CN112859815B - Method for monitoring and diagnosing abnormal furnace conditions in roasting process - Google Patents
Method for monitoring and diagnosing abnormal furnace conditions in roasting process Download PDFInfo
- Publication number
- CN112859815B CN112859815B CN202110078674.3A CN202110078674A CN112859815B CN 112859815 B CN112859815 B CN 112859815B CN 202110078674 A CN202110078674 A CN 202110078674A CN 112859815 B CN112859815 B CN 112859815B
- Authority
- CN
- China
- Prior art keywords
- variable
- variables
- temperature
- boiling layer
- monitoring
- 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
- 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
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Automation & Control Theory (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses a method for monitoring and diagnosing abnormal furnace conditions in a roasting process, which is a variable selection method based on correlation and redundancy, and is characterized in that autocorrelation and cross correlation variables which can reflect the changes of the furnace conditions of a roasting furnace at different sampling moments are found for key variables which can reflect the changes of the furnace conditions of the roasting furnace by utilizing historical effective data, so that the complex time sequence correlation among process variables is better described, and the redundant variables in the roasting process are obviously reduced; establishing a distributed dynamic PCA monitoring model on the basis of each key variable, and fusing the monitoring results of all sub-models into a probability type monitoring index on the basis of Bayesian inference so as to rapidly judge the occurrence of faults; the invention can quickly and effectively detect the abnormal condition of the roasting furnace, accurately find the fault source position and reduce the influence of the abnormal condition on the running state of the roasting furnace.
Description
Technical Field
The invention relates to the technical field of industrial control, in particular to a method for monitoring and diagnosing abnormal furnace conditions in a roasting process.
Background
The roasting process is a preposed process of the zinc hydrometallurgy process and is also the most important process, and in the process, the mixed zinc concentrate is thrown into a roasting furnace through a thrower to carry out fluidized combustion to generate products such as zinc calcine, smoke and the like. The main purpose of the roasting process is to increase the conversion of zinc sulphide in the zinc concentrate and at the same time to reduce the content of insoluble impurities. In the production process of the roasting furnace, due to the fact that zinc concentrate components are complex and changeable, misoperation of personnel, equipment abnormity and the like occur, the abnormity of the roasting furnace can occur frequently. Once the abnormality occurs in the operation process of the roasting furnace, the impurity content in the product is high, the zinc sulfide conversion rate is low, and shutdown or large-scale production accidents are caused if the abnormality occurs, so that the production benefits of enterprises are directly affected greatly, and further, the immeasurable loss is brought. Therefore, if the furnace condition of the roasting furnace can be accurately forecasted and diagnosed, the monitoring of the furnace condition of the roasting furnace and the accurate forecasting of the abnormal furnace condition are realized, the production benefit can be improved, and the normal operation of the roasting furnace can be guaranteed.
The fault detection and diagnosis method in the industrial process mainly comprises a method based on expert knowledge, a mechanism model and data driving. However, the method based on expert knowledge requires time and labor for establishing a huge expert knowledge base, and the knowledge of experience in some scenes cannot ensure high accuracy and reliability; the method based on the mechanism model needs to establish the mechanism model for the process object, is only suitable for a small system with enough mechanism knowledge, and is greatly limited in application range of large-scale process industry; data-driven based approaches require mining useful information in historical operating data to build a model. Because the components of the zinc concentrate are complex and changeable, and the process requirements are also constantly changed, the reaction inside the roasting furnace is complex and changeable, and therefore, the mechanism model and the experience knowledge also need to be continuously updated, and therefore, the method based on the mechanism model and the expert system based on the artificial experience cannot be applied to abnormal furnace condition detection and diagnosis in the roasting process. The zinc smelting roasting process belongs to a complex production process with multivariable, large lag and strong coupling, the dimension of process variables is high, redundant variables are more, the production period of the whole process is long, so that complex time sequence correlation exists among the variables, and useful information of key variables capable of reflecting the changes of the furnace conditions of the roasting furnace is dispersed, so that the occurrence of abnormal furnace conditions is difficult to detect by directly using the traditional PCA method.
Disclosure of Invention
Technical problem to be solved
Based on the problems, the invention provides a method for monitoring and diagnosing abnormal furnace conditions in the roasting process, which solves the problem that the abnormal furnace conditions in the operation process of a roasting furnace are difficult to timely and accurately detect.
