault patterns, increasing the weight of classified error samples and reducing the weight of classified correct samples, so that each time of classified error samples is paid more attention. And after the training samples with updated weights are obtained, training the basic classifier 2, iterating for a plurality of times in this way, terminating the training until the number of the basic classifiers reaches the preset number T, combining the T basic classifiers through a set strategy, and finally upgrading the T basic classifiers into a strong classifier.
In order to verify the feasibility of the method, the fault type in the Tennessee Eastman Process (TEP) is taken as a research object, and fault diagnosis model establishment is carried out based on an FDA method, so that the method aims to diagnose the fault of the Process industry in time, reduce economic loss caused by the fault and realize efficient and safe production. TEP is a realistic industrial process simulation created by eastman chemicals, usa, and provides a model paradigm for scientific research of process monitoring. The simulation process comprises five core composition units of a reactor, a condenser, a compressor, a separator and a stripping tower, and material flow relates to eight chemical components. The experimental result of the fault diagnosis model based on the TEP shows that compared with other models, the improved FDA fault diagnosis model based on the integrated learning method can realize higher-precision fault diagnosis.
Disclosure of Invention
In order to solve the limitations and defects of the prior art, the invention provides an improved FDA process industrial fault diagnosis method based on an ensemble learning method, which comprises the steps of preprocessing the obtained fault data, constructing a dynamic input process, reducing the FDA data dimension and extracting the features, classifying the mode of the Adaboost M2 method by the ensemble learning method, and constructing a fault diagnosis model;
the method for classifying patterns by the ensemble learning method Adaboost M2 comprises the following steps:
given a training set S and a base classifier space Ψ, the expression is as follows:
Figure BDA0002807777440000031
wherein z isiIs the row vector of data Z, y is the class label, c is the fault class number, phi represents the basic classifier;
training set weights are initialized, and the expression is as follows:
Figure BDA0002807777440000032
for each weak classifier, dividing the sample space Z to obtain Z1，Z2，...，ZmIn aProbability distribution DfThe sample class attribute probability is calculated as follows:
Figure BDA0002807777440000033
setting the output of a basic classifier in the divided sample space Z, wherein the expression is as follows:
Figure BDA0002807777440000034
calculating to obtain a normalization factor, wherein the expression is as follows:
Figure BDA0002807777440000035
wherein there are F features, F1, 2,., F;
the basic classifier is chosen such that the normalization factor is minimal, the expression is as follows:
Figure BDA0002807777440000041
calculating error fraction, and expressing the following expression:
Figure BDA0002807777440000042
wherein, Tm(z) is the sample label predicted by the base classifier;
calculating the weight of each basic classifier, and expressing the following expression:
Figure BDA0002807777440000043
and updating the sample weight, wherein the expression is as follows:
Figure BDA0002807777440000044
obtaining a combined classifier, wherein the expression is as follows:
Figure BDA0002807777440000045
wherein the content of the first and second substances,
Figure BDA0002807777440000046
optionally, the step of preprocessing the obtained fault data includes: selecting 52 process variables of the Tennessee-Ismann process as input, and carrying out fault diagnosis on 21 faults of the Tennessee-Ismann process;
obtaining a normalized data set X of a failed original training sample0The expression is as follows:
Figure BDA0002807777440000047
wherein n represents the number of samples, and m represents the number of variables;
the step of constructing a dynamic input process comprises: for the normalized data set X0Performing dynamic processing, and selecting a time lag constant L to be 2 to obtain an augmentation matrix X, wherein the expression is as follows:
Figure BDA0002807777440000051
the FDA data dimension reduction and feature extraction steps comprise: and reducing the dimension of the augmentation matrix X, calculating an FDA feature vector, and performing L2 norm normalization.
Optionally, the step of performing dimension reduction on the augmented matrix X, calculating an FDA eigenvector, and performing norm normalization by using L2 includes:
and constructing an overall dispersion matrix, wherein the expression is as follows:
St＝Sw+Sb (3)
wherein S iswRepresenting an intra-class dispersion matrix, SbRepresenting an inter-class dispersion matrix;
calculating an FDA vector, and maximizing the inter-class dispersion while minimizing the intra-class dispersion, wherein the expression is as follows:
Figure BDA0002807777440000052
using lagrange multiplier method, obtain FDA vector wk by solving generalized eigenvalues, the expression is as follows:
Sbwk＝λpSwwk (5)
for the FDA vector wkL2 norm normalization is performed, and the expression is as follows:
Figure BDA0002807777440000053
r FDA vectors are selected to construct a linear transformation matrix WnewThe expression is as follows:
Wnew＝(w2 1，w2 2，...，，w2 r)T (7)
the process industry data dimension is carried out, and the expression is as follows:
Z＝WT new×X (8)。
the invention has the following beneficial effects:
