CN113219942B - Blast furnace fault diagnosis method based on weighted joint distribution adaptive neural network - Google Patents
Blast furnace fault diagnosis method based on weighted joint distribution adaptive neural network Download PDFInfo
- Publication number
- CN113219942B CN113219942B CN202110441962.0A CN202110441962A CN113219942B CN 113219942 B CN113219942 B CN 113219942B CN 202110441962 A CN202110441962 A CN 202110441962A CN 113219942 B CN113219942 B CN 113219942B
- Authority
- CN
- China
- Prior art keywords
- blast furnace
- data
- historical data
- fault
- neural network
- 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
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0243—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults model based detection method, e.g. first-principles knowledge model
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/20—Pc systems
- G05B2219/24—Pc safety
- G05B2219/24065—Real time diagnostics
Abstract
The invention discloses a blast furnace fault diagnosis method based on a weighted joint distribution adaptive neural network. Firstly, performing first-time feature extraction on the historical data and the data to be detected of the blast furnace by using a deep neural network and generating a label of the data to be detected. Based on the label value, the ratio of the to-be-measured data of the blast furnace to the number of samples of the corresponding fault category in the historical data is calculated, and the obtained ratio is used as the corresponding weight of each type of fault of the blast furnace to be combined with a joint distribution adaptation method. And finishing the extraction of the second characteristic by weighted joint distribution adaptation and obtaining a new label value. And finally, carrying out iterative solution on the process of generating labels in the weighted joint distribution adaptation, calculating weights and updating parameters to obtain a fault diagnosis result. The invention not only utilizes the deep neural network to improve the diagnosis precision, but also solves the problem of low accuracy of the traditional fault diagnosis method caused by less fault samples of the blast furnace and larger fluctuation of data distribution along with the change of working conditions through the adaptation of combined distribution and prior distribution.
Description
Technical Field
The invention belongs to the field of industrial process monitoring, modeling and simulation, and particularly relates to a blast furnace fault diagnosis method based on a weighted joint distribution adaptive neural network.
Background
Blast furnace ironmaking is a core unit for material flow conversion of ferrite and is the link with the largest energy consumption and the highest production cost in the steel manufacturing process. With the continuous progress of the process and the technology in the blast furnace ironmaking process and the continuous development of the instruments and the automation technology, the modern blast furnace ironmaking has the characteristics of large scale, complex structure, strong coupling between production units, huge investment and the like. Abnormal fluctuation (or accident) in the blast furnace ironmaking process is discovered untimely, which often causes serious reduction of product quality or delays normal execution of a production plan, and causes huge economic loss and even casualties. Therefore, the fault diagnosis of the blast furnace has important significance for ensuring the safe and high-efficiency production of the blast furnace.
In the actual production process of blast furnace iron making, in order to avoid serious consequences, when a certain fault precursor occurs in the operation of a blast furnace system, an operator can adjust an air supply system, a material distribution system or a furnace heat system so as to avoid the occurrence of faults. Therefore, under the existing operating system and operating conditions, the fault diagnosis system for constructing the blast furnace has the problems of few fault samples, unbalanced data, missing marks, expensive and time-consuming sample labeling and the like. In addition, because the production place of raw materials is not fixed, the blast furnace ironmaking feeding of domestic steel mills mostly adopts a 'hundred-family mine' form. At different times, the type and the proportion of the fed materials can be obviously changed; secondly, the production operation process of the blast furnace has various working condition switching. Due to the factors, the blast furnace data changes along with the time change, the data distribution fluctuation is large, the distribution difference exists between the training data and the data to be detected, and the reliability and the accuracy of fault diagnosis are influenced.
The existing fault diagnosis methods applied to the blast furnace can be roughly divided into two types, namely an expert system and an intelligent fault diagnosis method based on data driving, wherein the expert system has higher requirements on prior knowledge such as related knowledge and rules, the physical and chemical reactions involved in the blast furnace are extremely complex, and the accurate condition of the internal actual reaction is difficult to know. And with the widespread use of various intelligent instruments and control devices of distributed control systems in modern industrial processes, a large amount of process data is collected and stored. However, these data containing process operating state information are often not utilized efficiently in expert systems.