(II) technical scheme
Based on the technical problem, the invention provides a method for monitoring and diagnosing abnormal furnace conditions in a roasting process, which comprises the following steps:
s1, collecting all measurement variables in the history data of the zinc smelting roasting process, and selecting sample data of a normal process from the collected history data as a training data set X belonging to Rn×mN is the number of samples, m is the number of measurement variables, and the measurement variables comprise data acquired by a sensor, including hearth temperature A, hearth temperature B, temperature A on a boiling layer, flue gas outlet temperature, air inlet pipe temperature, flue gas outlet pressure, air volume and boiler barrel water supply flow;
s2, introducing S delay measurement values into each measurement variable in the X to form an augmentation matrix Xa∈R(n -s)×(s+1)mAnd X isaEach variable in (a) is normalized to data of zero mean and unit standard deviation;
s3, analyzing the roasting process technology mechanism and the actual operation data, and determining the content of the X epsilon Rn×mSelecting key variable set X capable of reflecting furnace condition of roasting furnaceL∈RL×m(L ≦ m) and selecting method from said X by correlation and redundancy based variable selection methodaFinding all related variables with small redundancy of each key variable from m (S +1) variables to obtain L variable subsets Si(i=1,2,…L);
S4, using PCA method to perform subset S on the L variablesi(i 1,2, … L) respectively establishing fault detection models, and calculating control limits of the corresponding modelsAnd SPElim(i) The PCA method is a principal component analysis method;
s5, in the process of on-line monitoring, introducing the on-line measured value into the corresponding delay measured value, and according to the L variable subsets S in the step S3i(i-1, 2, … L), and the variables are partitioned into blocks, and the blocks are substituted into the failure detection models, and bayesian inference is used to determine the failure of the failure detection modelsFusing the detection results of all fault detection models into a monitoring index, judging whether the current sample has a fault, if so, entering the step S6, otherwise, replacing the new test sample data, and repeating the step S5;
and S6, finding the fault source by using the weighted contribution graph.
Further, the method for selecting variables based on correlation and redundancy in step S3 includes:
s3.1, determining a key variable set X by analyzing a roasting process technological mechanism and actual operation dataL∈RL×m(L is less than or equal to m) and candidate variable set Xa∈R(n-s)×(s+1)mAnd a set S of related variables for each key variablei(i=1,2,…L);
S3.2 for each key variable xi∈XLComputing x using mutual informationiAnd XaEach variable x inj(j ═ 1,2, …, m (s +1)) correlation I (x)i,xj):
Wherein p (x)i,xj) Is xiAnd xjOf (2) a joint probability density function of p (x)i) And p (x)j) Are respectively xiAnd xjThe edge probability density function of (a);
s3.3 for XaEach variable x injCalculating xjAnd SiRedundancy of all variables in (A), redundancy beta (x)j,Si) The calculation formula of (2):
s3.4, for XaEach variable x injSynthesis of xjAnd xiCorrelation of (2) and xjAnd SiRedundancy of selected variables, calculating xjEvaluation value J (x) ofj,xi) The specific calculation formula of (2) is as follows:
s3.5, for the key variable xiFrom the resulting evaluation value set { J (x)j,xi) Finding the variable corresponding to the maximum value in (j ═ 1,2, …, m (s +1)), that is, the selected variable is: y is argmax (J (x)j,xi) Adding the variable y to SiFrom XaRemoving;
s3.6, for the selected variable SiAll the variables in (1), calculating the sum of the variables and the key variable xiThe calculation formula:
s3.7, calculating D, if D is less than or equal to T, continuing to execute the step S3.8, otherwise, returning to the step S3.2, wherein D is the ratio of the evaluation value of the optimal variable in the candidate variables to the evaluation value of the relevant variable set, T is a threshold value for ending selection,
s3.8, returning to the step S3.1, circulating for L times in total, and finding each key variable xiSet of related variables Si(i=1,2,…L)。
Wherein p isiFor each sub-model, n is the number of principal elementsNumber of samples, Fα(pi,n-pi) Is a degree of freedom of piAnd n-piA is the confidence level,represents a weight of gi=ai/2biWith a degree of freedom of hi=2ai 2/biIs weighted χ2Distribution aiAnd biMean and variance of SPE statistics for each sub model are represented separately.
Further, the S5 includes the following steps:
s5.1, for a new sample data set yt∈Rm×1Firstly introducing the sampling values of s previous moments to obtain an expansion matrix ya=[yt,yt-1,…yt-s]And standardizing the same;
s5.2, according to L variable subsets S in the step S3i(i-1, 2, … L) variable partitioning method y is also dividedtDivision into y1,y2,…,yLAnd respectively calculating the monitoring index T of the model under the ith PCA modeli 2And SPEiThe specific calculation formula is as follows:
wherein P isiFor the top p in each submodeliEigenvectors, Lambda, corresponding to the largest eigenvaluesi=diag{λ1,λ2,…,λpiIs a diagonal matrix;
s5.3, obtaining L T numbers through Bayesian inference2Or fusing the SPE statistical value to obtain a probability type monitoring indexOr BICSPEAnd judging whether the current sample has a fault.