according to the improved FDA process industrial fault diagnosis method based on the ensemble learning method, provided by the invention, after data are subjected to dynamic processing, the time sequence relevance and more useful information of an industrial process are fully reserved among the data, and the FDA method of L2 norm normalization is convenient for feature extraction. According to the technical scheme provided by the invention, under the action of the integrated learning method, the time for establishing the comprehensive fault diagnosis model is shortened, and the fault diagnosis efficiency is improved. Compared with a fault diagnosis model which is not subjected to dynamic processing and is based on a Bayesian method, the fault diagnosis method has the obvious advantages.
Drawings
Fig. 1 is a flowchart of a TEP according to an embodiment of the present invention.
Fig. 2 is a schematic diagram of an average classification accuracy curve in a training process according to an embodiment of the present invention.
Fig. 3 is a diagram illustrating a sample point distribution effect after dimensionality reduction of a data set according to an embodiment of the present invention.
Fig. 4 is a schematic diagram of a confusion matrix of the fault diagnosis result according to an embodiment of the present invention.
Fig. 5 is a comparison graph of fault diagnosis results of different methods according to the first embodiment of the present invention.
Detailed Description
In order to make those skilled in the art better understand the technical solution of the present invention, the improved FDA process industrial fault diagnosis method based on the ensemble learning method provided by the present invention is described in detail below with reference to the accompanying drawings.
Example one
The embodiment provides a novel fault diagnosis method based on Adaboost M2 dynamic L2 norm normalized Fisher discriminant analysis (D-L2-FDA), and a fault data set is dynamically processed in consideration of time sequence characteristics of collected process data in a constructed D-L2-FDA model. The L2 norm normalized FDA is used to extract data features and reduce the dimensionality of the fault data. In addition, in order to classify different types of failure modes, the present embodiment adopts an ensemble learning method such as Adaboost M2. In order to verify the effectiveness of the proposed D-L2-FDA model based on Adaboost M2 in fault diagnosis, the present example was subjected to case study based on Tennessman process. Simulation results show that the diagnostic performance of the D-L2-FDA model based on Adaboost M2 is superior to other methods.
Fig. 1 is a flowchart of a TEP according to an embodiment of the present invention. In order to improve the fault diagnosis accuracy, the embodiment provides an improved FDA process industrial fault diagnosis method based on an integrated learning method, so as to accurately diagnose a process fault and ensure production safety. The embodiment comprises the following steps: the method comprises the steps of obtaining data, conducting data preprocessing, constructing a dynamic input process, FDA data dimension reduction and feature extraction, conducting mode classification by an integrated learning method Adaboost M2, and constructing a fault diagnosis model. According to the technical scheme provided by the embodiment, the building time of the comprehensive fault diagnosis model is shortened under the action of the integrated learning method, and the fault diagnosis efficiency is improved.
The embodiment is based on TEP simulation, actual factory production is truly simulated, an obtained data set is divided into a training set and a testing set, the training set is composed of 480 samples, and the testing set is composed of 960 samples. The data in the training set and the test set contain information of 52 variables in total, namely the control variable and the measurement variable, at different sampling moments, and the variable information is shown in table 1. All data sets cover 22 different conditions, including 1 fault-free set and 21 fault sets, and the fault conditions are described and their types are shown in table 2.
TABLE 1 Process variables and their description
Figure BDA0002807777440000071
TABLE 2 Fault and description thereof
Figure BDA0002807777440000081
Fig. 2 is a schematic diagram of an average classification accuracy curve in a training process according to an embodiment of the present invention. In this embodiment, data preprocessing is performed, and in order to consider continuity and dynamic property of the process, dynamic processing is performed on the original data set.
Fig. 3 is a diagram illustrating a sample point distribution effect after dimensionality reduction of a data set according to an embodiment of the present invention. In this embodiment, the dimension reduction and feature extraction are performed on the fault data set, an improved FDA method is adopted, after an FDA feature vector is solved, L2 norm normalization is performed to obtain a feature vector direction, while the magnitude of the feature vector is ignored, so that the model calculation complexity is simplified.
In this embodiment, after the fault mode classification is performed to obtain the specification data after the dimension reduction, the fault mode classification is performed by using an ensemble learning method Adaboost M2 based on a decision tree to complete fault diagnosis.
Fig. 4 is a schematic diagram of a confusion matrix of the fault diagnosis result according to an embodiment of the present invention. Fig. 5 is a comparison graph of fault diagnosis results of different methods according to the first embodiment of the present invention. The embodiment is an improved FDA fault diagnosis model based on an ensemble learning method, which is used for diagnosing process faults of the process industry, the initial data set is dynamically processed, L2 norm normalization is carried out on the obtained FDA feature vector to obtain a fault feature extraction model, fault mode classification is carried out by combining an Adaboost M2 algorithm based on decision tree integration, and finally fault diagnosis of the process industry is realized.