On the other hand, the successful application of the traditional intelligent fault diagnosis method based on data driving has two preconditions: 1) Large amount of labeled data, 2) training and test data from the same data distribution. However, in the production process of the blast furnace, the labeled fault samples are few and difficult to obtain, and due to the fact that the grade of ore raw materials is different from the production working condition, data can fluctuate greatly, and therefore training data and testing data cannot meet the condition of the same distribution. Therefore, the existing abnormal furnace condition diagnosis method has a large gap from ideal practical application.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention aims to provide a blast furnace fault diagnosis method based on a weighted joint distribution adaptive neural network. The method comprises the steps of firstly, using a neural network to carry out first-time feature extraction on historical data and to-be-detected data of a blast furnace and generate an initial tag of the to-be-detected data. Based on the label value, the ratio of the to-be-measured data of the blast furnace to the number of samples of the corresponding fault category in the historical data is calculated, and the obtained ratio is used as the corresponding weight of each type of fault of the blast furnace to be combined with a joint distribution adaptation method. And finishing the extraction of the second characteristic by weighted joint distribution adaptation and obtaining a new label value. And finally, carrying out iterative solution on the process of generating labels in the weighted joint distribution adaptation, calculating weights and updating parameters to obtain a fault diagnosis result. The method can be widely applied to industrial systems with high reliability and accuracy requirements on fault diagnosis.
A blast furnace fault diagnosis method based on a weighted joint distribution adaptive neural network comprises the following steps:
the method comprises the following steps: performing weight training on a deep neural network by using blast furnace historical data and blast furnace data to be tested, wherein the data value obtained by the last full-connection layer in the neural network is the extracted feature, taking the sum of the fault diagnosis error of the blast furnace historical data and the distance between the extracted features of the two groups of data as a loss function, and fixing the weight after the training reaches a preset iteration number or the loss function is lower than a preset value;
step two: performing first feature extraction on the historical data and the data to be detected of the blast furnace by adopting a neural network fixed in the first step, and generating an initial label value on the data to be detected of the blast furnace, namely preliminarily forming a nonlinear mapping from process variables of the blast furnace to fault categories of the blast furnace by utilizing blast furnace fault diagnosis knowledge learned from the historical data and the data to be detected of the blast furnace on the basis of performing feature migration on the historical data and the data to be detected of the blast furnace;
step three: respectively calculating the proportion of the to-be-detected data of the blast furnace to each fault category in the historical data based on the label value of the to-be-detected data, comparing the proportion of the to-be-detected data of the blast furnace to the proportion of the corresponding category in the historical data, multiplying the obtained ratio serving as the category prior distribution weight of the to-be-detected data of the blast furnace by the corresponding blast furnace historical data feature extracted in the step two, and forming a feature variable matrix together with the feature of the to-be-detected data of the blast furnace extracted in the step two, wherein after weighting, the fault category distribution of the blast furnace historical data tends to be consistent with the fault category distribution of the to-be-detected data of the blast furnace, and the prior category distribution adaptation of the blast furnace historical data and the fault category distribution of the to-be-detected data of the blast furnace is realized;
step four: introducing a kernel method, mapping the characteristic variables to obtain new characteristic variables, and transforming the characteristic variables in a kernel space to ensure that the sum of distances of characteristic vectors extracted from the blast furnace historical data and the data to be tested on edge distribution and condition distribution is minimum, thereby realizing the joint distribution adaptation of the blast furnace historical data and the data to be tested by the kernel method and a transformation matrix method;
step five: finally, the transformed characteristic variables are used as input of a classifier, the connection weight between the characteristic variables and the classifier needs to be trained by taking the classification accuracy as a target function, and after convergence, the classification result of the classifier on the data to be detected, namely the blast furnace fault category, is used as a new label value to be distributed to the data to be detected;
step six: and circularly iterating the third step to the fifth step until the distance of the blast furnace historical data and the characteristic vector of the data to be detected on the joint distribution and the classification accuracy tend to be stable, fixing the model parameters, and carrying out judgment processing on the data to be detected to generate a fault diagnosis result.