Further, the method for determining whether the current sample has a fault in step S5.3 includes the following steps:
s5.3.1 sample ytThe probability of failure is:
wherein N and F represent normal and fault, respectively,andalpha and 1-alpha, respectively, conditional probabilityAndthe calculation of (a) can be calculated in the form of an empirical distribution, specifically:
s5.3.2, fusing according to a weighting form to obtain a final probability type monitoring indexOr BICSPE:
S5.3.3 current probability type monitoring indexOr BICSPEWhen the value is more than 1-alpha, the sample y is consideredtA fault has occurred, otherwise the process is considered normal, where α is the confidence level.
Further, the step S6 includes:
s6.1, supposeIs the b sub-block, which contains a variables, then the variables areScore the ith principal elementThe contribution degree of (A) is as follows:
whereinFor loading a matrix P in the b-th blockbThe elements (A) and (B) in (B),is a diagonal matrix ΛbCorresponding elements in the Chinese character;
wherein k isbIndicating the number of the main elements reserved in the b block;
s6.3, calculating variablesAnd all its autocorrelation variables in the b-th blockAverage contribution to the b-th block:
wherein q is a variableAnd its corresponding autocorrelation variableThe number of variables in the b-th block;
s6.4, variable x existing in r blocksjThe total contribution of (a) is a weighted sum of the contributions in each block:
s6.5, likewise, the variable xjThe contribution to the SPE statistics is:
And S6.6, taking the average value of the contribution degrees of the variables in the given interval time as a diagnosis result.
Further, the key variables include: the system comprises a hearth temperature A, a boiling layer middle temperature E, a hearth temperature B, a boiling layer lower temperature A, a boiling layer upper temperature A, a boiling layer lower temperature B, a boiling layer upper temperature B, a boiling layer lower temperature C, a boiling layer upper temperature C, a boiling layer lower temperature D, a boiling layer upper temperature D, a boiling layer lower temperature E, a boiling layer upper temperature E, a flue gas outlet pressure, a boiling layer middle temperature A, a wind box pressure, a boiling layer middle temperature B, flue gas sulfur, a boiling layer middle temperature C, buried pipe circulating water flow, a boiling layer middle temperature D and a waste heat boiler convection zone inlet temperature.
(III) advantageous effects
The technical scheme of the invention has the following advantages:
(1) the method overcomes the defects that the components of the zinc concentrate are complex and changeable, the process requirements are also changed continuously, the occurrence of faults is difficult to be detected timely and accurately through a mechanism model and experience knowledge, and the fault sources cannot be found rapidly at the same time;
(2) by taking key variables reflecting abnormal furnace conditions of the roasting furnace as a reference, autocorrelation and cross-correlation variables of the roasting furnace at different sampling moments can be found out by a variable selection method based on correlation and redundancy, so that the complex time sequence correlation among process variables is better described, and the redundancy variables of the roasting process are obviously reduced;
(3) the distributed process monitoring model is established for the whole roasting process, so that local abnormal conditions can be better monitored, the accuracy and the sensitivity of the monitoring model are ensured, meanwhile, the fault variable can be quickly found by using the method of the right contribution diagram, operators can conveniently monitor the production process of the roasting furnace, the furnace condition fluctuation is accurately judged, measures are timely taken for treatment and correction, and the stability, the high efficiency and the safe smooth operation of the blast furnace are ensured, and the method is particularly important for ensuring the roasting quality.
Drawings
The features and advantages of the present invention will be more clearly understood by reference to the accompanying drawings, which are illustrative and not to be construed as limiting the invention in any way, and in which:
FIG. 1 is a flow chart of a calcination process;
FIG. 2 is a block diagram of a distributed dynamic PCA process monitoring set fault diagnosis method selected based on correlation and redundancy variables;
FIG. 3 is normal operating process data;
FIG. 4 is abnormal process data;
FIG. 5 shows T for two abnormal situations2And SPE statistic monitoring results;
FIG. 6 is a diagnostic result of an anomaly of the fluidization cooler;
fig. 7 is a diagnostic result of an abnormality of the heat recovery steam generator.
Detailed Description
The following detailed description of embodiments of the present invention is provided in connection with the accompanying drawings and examples. The following examples are intended to illustrate the invention but are not intended to limit the scope of the invention.