This embodiment obtains a normalized data set X of the original training sample of the failure0：
Figure BDA0002807777440000091
Where n represents the number of samples and m represents the number of variables.
This example is for X0Performing dynamic processing, and selecting a time lag constant L to be 2 to obtain an augmentation matrix X:
Figure BDA0002807777440000092
in this embodiment, dimension reduction is performed on the augmented matrix, FDA eigenvectors are calculated, and L2 norm normalization is performed. This embodiment constructs the total dispersion matrix:
St＝Sw+Sb (3)
wherein S iswRepresenting an intra-class dispersion matrix, SbRepresenting an inter-class dispersion matrix.
The present embodiment calculates the FDA vector, and maximizes the inter-class dispersion while minimizing the intra-class dispersion:
Figure BDA0002807777440000093
the present embodiment applies the lagrangian multiplier method to obtain the FDA vector w by solving the generalized eigenvalue of the following equationk：
Sbwk＝λpSwwk (5)
In this embodiment, a series of FDA vectors are normalized by an L2 norm:
Figure BDA0002807777440000094
in the embodiment, r FDA vectors are selected to construct a linear transformation matrix Wnew：
Wnew＝(w2 I，w2 2，...，w2 r)T (7)
In this embodiment, the process industry data dimension reduction is performed:
Z＝WT new×X (8)
in this embodiment, an ensemble learning method Adaboost M2 is applied to classify the failure modes, and the generation process of the combination classifier is as follows:
this embodiment gives a training set S and a base classifier space Ψ:
Figure BDA0002807777440000101
wherein z isiIs a row vector of data Z, representing one sample. y is the class label, c is the number of failed classes, and φ represents the basic classifier.
This embodiment initializes the training set weights:
Figure BDA0002807777440000102
this embodiment assumes that a total of F features need to be obtained, for F ═ 1,2, …, F:
for each weak classifier, the present embodiment performs the following operations:
this example divides the sample space Z to obtain Z1，Z2，...，Zm；
Under probability distribution, the present embodiment calculates a sample class attribute probability:
Figure BDA0002807777440000103
the present embodiment sets the output of the basic classifier in the above division:
Figure BDA0002807777440000104
in this embodiment, a normalization factor is calculated:
Figure BDA0002807777440000105
the embodiment selects a basic classifier such that the normalization factor is minimal:
Figure BDA0002807777440000106
this embodiment calculates the error fraction:
Figure BDA0002807777440000107
wherein, Tm(z) is the sample label predicted by the basic classifier.
The present embodiment calculates the weight of each basic classifier:
Figure BDA0002807777440000111
the present embodiment updates the sample weight:
Figure BDA0002807777440000112
this embodiment results in a combined classifier:
Figure BDA0002807777440000113
wherein the content of the first and second substances,
Figure BDA0002807777440000114
the combined classifier obtained by the expression (18) integrates a plurality of basic classifiers with relatively weak classification capability into a strong classifier, which is called as an ensemble learning method in the embodiment. Adaboost M2, as an ensemble learning method, provides a novel pattern classifier that provides a new idea for fault diagnosis by predicting the labels of test set data based on extracted process data features to implement pattern classification. The idea of the Adaboost M2 method is to construct a first simple classifier based on the original data distribution. Then, bootstrap generates T weak classifiers, and finally, the different classifiers are combined together to construct a strong classifier with better performance. As can be seen from the algorithm flow, in the weight updating process, the weight of the misclassified sample is increased, and the weight of the correctly classified sample is decreased. Thus, Adaboost M2 enhances learning of misclassified samples.
An improved FDA fault diagnosis model based on an ensemble learning method Adaboost M2 is finally designed, and from the experimental result of TEP fault diagnosis of the model, the embodiment has high fault diagnosis accuracy and certain practical application value.
According to the improved FDA process industrial fault diagnosis method based on the ensemble learning method, after data are subjected to dynamic processing, time sequence relevance and more useful information of an industrial process are fully reserved among the data, and feature extraction is conveniently performed by the FDA method of L2 norm normalization. According to the technical scheme provided by the embodiment, the building time of the comprehensive fault diagnosis model is shortened under the action of the integrated learning method, and the fault diagnosis efficiency is improved. Compared with a fault diagnosis model which is not subjected to dynamic processing and is based on a Bayesian method, the technical scheme provided by the embodiment has significant advantages through experimental simulation results.
It will be understood that the above embodiments are merely exemplary embodiments taken to illustrate the principles of the present invention, which is not limited thereto. It will be apparent to those skilled in the art that various modifications and improvements can be made without departing from the spirit and substance of the invention, and these modifications and improvements are also considered to be within the scope of the invention.