The structure of the deep neural network in the step one is as follows: the deep neural network comprises three parts, namely an input layer, a hidden layer and an output layer, wherein the input layer is a blast furnace process variable parameter input layer and comprises industrial process parameters representing the production state of a blast furnace, such as air permeability index, cold air flow, hot air flow, top pressure, cold air pressure, hot air pressure and the like, the output layer is a blast furnace fault category layer and comprises blast furnace faults related to the production process of the blast furnace, such as difficulty, hanging materials, pipelines, material collapse, furnace heat, furnace cool and the like, and the hidden layer is used for establishing a nonlinear mapping from blast furnace process variables to blast furnace fault categories, so that blast furnace fault diagnosis knowledge can be learned from blast furnace historical fault data, and a blast furnace fault diagnosis model can be established. The neurons in the same layer are not connected, the neurons between layers are fully connected, and each connection has a weight value to represent the strength of the connection degree between the neurons. For different industrial application fields, the requirements on the number of layers of hidden layers of a deep neural network are different, the neural network with the hidden layer being more than or equal to 2 is defined as the deep neural network, and the mathematical model of the deep neural network is as follows:
wherein the content of the first and second substances,
for the output of the ith hidden layer unit of the ith layer of the neural network, note h i At the ith layer of the neural network, h 0 As the input layer of the neural network, h k+1 Outputting a layer for the neural network; j is determined according to the number of the neurons of the ith layer of the network, and the number of the neurons of the ith layer is recorded as z i Then each layer j takes on a value from 1 to z i (ii) a W (i, j) is a weight matrix corresponding to the jth neuron of the ith layer;
bias term for the jth neuron at level i, b k+1 Bias terms corresponding to the output layer units; y represents the output of the neural network, M is the total number of samples of the blast furnace historical data, N is the total number of samples of the blast furnace data to be measured, f (-) and g (-) are the activation functions of the hidden layer unit and the output unit respectively,
represents the firstMaximum of i samples in output layer neurons, s j Representing the value of the jth neuron in the output layer, extracting the data of the full connection layer as characteristic vectors, and respectively recording the extracted characteristic vectors of blast furnace historical data and data to be detected as x s And x t Taking the sum of the distance between the fault diagnosis error of the blast furnace historical data and the extracted features of the two groups of data as a loss function, namely the following formula:
the weighting steps in the third step are as follows: recording the C type of the blast furnace faults, wherein C represents the type of the blast furnace faults, and when C is a real number from 1 to C, the type represents the corresponding specific fault type, such as: pipeline, descending, difficult operation, suspension and the like, wherein M is the total number of samples of the historical data of the blast furnace, and the number of samples belonging to the type c fault is M C Correspondingly, N is the total number of samples of the data to be tested of the blast furnace, wherein the number of samples belonging to the type c fault is N C The tag value of the history data is recorded as y s And the label value of the data to be tested is recorded as y t And the distribution of the blast furnace historical data and the data to be measured is respectively marked as p s (. O) and p t The ratios of various fault samples in the blast furnace historical data and the data to be detected are respectively as follows:
the corresponding weight of each type of fault data in the blast furnace historical data is
After multiplying the weight by the blast furnace historical data, the prior distribution of the blast furnace historical data is as follows:
therefore, the fault category distribution of the blast furnace historical data and the category prior distribution of the blast furnace data to be detected tend to be consistent, and the prior category distribution adaptation of the blast furnace historical data and the category prior distribution adaptation of the blast furnace data to be detected are realized. After weighting, the blast furnace historical data characteristic matrix X s And the characteristic matrix X of the data to be measured t A feature matrix X is formed.
Step four, the step of core mapping and joint distribution adaptation is as follows: selecting a kernel function such as a Gaussian kernel function, and performing nonlinear mapping on the features, namely:
wherein
In the case of a nonlinear mapping function, the distance between the blast furnace historical data and the data to be measured in the joint distribution in the nuclear space is:
introducing a kernel matrix K:
wherein:
and if the obtained transformation matrix is W, the joint distribution distance of the blast furnace historical data and the data to be measured is as follows:
in conjunction with the kernel matrix, the minimization problem of the above equation can be translated into:
s.t W T KHK T W＝I
wherein: h = I M+N -1/(M+N)11 T
The solution of this equation, the transformation matrix W, is obtained by eigenvalue decomposition.
The step six of the iterative updating comprises the following steps: and (4) carrying out iterative solution on the steps from the third step to the fifth step, namely carrying out iteration on the process of generating labels in the weighted combined distribution adaptation, calculating the weight and updating the parameters to obtain a fault diagnosis result.
The invention has the beneficial effects that:
aiming at the characteristics and the problems of large data fluctuation, label loss, unbalanced data and the like in the blast furnace ironmaking process, the blast furnace fault diagnosis method based on the weighted joint distribution adaptive neural network is constructed, the blast furnace data is subjected to the adaptation of prior distribution and joint distribution, the knowledge contained in the data is fully mined, the problems that the blast furnace has more historical data and is difficult to directly train a model for the data to be tested are solved, the advantages of high reliability and high accuracy are realized, and the automation and intelligence level of the ironmaking process is improved.
Drawings
FIG. 1 is a block flow diagram of the method of the present invention.
FIG. 2 shows the result display of the original distribution of the data to be measured visualized by t-sne.
FIG. 3 shows the visualized result display of t-sne after the method of the present invention classifies the blast furnace fault of the data to be measured.
Detailed Description
The invention is further elucidated with reference to the following figures.