The embodiment of the invention relates to a method for monitoring and diagnosing abnormal furnace conditions in a roasting process of distributed dynamic PCA (principal component analysis) based on relevance and redundancy variable selection, which takes a zinc hydrometallurgy roasting process of a certain smelting plant as an example, a process flow diagram is shown in figure 1, and the roasting process comprises a drying system, a feeding system, a liquid roasting furnace, a water supply system, an air supply system, a flue gas treatment system, a roasted sand cooling system, a ball milling system and other subsystems for drying zinc concentrate. In the roasting process, zinc concentrate is uniformly mixed in a disc feeder according to the proportion, is dried in a rotary drum drying kiln and is thrown into a roasting furnace through a thrower, a large amount of air is fed into the lower part of the roasting furnace through an air blower, the air forms vertical upward airflow in a fluidized bed furnace after passing through air caps uniformly distributed in an air box, the zinc concentrate fed into the furnace is in a suspension state in a fluidized bed at the moment, is fully in contact reaction with oxygen in the air to produce zinc oxide, and most of products flow out through an overflow port and are sent to a ball milling workshop after being cooled by a fluidized cooler and a cylinder cooler; and part of products are discharged from the upper part of the furnace body along with smoke in a smoke form, the smoke after being subjected to waste heat boiler, cyclone dust collection and electric dust collection is sent to a ball milling workshop, and the smoke is sent to an acid making process. The roasting furnace body is the most important system in the whole process, various complex chemical reactions and substance conversion processes in the whole process are completed in the fluidized roasting furnace, so that the operation state of the roasting furnace has great influence on the yield and the product quality in the whole production process, but when any link in the roasting process is abnormal, the operation state of the roasting furnace is greatly influenced. All the purposes of the invention are to quickly and effectively detect the abnormal condition of the roasting furnace, accurately find the fault source and reduce the influence of the abnormal condition of the roasting process on the running state of the roasting furnace.
The implementation framework of the scheme is shown in fig. 2, and the specific implementation process is as follows:
s1, collecting all measurement variables in the history data of the zinc smelting roasting process, and selecting sample data of a normal process from the collected history data as a training data set X belonging to Rn×mN is the number of samples, and m is the number of measurement variables.
In this embodiment, all available measurement variables in the zinc smelting and roasting process are 62, sample data from 12/month 1/2019 to 2020/month 1/31 is processed, and 15000 sample data in a normal process are selected as a training set, and specific measurement variables are shown in the following table:
s2, introducing S time-delay measured values into each measured variable in the training data set X to form an augmentationMatrix Xa∈R(n-s)×(s+1)mAnd X isaIs normalized to zero mean, unit standard deviation data.
In this embodiment, each measurement variable of X in the training data set is introduced into 15 delay measurement values to form an augmented matrix Xa∈R14985×992And combining XaIs normalized to zero mean, unit standard deviation data.
S3 from X ∈ Rn×mSelecting key variable set X capable of reflecting furnace condition of roasting furnaceL∈RL×m(L.ltoreq.m) and is selected from X by a variable selection method based on correlation and redundancyaAnd finding all related variables with small redundancy of each key variable from m (s +1) variables to obtain L variable subsets.
The variable selection method based on the correlation and the redundancy considers Xa∈R(n-s)×(s+1)mThe correlation of each variable with each key variable, also taking into account the redundancy between the selected variables, comprising the steps of:
s3.1, determining a key variable set X by analyzing a roasting process technological mechanism and actual operation dataL∈RL×m(L is less than or equal to m) and candidate variable set Xa∈R(n-s)×(s+1)mAnd a set S of related variables for each key variablei(i=1,2,…L)。
S3.2 for each key variable xi∈XLCalculating the sum of X and the sum of X by using the mutual informationaEach variable x inj(j ═ 1,2, …, m (s +1)) correlation I (x)i,xj) The specific calculation formula is as follows:
wherein p (x)i,xj) Is xiAnd xjOf (2) a joint probability density function of p (x)i) And p (x)j) Are respectively xiAnd xjThe edge probability density function of (1).
S3.3 for XaEach variable x injTaking into account the sum of S andiredundancy of all variables in (A) and evaluation of redundancy, β (x)j,Si) The specific calculation formula of (2) is as follows:
s3.4, for XaEach variable x injTaken together with xiCorrelation with S andiredundancy of selected variables, calculating xjEvaluation value J (x) ofj,xi) The specific calculation formula of (2) is as follows:
s3.5, for the key variable xiFrom the resulting evaluation value set { J (x)j,xi) Finding the variable corresponding to the maximum value in (j ═ 1,2, …, m (s +1)), that is, the selected variable is: y is argmax (J (x)j,xi) Adding the variable y to SiFrom XaAnd (5) removing.
S3.6, for the selected variable SiAll the variables in (1), calculating the sum of the variables and the key variable xiThe specific calculation formula of the total evaluation value of (1) is as follows:
s3.7, calculating D through the following formula, judging whether the selection meets the termination condition, if so, continuing to execute the next step, otherwise, repeatedly executing S3.2-S3.7 until all selected variables meet the performance requirement, namely, stopping selection when the information provided by the rest variables is far lower than the redundancy caused by the selected variables, thereby obtaining the key variable xiAll related and less redundant sets of variables S ofi。
Where D is the ratio of the evaluation value of the optimal variable in the candidate variables to the evaluation value of the relevant variable set, and T is the threshold for ending the selection, where the selected threshold is 0.1 in this example.