Claims (3)

1. An improved FDA process industrial fault diagnosis method based on an ensemble learning method is characterized by comprising the steps of preprocessing obtained fault data, constructing a dynamic input process, FDA data dimension reduction and feature extraction, carrying out mode classification by using an ensemble learning method Adaboost M2 method, and constructing a fault diagnosis model;
the method for classifying patterns by the ensemble learning method Adaboost M2 comprises the following steps:
given a training set S and a base classifier space Ψ, the expression is as follows:
Figure FDA0002807777430000011
wherein z isiIs the row vector of data Z, y is the class label, c is the fault class number, phi represents the basic classifier;
training set weights are initialized, and the expression is as follows:
Figure FDA0002807777430000012
for each weak classifier, dividing the sample space Z to obtain Z1,Z2,...,ZmIn the probability distribution DfThe sample class attribute probability is calculated as follows:
Figure FDA0002807777430000013
setting the output of a basic classifier in the divided sample space Z, wherein the expression is as follows:
Figure FDA0002807777430000014
calculating to obtain a normalization factor, wherein the expression is as follows:
Figure FDA0002807777430000015
wherein there are F features, F is 1,2, …, F;
the basic classifier is chosen such that the normalization factor is minimal, the expression is as follows:
Figure FDA0002807777430000016
calculating error fraction, and expressing the following expression:
Figure FDA0002807777430000021
wherein, Tm(z) is the sample label predicted by the base classifier;
calculating the weight of each basic classifier, and expressing the following expression:
Figure FDA0002807777430000022
and updating the sample weight, wherein the expression is as follows:
Figure FDA0002807777430000023
obtaining a combined classifier, wherein the expression is as follows:
Figure FDA0002807777430000024
wherein the content of the first and second substances,
Figure FDA0002807777430000025
2. the improved FDA process industry fault diagnosis method based on integrated learning method according to claim 1, wherein the step of preprocessing the obtained fault data comprises: selecting 52 process variables of the Tennessee-Ismann process as input, and carrying out fault diagnosis on 21 faults of the Tennessee-Ismann process;
obtaining a normalized data set X of a failed original training sample0The expression is as follows:
Figure FDA0002807777430000026
wherein n represents the number of samples, and m represents the number of variables;
the step of constructing a dynamic input process comprises: for the normalized data set X0Performing dynamic processing, and selecting a time lag constant L to be 2 to obtain an augmentation matrix X, wherein the expression is as follows:
Figure FDA0002807777430000031
the FDA data dimension reduction and feature extraction steps comprise: and reducing the dimension of the augmentation matrix X, calculating an FDA feature vector, and performing L2 norm normalization.
3. The improved FDA process industrial fault diagnosis method based on ensemble learning method according to claim 2, wherein the step of performing dimension reduction on the augmented matrix X, calculating FDA feature vector, and performing L2 norm normalization comprises:
and constructing an overall dispersion matrix, wherein the expression is as follows:
St＝Sw+Sb (3)
wherein S iswRepresenting an intra-class dispersion matrix, SbRepresenting an inter-class dispersion matrix;
calculating an FDA vector, and maximizing the inter-class dispersion while minimizing the intra-class dispersion, wherein the expression is as follows:
Figure FDA0002807777430000032
obtaining the FDA vector w by solving for the generalized eigenvalues using the Lagrangian multiplier methodkThe expression is as follows:
Sbwk＝λpSwwk (5)
for the FDA vector wkL2 norm normalization is performed, and the expression is as follows:
Figure FDA0002807777430000033
r FDA vectors are selected to construct a linear transformation matrix WnewThe expression is as follows:
Wnew＝(w2 I，w2 2，...，w2 r)T (7)
and (3) performing process industrial data dimension reduction, wherein the expression is as follows:
Z＝WT new×X (8)。
CN202011374338.5A 2020-11-30 2020-11-30 Improved FDA process industrial fault diagnosis method based on ensemble learning method Pending CN112598022A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011374338.5A CN112598022A (en) 2020-11-30 2020-11-30 Improved FDA process industrial fault diagnosis method based on ensemble learning method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011374338.5A CN112598022A (en) 2020-11-30 2020-11-30 Improved FDA process industrial fault diagnosis method based on ensemble learning method