The invention aims to provide a blast furnace fault diagnosis method based on a weighted joint distribution adaptive neural network, a flow diagram is shown in figure 1, the nonlinear and non-Gaussian properties of blast furnace production information are considered, a deep neural network can infinitely approximate the advantages of a nonlinear function to carry out first-time feature extraction, the edge distribution distance between blast furnace historical data and data to be detected is measured through a maximum average distance (MMD), the sum of classification accuracy and distance is used as a loss function to carry out network training and weight fixing, a full connection layer of the deep neural network is used as a feature vector extracted by the neural network, the classification result of the neural network is used as an initial label value of the neural network, the ratio of the number of corresponding fault category samples of the blast furnace historical data and the data to be detected is used as a category prior distribution weight to carry out weighting, then the feature vector is mapped in a kernel function mode, the distance of the joint distribution of the feature vector of the blast furnace historical data and the data to be detected is measured through a weighted joint maximum average distance (WJMMD), a transformation matrix is obtained through minimum finding, and a new classification result is obtained through a classifier (softm) of the distance. And finally, carrying out iterative solution on the process of generating labels in the weighted joint distribution adaptation, calculating weights and updating parameters to obtain a fault diagnosis result. The method is beneficial to realizing the enhancement of knowledge and decision in the fault diagnosis of the blast furnace and ensuring the reliability and the accuracy of the fault diagnosis of the blast furnace.
The effectiveness of the method is verified by using the blast furnace fault data collected by the No. 2 blast furnace of a certain steel plant. The blast furnace is divided into five parts, namely a furnace throat, a furnace body, a furnace waist, a furnace belly and a furnace hearth from top to bottom, and different parts in the furnace can undergo different changes in the descending process of coke, ore and flux until the bottom of the furnace hearth is reached and the coke, the ore and the flux are completely converted into molten iron and slag. Because of the large size of the blast furnace and the complex chemical reaction in the furnace, it is very important to ensure the safe and stable operation of the blast furnace. Blast furnace faults are mainly classified into 4 types: difficult (difficult), hanging (hanging), piping (channeling), disintegrating (collapsing). The data collected in the production process comprises 29 parameters such as air permeability index, cold air flow, hot air flow, top pressure, cold air pressure, hot air pressure and the like. In actual production, a three-shift system is adopted to organize workers to monitor and operate and manage the blast furnace ironmaking process, so that great labor cost is consumed, the control mode is relatively extensive, furnace condition judgment is mainly carried out by means of a plurality of parameters, and problems existing in the blast furnace operation process are difficult to diagnose in time and accurate control is carried out in time. The method can solve the problem to a certain extent and has practical application value.
The following describes the implementation steps of the present invention in detail with reference to the specific process.
1. Constructing a deep neural network for the first feature extraction and generating an initial label value
(1) The structure of the deep neural network is shown in a neural network characteristic extraction part in figure 1 and comprises an input layer, a hidden layer and an output layer, wherein the input layer is a blast furnace process variable parameter input layer and comprises industrial process parameters representing the production state of a blast furnace, such as air permeability index, cold air flow, hot air flow, top pressure, cold air pressure, hot air pressure and the like, the output layer is a blast furnace fault category layer and comprises blast furnace faults related to the production process of the blast furnace, such as difficulty, hanging materials, pipelines, material collapse, furnace heat, furnace cool and the like, and the hidden layer is used for establishing a nonlinear mapping from the blast furnace process variables to the blast furnace fault categories, so that blast furnace fault diagnosis knowledge can be learned from blast furnace historical fault data and a blast furnace fault diagnosis model can be established. The neurons in the same layer are not connected, the neurons between layers are fully connected, and each connection has a weight value to represent the strength of the connection degree between the neurons.
For different industrial application fields, the requirements on the number of layers of hidden layers of a deep neural network are different, the neural network with the hidden layer being more than or equal to 2 is defined as the deep neural network, and the mathematical model of the deep neural network is as follows:
wherein the content of the first and second substances,
for the output of the ith hidden layer unit of the ith layer of the neural network, note h i At the ith layer of the neural network, h 0 As the input layer of the neural network, h k+1 Outputting a layer for the neural network; j is determined according to the number of the neurons of the ith layer of the network, and the number of the neurons of the ith layer is recorded as z i Then each layer j takes on a value of 1 to z i (ii) a W (i, j) is a weight matrix corresponding to the jth neuron of the ith layer;
bias term for the jth neuron at level i, b k+1 A bias term corresponding to the output layer unit; y represents the output of the neural network, M is the total number of samples of the blast furnace historical data, N is the total number of samples of the blast furnace data to be measured, f (-) and g (-) are the activation functions of the hidden layer unit and the output unit respectively,
represents the maximum value of the ith sample in the output layer neurons, s j Representing the value of the jth neuron in the output layer, extracting the data of the full connection layer as characteristic vectors, and respectively recording the extracted characteristic vectors of blast furnace historical data and data to be detected as x s And x t Taking the sum of the fault diagnosis error of the blast furnace historical data and the distance between the extracted features of the two groups of data as a loss function, namely the following formula:
(2) And (2) performing first feature extraction on the historical data and the data to be detected of the blast furnace by adopting the neural network fixed in the step (1) and generating an initial label value for the data to be detected of the blast furnace, namely preliminarily forming a nonlinear mapping from the process variables of the blast furnace to the fault category of the blast furnace by using the fault diagnosis knowledge of the blast furnace learned from the historical data and the data to be detected of the blast furnace on the basis of performing feature migration on the historical data and the data to be detected of the blast furnace.