S3.8, repeating S3.1-S3.7 for L times in total, and finding each key variable xiSet of related variables Si(i=1,2,…L)。
In this example, 22 key variables reflecting changes in the furnace conditions of the roasting furnace were selected from 62 variables in the roasting process, and specific key variables were shown in the following table and selected from X by a variable selection method based on correlation and redundancyaFinds out all related variables with small redundancy of each key variable from 992 variables in the database, and obtains 22 variable subsets Si|i=1,2,…,22}。
S4, using PCA method to perform subset S on L variablesi(i-1, 2, … L) respectively establishing fault detection models, and calculating control limits of the corresponding modelsAnd SPElim(i)。
Wherein p isiFor the number of principal elements of each submodel, n is the number of samples, Fα(pi,n-pi) Is a degree of freedom of piAnd n-piA is the confidence level,represents a weight of gi=ai/2biWith a degree of freedom of hi=2ai 2/biIs weighted χ2Distribution aiAnd biMean and variance of each submodel SPE statistic are separately represented. The PCA method is a principal component analysis method.
In this embodiment, 22 variable subsets S are processed by the PCA methodi(i 1,2, … 22) respectively establishing fault detection models, and calculating control limits of the corresponding modelsAnd SPElim(i)。
S5, in the process of on-line monitoring, introducing the on-line measured value into the corresponding delay measured value, and according to the L variable subsets S in the step S3iDividing variables (i is 1,2, … L), performing corresponding partitioning, substituting into respective fault detection models, fusing detection results of all models into a monitoring index by using Bayesian inference, and judging whether a fault occurs in a current sample; y isa=[yt,yt-1,…yt-s]
S5.1, for a new test sample data set yt∈Rm×1Firstly introducing the sampling values of s previous moments to obtain an expansion matrix ya=[yt,yt-1,…yt-s]And standardizing the same;
s5.2, according to L variable subsets S in the step S3i(i-1, 2, … L) variable partitioning method y is also dividedtDivision into y1,y2,…,yLHere, it is not necessary to repeat the step S3, and the variables are divided into L sub-blocks according to the step S3, and the monitoring indexes T under the ith PCA model are calculated respectivelyi 2And SPEiThe specific calculation formula is as follows:
wherein P isiFor the top p in each submodeliThe feature vector corresponding to the largest feature value,is a diagonal matrix.
S5.3, obtaining L T numbers through Bayesian inference2(or SPE) statistic value is fused to obtain a probability type monitoring index(or BICSPE) And judging whether the current sample has a fault, wherein the specific process comprises (by T)2For example):
s5.3.1 sample ytThe probability of failure is:
wherein N and F represent normal and fault, respectively,andalpha and 1-alpha, respectively, conditional probabilityAndthe calculation of (a) can be calculated in the form of an empirical distribution, specifically:
s5.3.2, and finally, fusing according to a weighting form to obtain a final probability type monitoring index
Similarly, taking SPE as an example, according to S5.3.1 and S5.3.2, the final probabilistic type monitoring index BIC can be obtainedSPE:
S5.3.3 current probability type monitoring indexOr BICSPEWhen the value is more than 1-alpha, the sample y is consideredtIf the fault occurs, the process is considered to be in a normal state; α is the confidence level, in this example, α is 0.99.
In the embodiment, multiple abnormal conditions occur from 1/12/2019 to 31/2020/1, and one section of data including two abnormal conditions (abnormal fluidization cooler and abnormal waste heat boiler) is selected as a test sample Y e R5800×62The specific data case is shown in FIG. 4. similarly, 15 delay measurements are introduced for each sample of the segment of data to form an amplification matrix Ya∈R5785×992And carrying out corresponding standardization processing.
Will Ya∈R5785×992Dividing into 22 subsets according to the method of step S3, substituting into 22 submodels to calculate monitoring results, and fusing the monitoring results by Bayesian inference to obtain T2And SPE statistic monitoring results are shown in fig. 5, it can be seen from the monitoring results that when a fault occurs, the abnormal condition can be rapidly monitored by using the schemeThe occurrence of the fault is frequently detected after a long period of time is needed by manual experience judgment, and corresponding processing is performed, so that great time delay is realized.
And S6, finding out the fault source by using the weighted contribution graph after the fault is detected to occur.
S6.1, supposeIs the b sub-block, which contains a variables, then the variables areScoring the ith pivotThe contribution degree of (A) is as follows:
whereinFor loading a matrix P in the b-th blockbThe elements (A) and (B) in (B),is a diagonal matrix ΛbThe corresponding elements in (1).
wherein k isbIndicating the number of pivot elements reserved in the b-th block.
S6.3, calculating variablesAnd all its autocorrelation variables in the b-th blockAverage contribution to the b-th block:
wherein q is a variableAnd its corresponding autocorrelation variableThe number of variables in the b-th block.