Publications (1)

Publication Number Publication Date
CN112598022A true CN112598022A (en) 2021-04-02

Family

ID=75187312

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011374338.5A Pending CN112598022A (en) 2020-11-30 2020-11-30 Improved FDA process industrial fault diagnosis method based on ensemble learning method

Country Status (1)

Country Link
CN (1) CN112598022A (en)

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105260601A (en) * 2015-10-10 2016-01-20 沈阳化工大学 Polymerization reactor fault diagnosis method based on combination of DKPCA (dynamic kernel principal component analysis) and FDA (Fisher's discriminant analysis)
CN106649789A (en) * 2016-12-28 2017-05-10 浙江大学 Integrated semi-supervised Fisher's discrimination-based industrial process fault classifying method
CN106843195A (en) * 2017-01-25 2017-06-13 浙江大学 Based on the Fault Classification that the integrated semi-supervised Fei Sheer of self adaptation differentiates
CN108446529A (en) * 2018-06-22 2018-08-24 太原理工大学 Organic rankine cycle system fault detection method based on broad sense cross-entropy-DPCA algorithms
CN109583482A (en) * 2018-11-13 2019-04-05 河海大学 A kind of infrared human body target image identification method based on multiple features fusion Yu multicore transfer learning
CN109784356A (en) * 2018-07-18 2019-05-21 北京工业大学 Matrix variables based on Fisher discriminant analysis are limited Boltzmann machine image classification method
CN110765587A (en) * 2019-09-30 2020-02-07 北京化工大学 Complex petrochemical process fault diagnosis method based on dynamic regularization judgment local retention projection
CN111507504A (en) * 2020-03-18 2020-08-07 中国南方电网有限责任公司 Adaboost integrated learning power grid fault diagnosis system and method based on data resampling
CN111738309A (en) * 2020-06-03 2020-10-02 哈尔滨工业大学 Gas sensor fault mode identification method based on multi-scale analysis and integrated learning