2. Weighting the extracted feature vectors to complete class prior distribution adaptation
The weighting steps of the characteristic vectors extracted from the blast furnace historical data are as follows: recording the C type of the blast furnace faults, wherein C represents the type of the blast furnace faults, and when C takes a real number from 1 to C, the type of the corresponding specific faults is represented, such as: pipelines, descending, difficult operation, hanging materials and the like, wherein M is the total number of samples of the historical data of the blast furnace, and the number of the samples belonging to the type c fault is M C Correspondingly, N is the total number of samples of the data to be tested of the blast furnace, wherein the number of samples belonging to the type c fault is N C The tag value of the history data is recorded as y s And the label value of the data to be tested is recorded as y t And the distribution of the blast furnace historical data and the data to be measured is respectively marked as p s (. Cndot.) with p t The ratio of various fault samples in the blast furnace historical data to the data to be detected is respectively as follows:
the corresponding weight of each type of fault data in the blast furnace historical data is
After multiplying the weight by the blast furnace historical data, the prior distribution of the blast furnace historical data is as follows:
therefore, the fault class distribution of the blast furnace historical data and the class prior distribution of the blast furnace data to be measured tend to be consistent, and the prior class distribution adaptation of the blast furnace historical data and the class prior distribution adaptation of the blast furnace data to be measured are realized. After weighting, the blast furnace historical data characteristic matrix X s And the characteristic matrix X of the data to be measured t A feature matrix X is formed.
3. Kernel mapping and joint distribution adaptation for feature vectors
(1) Selecting a kernel function such as a Gaussian kernel function, and performing nonlinear mapping on the features, namely:
wherein
And if the mapping function is a nonlinear mapping function, the distance between the blast furnace historical data and the data to be measured in the nuclear space in the joint distribution is as follows:
introducing a kernel matrix K:
wherein:
and if the obtained transformation matrix is W, the joint distribution distance of the blast furnace historical data and the data to be measured is as follows:
in conjunction with the kernel matrix, the minimization problem of the above equation can be translated into:
s.t W T KHK T W＝I
wherein: h = I M+N -1/(M+N)11 T
The solution of this equation, the transformation matrix W, is obtained by eigenvalue decomposition.
(2) And taking the transformed characteristic variable as the input of a classifier, training the connection weight between the characteristic variable and the classifier by taking the classification accuracy as a target function, and distributing the classification result of the classifier on the data to be detected, namely the blast furnace fault category, as a new label value to the data to be detected after convergence.
4. Performing iterative solution
And performing loop iteration on the step two and the step three, namely performing iteration on the process of generating labels in the weighted joint distribution adaptation, calculating the weight and updating the parameters to obtain a fault diagnosis result.
5. Substituting industrial actual data for verification
We take the volume of a certain ironworks as 2650m 3 The production data of the blast furnace No. 2 in 10 months in 2020 is used as blast furnace historical data, namely source domain data, and the blast furnace production data in 12 months is used as target domain data to be measured, wherein the target domain data comprises 29 parameters in total, namely oxygen enrichment rate ', ' air permeability index ', ' CO ', ' H2', ' CO2', ' standard wind speed ', ' oxygen enrichment flow ', ' cold wind flow ', ' air blast kinetic energy ', ' furnace belly coal gas quantity ', ' furnace belly coal gas index ', ' theoretical combustion temperature ', ' top pressure ', ' oxygen enrichment pressure ', ' cold wind pressure ', ' total pressure difference ', ' hot wind pressure ', ' actual wind speed ', ' cold wind temperature ', ' hot wind temperature ', ' top temperature northeast ', ' top temperature southwest ', ' top temperature northwest ', ' top temperature southwest ', ' top temperature down pipe ', ' resistance coefficient of air blast humidity ', ' actual coal injection amount ' in this hour, and ' actual coal injection amount in last hour ', and the sampling rate are consistent. And carrying out validity verification on the model obtained by training in the data to be tested, namely the blast furnace production data of 12 months.
FIGS. 2 and 3 show the original distribution of the blast furnace data to be tested in 12 months, and the visualization result display of the blast furnace data to be tested in t-sne after the blast furnace fault classification by the method of the present invention. The fault diagnosis result shows that the model has good effect. The classification effect is obvious, can accurately classify the blast furnace fault samples, and therefore can be applied to actual industrial production.