S6.4, due to variableNot only exists only in the b-th block, all blocks should be considered together to calculate the variable xjThe contribution degree of (c). Assume variable xjAmong the r blocks, the total contribution of the variable is a weighted sum of the contributions in each block.
S6.5, likewise, variable xjThe contribution to the SPE statistics is:
the only difference is that
And S6.6, because the contribution graph is sensitive to the samples, calculating the mean value of the contribution degree of each variable in a period of time to serve as a diagnosis result.
In this embodiment, after a fault occurs, the method of the weighted contribution graph can be used to quickly find the location of the fault source, and as shown in fig. 6 and 7, it can be seen from the graphs that after a fault occurs, the method can be used to timely and accurately find the location of the fault source.
In summary, the method for monitoring and diagnosing abnormal furnace conditions in the roasting process has the following effects:
(1) the method overcomes the defects that the components of the zinc concentrate are complex and changeable, the process requirements are also changed continuously, the occurrence of faults is difficult to be detected timely and accurately through a mechanism model and experience knowledge, and the fault sources cannot be found rapidly at the same time;
(2) by taking key variables reflecting abnormal furnace conditions of the roasting furnace as a reference, autocorrelation and cross-correlation variables of the roasting furnace at different sampling moments can be found out by a variable selection method based on correlation and redundancy, so that the complex time sequence correlation among process variables is better described, and the redundancy variables of the roasting process are obviously reduced;
(3) the distributed process monitoring model is established for the whole roasting process, so that local abnormal conditions can be better monitored, the accuracy and the sensitivity of the monitoring model are ensured, meanwhile, the fault variable can be quickly found by using the method of the weighted contribution diagram, operators can conveniently monitor the production process of the roasting furnace, the furnace condition fluctuation is accurately judged, measures are timely taken for treatment and correction, and the stability, the high efficiency and the safe smooth operation of the blast furnace are ensured, and the method is particularly important for ensuring the roasting quality. .
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations fall within the scope defined by the appended claims.
Claims (7)
1. A method for monitoring and diagnosing abnormal furnace conditions in a roasting process is characterized by comprising the following steps:
s1, collecting all measurement variables in the history data of the zinc smelting roasting process, and selecting sample data of a normal process from the collected history data as a training data set X belonging to Rn×mN is the number of samples, m is the number of measurement variables, and the measurement variables comprise data acquired by a sensor, including hearth temperature A, hearth temperature B, temperature A on a boiling layer, flue gas outlet temperature, air inlet pipe temperature, flue gas outlet pressure, air volume and boiler barrel water supply flow;
s2, introducing S delay measurement values into each measurement variable in X to form an amplification matrix Xa∈R(n-s)×(s+1)mAnd X isaEach variable in (1) is normalized to data with zero mean and unit standard deviation;
s3, analyzing the roasting process technology mechanism and the actual operation data, and determining the content of the X epsilon Rn×mSelecting key variable set X capable of reflecting furnace condition of roasting furnaceL∈RL×m(L ≦ m) and selecting method from said X by correlation and redundancy based variableaFinding all related variables with small redundancy of each key variable from m (S +1) variables to obtain L variable subsets Si(i=1,2,…L);
S4, using PCA method to perform subset S on the L variablesi(i 1,2, … L) respectively establishing fault detection models, and calculating control limits of the corresponding modelsAnd SPElim(i) The PCA method is a principal component analysis method;
s5, in the process of on-line monitoring, measuring on lineThe values are introduced into corresponding delay measurements and are according to L subsets S of variables in step S3iDividing variables (i is 1,2, … L), performing corresponding partitioning, substituting into the respective fault detection models, fusing detection results of all the fault detection models into a monitoring index by using Bayesian inference, judging whether the current sample has a fault, if so, entering step S6, otherwise, updating new test sample data, and repeating step S5;
and S6, finding the fault source by using the weighted contribution diagram.