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105260601A (en) * 2015-10-10 2016-01-20 沈阳化工大学 Polymerization reactor fault diagnosis method based on combination of DKPCA (dynamic kernel principal component analysis) and FDA (Fisher's discriminant analysis)
CN106649789A (en) * 2016-12-28 2017-05-10 浙江大学 Integrated semi-supervised Fisher's discrimination-based industrial process fault classifying method
CN106843195A (en) * 2017-01-25 2017-06-13 浙江大学 Based on the Fault Classification that the integrated semi-supervised Fei Sheer of self adaptation differentiates
CN108446529A (en) * 2018-06-22 2018-08-24 太原理工大学 Organic rankine cycle system fault detection method based on broad sense cross-entropy-DPCA algorithms
CN109784356A (en) * 2018-07-18 2019-05-21 北京工业大学 Matrix variables based on Fisher discriminant analysis are limited Boltzmann machine image classification method
CN109583482A (en) * 2018-11-13 2019-04-05 河海大学 A kind of infrared human body target image identification method based on multiple features fusion Yu multicore transfer learning
CN110765587A (en) * 2019-09-30 2020-02-07 北京化工大学 Complex petrochemical process fault diagnosis method based on dynamic regularization judgment local retention projection
CN111507504A (en) * 2020-03-18 2020-08-07 中国南方电网有限责任公司 Adaboost integrated learning power grid fault diagnosis system and method based on data resampling
CN111738309A (en) * 2020-06-03 2020-10-02 哈尔滨工业大学 Gas sensor fault mode identification method based on multi-scale analysis and integrated learning

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
YAN-LIN HE.ET AL: "Fault diagnosis using novel AdaBoost based discriminant locality preserving projection with resample", 《ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE》, 8 April 2020 (2020-04-08), pages 1 - 11 *
侯杰等: "基于FDA的快速haar特征选取及其在级联AdaBoost人脸检测中的应用", 《中国自动化学会控制理论专业委员会D卷》, 31 December 2011 (2011-12-31), pages 3234 - 3238 *
陈定三等: "一种基于改进加权粗糙集的多模型软测量建模方法", 《化工自动化及仪表》, no. 01, 10 January 2010 (2010-01-10) *
魏雪倩等: "基于AdaBoost多分类算法变压器故障诊断", 《西安工程大学学报》, no. 02, 31 December 2016 (2016-12-31) *

Similar Documents

Publication Publication Date Title
Ma et al. A novel bearing fault diagnosis method based on 2D image representation and transfer learning-convolutional neural network
Li et al. Fault diagnosis of rotating machinery based on combination of deep belief network and one-dimensional convolutional neural network
Xiong et al. A Wasserstein gradient-penalty generative adversarial network with deep auto-encoder for bearing intelligent fault diagnosis
Zheng et al. A rolling bearing fault diagnosis method based on multi-scale fuzzy entropy and variable predictive model-based class discrimination
Qian et al. An intelligent fault diagnosis framework for raw vibration signals: adaptive overlapping convolutional neural network
Guo et al. Deep convolution feature learning for health indicator construction of bearings
CN108564107A (en) The sample class classifying method of semi-supervised dictionary learning based on atom Laplce's figure regularization
Zheng et al. Sucker rod pumping diagnosis using valve working position and parameter optimal continuous hidden Markov model
CN108764265A (en) A kind of method for diagnosing faults based on algorithm of support vector machine
CN110765587A (en) Complex petrochemical process fault diagnosis method based on dynamic regularization judgment local retention projection
CN108717149A (en) Diagnosis Method of Transformer Faults based on M-RVM fusion dynamic weightings AdaBoost
CN105116397A (en) Radar high-resolution range profile target recognition method based on MMFA model
Guo et al. Improved adversarial learning for fault feature generation of wind turbine gearbox
CN112881942B (en) Abnormal current diagnosis method and system based on wavelet decomposition and empirical mode decomposition
Yang et al. Enhanced hierarchical symbolic dynamic entropy and maximum mean and covariance discrepancy-based transfer joint matching with Welsh loss for intelligent cross-domain bearing health monitoring
Masood et al. Differential evolution based advised SVM for histopathalogical image analysis for skin cancer detection
CN108898157B (en) Classification method for radar chart representation of numerical data based on convolutional neural network
CN112598022A (en) Improved FDA process industrial fault diagnosis method based on ensemble learning method
Sipola et al. Gear classification and fault detection using a diffusion map framework
CN112183569A (en) FDA and SOM based intermittent industrial process reaction phase clustering and fault classification visualization
Baruque et al. THE S 2-ENSEMBLE FUSION ALGORITHM
Ye et al. Monitoring and diagnosis of multi-channel profile data based on uncorrelated multilinear discriminant analysis
Zhu et al. Novel imbalanced fault diagnosis method based on CSMOTE integrated with LSDA and LightGBM for industrial process
Yang et al. Convolutional autoencoder-based sensor fault classification
CN114139638A (en) Fan blade icing fault diagnosis method considering multivariable correlation

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