Claims (5)
1. A blast furnace fault diagnosis method based on a weighted joint distribution adaptive neural network is characterized by comprising the following steps:
the method comprises the following steps: performing weight training on a deep neural network by using blast furnace historical data and blast furnace data to be tested, wherein a data value obtained by the last full-connection layer in the deep neural network is the extracted feature, taking the sum of the fault diagnosis error of the blast furnace historical data and the distance between the extracted features as a loss function, and fixing the weight after the training reaches a preset iteration number or the loss function is lower than a preset value;
step two: performing first-time feature extraction on the historical data and the data to be detected of the blast furnace by adopting the fixed deep neural network in the first step and generating a label value on the data to be detected of the blast furnace, namely preliminarily forming a nonlinear mapping from process variables of the blast furnace to fault categories of the blast furnace by utilizing blast furnace fault diagnosis knowledge learned from the historical data and the data to be detected of the blast furnace on the basis of performing feature migration on the historical data and the data to be detected of the blast furnace;
step three: respectively calculating the proportion of the to-be-detected data of the blast furnace to each fault category in the historical data based on the label value of the to-be-detected data, comparing the proportion of the to-be-detected data of the blast furnace to the proportion of the corresponding category in the historical data, multiplying the obtained ratio serving as the category prior distribution weight of the to-be-detected data of the blast furnace by the corresponding blast furnace historical data feature extracted in the step two, and forming a feature variable matrix together with the feature of the to-be-detected data of the blast furnace extracted in the step two, wherein after weighting, the fault category distribution of the blast furnace historical data tends to be consistent with the fault category distribution of the to-be-detected data of the blast furnace, and the prior category distribution adaptation of the blast furnace historical data and the fault category distribution of the to-be-detected data of the blast furnace is realized;
step four: introducing a kernel method, mapping the characteristic variables to obtain new characteristic variables, and transforming the characteristic variables in a kernel space to ensure that the sum of distances of characteristic vectors extracted from the blast furnace historical data and the data to be tested on edge distribution and condition distribution is minimum, thereby realizing the joint distribution adaptation of the blast furnace historical data and the data to be tested by the kernel method and a transformation matrix method;
step five: taking the transformed characteristic variable as the input of a classifier, training the connection weight between the characteristic variable and the classifier by taking the classification accuracy as a target function, and distributing the classification result of the classifier on the data to be detected, namely the blast furnace fault category, as a new label value to the data to be detected after convergence;
step six: and circularly iterating the third step to the fifth step until the distance of the blast furnace historical data and the characteristic vector of the data to be detected on the joint distribution and the classification accuracy tend to be stable, fixing the model parameters, and carrying out judgment processing on the data to be detected to generate a fault diagnosis result.
2. The method of claim 1, wherein the deep neural network of step one has the following structure: the deep neural network comprises three parts, namely an input layer, a hidden layer and an output layer, wherein the input layer is a blast furnace process variable parameter input layer and comprises industrial process parameters of air permeability indexes, cold air flow, hot air flow, top pressure, cold air pressure and hot air pressure representing the production state of the blast furnace, the output layer is a blast furnace fault category layer and comprises blast furnace faults related to the production process of the blast furnace, such as difficulty, hanging materials, pipelines, material collapse, furnace heat and furnace cool, the hidden layer is used for establishing a nonlinear mapping from blast furnace process variables to blast furnace fault categories, learning blast furnace fault diagnosis knowledge from blast furnace historical fault data and establishing a blast furnace fault diagnosis model; the neurons in the same layer are not connected, the neurons between layers are all connected, and each connection has a weight value to represent the strength of the connection degree between the neurons; defining a neural network with a hidden layer more than or equal to 2 as a deep neural network, wherein the mathematical model of the deep neural network is as follows:
wherein the content of the first and second substances,
for the output of the ith hidden layer unit of the ith layer of the neural network, note h i At the ith layer of the neural network, h 0 As the input layer of the neural network, h k+1 A neural network output layer; j is determined according to the number of the neurons of the ith layer of the network, and the number of the neurons of the ith layer is recorded as z i Then each layer j takes on a value of 1 to z i (ii) a W (i, j) is a weight matrix corresponding to the jth neuron of the ith layer;