2. The method as claimed in claim 1, wherein the correlation and redundancy-based variable selection method in step S3 includes:
s3.1, determining a key variable set X by analyzing a roasting process technological mechanism and actual operation dataL∈RL×m(L is less than or equal to m) and an amplification matrix Xa∈R(n-s)×(s+1)mAnd a subset of variables S for each key variablei(i=1,2,…L);
S3.2 for each key variable xi∈XLComputing x using mutual informationiAnd XaEach variable x injA correlation I (x +1) of (j ═ 1, 2.., m (s +/1))i,xj):
Wherein p (x)i,xj) Is xiAnd xjP (x) ofi) And p (x)j) Are respectively xiAnd xjThe edge probability density function of (a);
s3.3 for XaEach variable x injCalculating xjAnd SiRedundancy of all variables in (A), redundancy beta (x)j,Si) The calculation formula of (2):
s3.4 for XaEach variable x injSynthesis of xjAnd xiCorrelation of (2) and xjAnd SiRedundancy of selected variables, calculating xjEvaluation value of (2) J (x)j,xi) The specific calculation formula of (2) is as follows:
s3.5, for the key variable xiFrom the resulting evaluation value set { J (x)j,xi) Finding the variable corresponding to the maximum value in (j ═ 1,2, …, m (s +1)), that is, the selected variable is: y ═ arg max { J (x)j,xi) Adding the variable y to SiFrom XaRemoving;
s3.6, to variable subset SiAll the variables in (1), calculating the sum of the variables and the key variable xiThe calculation formula:
s3.7, calculating D, if D is less than or equal to T, continuing to execute the step S3.8, otherwise, returning to the step S3.2, wherein D is an augmentation matrix XaEvaluation value and variable subset S of medium-optimal variablesiT is a threshold value for the end of selection,
s3.8, returning to the step S3.1, circulating for L times in total, and finding each key variable xiIs variable subset Si(i=1,2,…L)。
3. According to claim 1The method for monitoring and diagnosing abnormal furnace conditions in the roasting process is characterized in that the method is described in step S4And SPElim(i) Is calculated by the formula
Wherein p isiFor the number of principal elements of each submodel, n is the number of samples, Fα(pi,n-pi) Is a degree of freedom of piAnd n-piA is the confidence level,represents a weight of gi=ai/2biDegree of freedom of hi=2ai 2/biIs weighted χ2Distribution aiAnd biMean and variance of SPE statistics for each sub model are represented separately.
4. The method as claimed in claim 1, wherein the step S5 includes the steps of:
s5.1, for a new test sample data set yt∈Rm×1Firstly introducing the sampling values of s previous moments to obtain an expansion matrix ya=[yt,yt-1,…yt-s]And standardizing the same;
s5.2, according to L variable subsets S in the step S3i(i-1, 2, … L) variable partitioning method y is also dividedtDivision into y1,y2,…,yLAnd respectively calculating the monitoring index T of the model under the ith PCA modeli 2And SPEiThe specific calculation formula is as follows:
wherein P isiFor the top p in each submodeliThe feature vector corresponding to the largest feature value,is a diagonal matrix;
5. The method for monitoring and diagnosing abnormal furnace conditions in the roasting process as claimed in claim 4, wherein the method for determining whether the current sample has a fault in step S5.3 comprises the following steps:
s5.3.1 sample ytThe probability of failure is:
wherein N and F represent normal and fault, respectively,andalpha and 1-alpha, respectively, conditional probabilityAndthe calculation of (a) can be calculated in the form of an empirical distribution, specifically:
s5.3.2, fusing according to a weighting form to obtain a final probability type monitoring indexOr BICSPE:
6. The method as claimed in claim 1, wherein the step S6 includes:
s6.1, supposeIs the b sub-block, which contains a variables, then the variables areScore the ith principal elementThe contribution degree of (A) is as follows:
whereinFor loading a matrix P in the b-th blockbThe elements (A) and (B) in (B),is a diagonal matrix ΛbCorresponding elements in the Chinese character;
wherein k isbIndicating the number of the main elements reserved in the b block;
s6.3, calculating variablesAnd all its autocorrelation variables in the b-th blockTo the b thBlock average contribution degree:
wherein q is a variableAnd its corresponding autocorrelation variableThe number of variables in the b-th block;
s6.4, variable x in r blocksjThe total contribution of (a) is a weighted sum of the contributions in each block:
s6.5, likewise, the variable xjThe contribution to the SPE statistics is:
And S6.6, taking the average value of the contribution degrees of the variables in the given interval time as a diagnosis result.