bias term for the jth neuron at level i, b k+1 Bias terms corresponding to the output layer units; y represents the output of the neural network, M is the total number of samples of the blast furnace historical data, N is the total number of samples of the blast furnace data to be measured, f (-) and g (-) are the activation functions of the hidden layer unit and the output unit respectively,
represents the maximum value of the ith sample in the output layer neurons, s j Representing the value of the jth neuron in the output layer, extracting the data of the full connection layer as characteristic vectors, and respectively recording the extracted characteristic vectors of the blast furnace historical data and the data to be detected as x s And x t Taking the sum of the fault diagnosis error of the blast furnace historical data and the distance between the extracted features of the two groups of data as a loss function, namely the following formula (3):
3. the method of claim 1, wherein the weighting step in step three is as follows: recording C types of blast furnace faults, wherein C represents the types of the blast furnace faults, when C takes a real number from 1 to C, the types of the corresponding specific faults are represented, the types of the specific faults comprise pipelines, descending, difficult traveling and hanging materials, M is the total number of samples of the historical data of the blast furnace, and the number of the samples belonging to the C types of the faults isIs M C Correspondingly, N is the total number of samples of the data to be tested of the blast furnace, wherein the number of samples belonging to the type c fault is N C The tag value of the history data is noted as y s And the label value of the data to be tested is recorded as y t And the distribution of the blast furnace historical data and the data to be measured is respectively recorded as p s (. O) and p t The ratio of various fault samples in the blast furnace historical data to the data to be detected is respectively as follows:
the corresponding weight of each fault data in the blast furnace historical data is
After multiplying the weight by the blast furnace historical data, the prior distribution of the blast furnace historical data is as follows:
therefore, the fault category distribution of the blast furnace historical data and the category prior distribution of the blast furnace data to be tested tend to be consistent, the prior category distribution adaptation of the fault category distribution and the category prior distribution are realized, and after weighting, the blast furnace historical data characteristic matrix X s And the characteristic matrix X of the data to be measured t A feature matrix X is formed.
4. The method of claim 1, wherein the step of adapting the kernel method to the joint distribution in the step four comprises: selecting a kernel function such as a Gaussian kernel function, and performing nonlinear mapping on the features, namely:
wherein
And if the mapping function is a nonlinear mapping function, the distance of the joint distribution of the blast furnace historical data and the data to be measured in the nuclear space is as follows:
introducing a kernel matrix K:
wherein:
and if the obtained transformation matrix is W, the joint distribution distance of the blast furnace historical data and the data to be measured is as follows:
in conjunction with the kernel matrix, the minimization problem of equation (14) can be translated into:
s.t W T KHK T W＝I (15)
wherein: h = I M+N -1/(M+N)11 T (16)
The transformation matrix in the eigenspace is solved by eigenvalue decomposition.
5. The method of claim 1, wherein the loop iteration of step six comprises the steps of: and (4) carrying out iterative solution on the steps from the third step to the fifth step, namely carrying out iteration on the process of generating labels in the weighted combined distribution adaptation, calculating the weight and updating the parameters to obtain a fault diagnosis result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110441962.0A CN113219942B (en) | 2021-04-23 | 2021-04-23 | Blast furnace fault diagnosis method based on weighted joint distribution adaptive neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110441962.0A CN113219942B (en) | 2021-04-23 | 2021-04-23 | Blast furnace fault diagnosis method based on weighted joint distribution adaptive neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113219942A CN113219942A (en) | 2021-08-06 |
| CN113219942B true CN113219942B (en) | 2022-10-25 |
Family
ID=77088950
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110441962.0A Active CN113219942B (en) | 2021-04-23 | 2021-04-23 | Blast furnace fault diagnosis method based on weighted joint distribution adaptive neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113219942B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116203929B (en) * | 2023-03-01 | 2024-01-05 | 中国矿业大学 | Industrial process fault diagnosis method for long tail distribution data |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108445861A (en) * | 2018-02-05 | 2018-08-24 | 华南理工大学 | A kind of goat fault detection method and system based on convolutional neural networks algorithm |
| CN108921230A (en) * | 2018-07-25 | 2018-11-30 | 浙江浙能嘉华发电有限公司 | Method for diagnosing faults based on class mean value core pivot element analysis and BP neural network |