7. The method for monitoring and diagnosing abnormal furnace conditions in a roasting process as claimed in claim 2, wherein the key variables comprise: the system comprises a hearth temperature A, a boiling layer middle temperature E, a hearth temperature B, a boiling layer lower temperature A, a boiling layer upper temperature A, a boiling layer lower temperature B, a boiling layer upper temperature B, a boiling layer lower temperature C, a boiling layer upper temperature C, a boiling layer lower temperature D, a boiling layer upper temperature D, a boiling layer lower temperature E, a boiling layer upper temperature E, a flue gas outlet pressure, a boiling layer middle temperature A, a wind box pressure, a boiling layer middle temperature B, flue gas sulfur, a boiling layer middle temperature C, buried pipe circulating water flow, a boiling layer middle temperature D and a waste heat boiler convection zone inlet temperature.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110078674.3A CN112859815B (en) | 2021-01-21 | 2021-01-21 | Method for monitoring and diagnosing abnormal furnace conditions in roasting process |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110078674.3A CN112859815B (en) | 2021-01-21 | 2021-01-21 | Method for monitoring and diagnosing abnormal furnace conditions in roasting process |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112859815A CN112859815A (en) | 2021-05-28 |
CN112859815B true CN112859815B (en) | 2022-05-13 |
Family
ID=76008372
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110078674.3A Active CN112859815B (en) | 2021-01-21 | 2021-01-21 | Method for monitoring and diagnosing abnormal furnace conditions in roasting process |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112859815B (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104731083A (en) * | 2015-02-03 | 2015-06-24 | 浙江大学 | Industrial fault diagnosis method and application based on self-adaption feature extraction |
JP2017128805A (en) * | 2016-01-19 | 2017-07-27 | Jfeスチール株式会社 | Operation method of blast furnace |
CN110378064A (en) * | 2019-07-29 | 2019-10-25 | 安徽工业大学 | A method of the prediction dead stock column temperature of blast furnace furnace core |
CN111612181A (en) * | 2020-05-22 | 2020-09-01 | 哈尔滨锅炉厂有限责任公司 | Fault tree-based boiler abnormal working condition diagnosis and operation optimization method |
CN111898794A (en) * | 2020-06-10 | 2020-11-06 | 山东科技大学 | Abnormal monitoring method for thermal efficiency of large coal-fired boiler |
-
2021
- 2021-01-21 CN CN202110078674.3A patent/CN112859815B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104731083A (en) * | 2015-02-03 | 2015-06-24 | 浙江大学 | Industrial fault diagnosis method and application based on self-adaption feature extraction |
JP2017128805A (en) * | 2016-01-19 | 2017-07-27 | Jfeスチール株式会社 | Operation method of blast furnace |
CN110378064A (en) * | 2019-07-29 | 2019-10-25 | 安徽工业大学 | A method of the prediction dead stock column temperature of blast furnace furnace core |
CN111612181A (en) * | 2020-05-22 | 2020-09-01 | 哈尔滨锅炉厂有限责任公司 | Fault tree-based boiler abnormal working condition diagnosis and operation optimization method |
CN111898794A (en) * | 2020-06-10 | 2020-11-06 | 山东科技大学 | Abnormal monitoring method for thermal efficiency of large coal-fired boiler |
Non-Patent Citations (1)
Title |
---|
"基于主元分析与支持向量机的方法及其在密闭鼓风炉过程监控诊断中的应用";蒋少华 等;《27th Chinese Control Conference》;20080718;全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN112859815A (en) | 2021-05-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112070109B (en) | Water chestnut kiln energy consumption abnormality detection method based on improved density peak value clustering | |
CN112001527B (en) | Industrial production process target data prediction method of multi-feature fusion depth neural network | |
Jämsä-Jounela et al. | A process monitoring system based on the Kohonen self-organizing maps | |
Bencheikh et al. | New reduced kernel PCA for fault detection and diagnosis in cement rotary kiln | |
CN106649919A (en) | Method and system for predicting carbon content of fly ash in coal-fired power plant boiler | |
CN111752147B (en) | Multi-working-condition process monitoring method with continuous learning capability and improved PCA (principal component analysis) | |
CN108647481A (en) | A kind of rotary kiln burning zone temperature flexible measurement method | |
CN111352408A (en) | Multi-working-condition process industrial process fault detection method based on evidence K nearest neighbor | |
CN106200624B (en) | Based on the Industrial Boiler method for diagnosing faults for intersecting segmentation PCA | |
Liang et al. | A hybrid first principles and data-driven process monitoring method for zinc smelting roasting process | |
CN104536396A (en) | Soft measurement modeling method used in cement raw material decomposing process in decomposing furnace | |
CN106354125A (en) | Method for utilizing block PCA (Principal Component Analysis) to detect fault of chemical process | |
CN112859815B (en) | Method for monitoring and diagnosing abnormal furnace conditions in roasting process | |
US20050137995A1 (en) | Method for regulating a thermodynamic process by means of neural networks | |
CN117308275B (en) | Temperature difference-based pipeline connection abnormality detection method and system | |
Matias et al. | Fault detection and replacement of a temperature sensor in a cement rotary kiln | |
CN112541296A (en) | SO2 prediction method based on PSO-LSSVM | |
CN108759479B (en) | Sintering machine air leakage fault online diagnosis method based on blind source separation unilateral detection method | |
CN116429269A (en) | Infrared intelligent analysis system for ethylene cracking furnace tube | |
CN114722888A (en) | Early fault detection method based on Wasserstein distance | |
CN113688895A (en) | Method and system for detecting abnormal firing zone of ceramic roller kiln based on simplified KECA | |
CN112598057A (en) | Coal-fired power plant boiler fault diagnosis method based on FEKNN strategy | |
CN111222540A (en) | Abnormal judgment dioxin detection method based on unsupervised learning of clustering | |
Wang et al. | Real-time Blast Furnace Monitoring based on Temporal Sub-mode Recognition | |
CN112380779B (en) | Robust soft measurement method and system for sintering end point |
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 |