| CN109325417A (en) * | 2018-08-23 | 2019-02-12 | 东北大学 | A kind of industrial process fault condition diagnostic method based on deep neural network |
| CN110222830A (en) * | 2019-06-13 | 2019-09-10 | 中国人民解放军空军工程大学 | A kind of depth feedforward network method for diagnosing faults based on self-adapted genetic algorithm optimization |
| KR102027389B1 (en) * | 2019-03-20 | 2019-10-01 | (주)브이엠에스 | Fault diagnosis system of mechanical devices using autoencoder and deep-learning |
| CN111381584A (en) * | 2020-03-25 | 2020-07-07 | 北京航空航天大学 | Aircraft cabin abnormal fault detection method based on two-stage gated loop network associative memory |
| CN111651931A (en) * | 2020-05-19 | 2020-09-11 | 浙江大学 | Blast furnace fault diagnosis rule derivation method based on deep neural network |
| CN111898095A (en) * | 2020-07-10 | 2020-11-06 | 佛山科学技术学院 | Deep migration learning intelligent fault diagnosis method and device, storage medium and equipment |
-
2021
- 2021-04-23 CN CN202110441962.0A patent/CN113219942B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108445861A (en) * | 2018-02-05 | 2018-08-24 | 华南理工大学 | A kind of goat fault detection method and system based on convolutional neural networks algorithm |
| CN108921230A (en) * | 2018-07-25 | 2018-11-30 | 浙江浙能嘉华发电有限公司 | Method for diagnosing faults based on class mean value core pivot element analysis and BP neural network |
| CN109325417A (en) * | 2018-08-23 | 2019-02-12 | 东北大学 | A kind of industrial process fault condition diagnostic method based on deep neural network |
| KR102027389B1 (en) * | 2019-03-20 | 2019-10-01 | (주)브이엠에스 | Fault diagnosis system of mechanical devices using autoencoder and deep-learning |
| CN110222830A (en) * | 2019-06-13 | 2019-09-10 | 中国人民解放军空军工程大学 | A kind of depth feedforward network method for diagnosing faults based on self-adapted genetic algorithm optimization |
| CN111381584A (en) * | 2020-03-25 | 2020-07-07 | 北京航空航天大学 | Aircraft cabin abnormal fault detection method based on two-stage gated loop network associative memory |
| CN111651931A (en) * | 2020-05-19 | 2020-09-11 | 浙江大学 | Blast furnace fault diagnosis rule derivation method based on deep neural network |
| CN111898095A (en) * | 2020-07-10 | 2020-11-06 | 佛山科学技术学院 | Deep migration learning intelligent fault diagnosis method and device, storage medium and equipment |
Non-Patent Citations (1)
| Title |
|---|
| 基于迁移学习与深度卷积的动车组滚动轴承故障诊断方法与研究;王碧瑶;《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》;20210315(第3期);C033-331 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113219942A (en) | 2021-08-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104651559B (en) | Blast furnace liquid iron quality online forecasting system and method based on multivariable online sequential extreme learning machine | |
| CN110066895B (en) | Stacking-based blast furnace molten iron quality interval prediction method | |
| CN104777830B (en) | A kind of multiple operating modes process monitoring method based on KPCA mixed models | |
| CN111651931B (en) | Blast furnace fault diagnosis rule derivation method based on deep neural network | |
| CN106249724B (en) | A kind of polynary molten steel quality forecast Control Algorithm of blast furnace and system | |
| CN105177199B (en) | Blast furnace gas generation amount soft measurement method | |
| CN109935280B (en) | Blast furnace molten iron quality prediction system and method based on ensemble learning | |
| CN104793606B (en) | Industrial method for diagnosing faults based on improved KPCA and HMM | |
| CN107299170B (en) | A kind of blast-melted quality robust flexible measurement method | |
| CN105608492A (en) | Robust random-weight neural network-based molten-iron quality multi-dimensional soft measurement method | |
| CN111652355B (en) | Method and device for predicting silicon content of blast furnace molten iron based on LSTM and DNN | |
| CN103440368A (en) | Multi-model dynamic soft measuring modeling method | |
| Jiang et al. | Classification of silicon content variation trend based on fusion of multilevel features in blast furnace ironmaking | |
| Xie et al. | Robust stochastic configuration network multi-output modeling of molten iron quality in blast furnace ironmaking | |
| CN102925602B (en) | Furnace profile maintenance method for blast furnace operation | |
| CN105574297B (en) | Self adaptation blast furnace molten iron silicon content trend prediction method | |
| CN113219942B (en) | Blast furnace fault diagnosis method based on weighted joint distribution adaptive neural network | |
| CN113761787A (en) | Blast furnace molten iron silicon content online prediction method and system based on deep migration network | |
| CN111444942A (en) | Intelligent forecasting method and system for silicon content of blast furnace molten iron | |
| CN104750902A (en) | Molten iron mass multivariant dynamic soft measurement method based on multi-output support vector regression machine | |
| Zhou et al. | Data-driven dynamic modeling for prediction of molten iron silicon content using ELM with self-feedback | |
| CN109934421B (en) | Blast furnace molten iron silicon content prediction and compensation method for fluctuating furnace condition | |
| Li et al. | A blast furnace coke ratio prediction model based on fuzzy cluster and grid search optimized support vector regression | |
| CN109359320B (en) | Blast furnace index prediction method based on multiple sampling rate autoregressive distribution hysteresis model | |
| CN115496124A (en) | Blast furnace fault diagnosis method based on minimum maximum entropy collaborative training |
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 |