Industrial furnace working condition identification method and system fusing process variables and expert knowledge
By integrating process variables and expert knowledge, objective and subjective evidence models are constructed and conflict redistribution and fusion are carried out, the problem of low accuracy in industrial kiln conditions is solved and more accurate identification results are achieved.
Patent Information
- Application Number
- CN202410984437.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-22
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-07-22
AI Technical Summary
The existing industrial kiln working condition recognition method has the problem of low accuracy, especially since model training requires a large number of samples and is difficult to update online, and cognitive uncertainty and evidence conflicts are prone to occur when considering process variables or expert knowledge alone, resulting in identification Inaccurate.
Using the method of integrating process variables and expert knowledge, by screening key process variables, constructing objective evidence theoretical models and subjective evidence theoretical models, and through conflict redistribution and secondary fusion of evidence, local contradictions between evidence are resolved, and industrial kilns are obtained. Accurately identify the results.
It realizes accurate identification of industrial kiln working conditions, improves identification accuracy, solves the problems of evidence conflicts and difficulty in model updates, and provides a data-knowledge joint recognition method.
Smart Images

Figure CN119089376A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of industrial furnace identification, and in particular to an industrial furnace operating condition identification method and system integrating process variables and expert knowledge. Background Art
[0002] As the core equipment in industrial production, the working status of industrial furnaces is directly related to the quality, output, safety and service life of products. Industrial furnace working condition identification refers to the identification of working status by monitoring and analyzing the state parameters of the furnace, combined with professional knowledge, to assess whether there is a risk of abnormal working conditions. As one of the most important equipment in many industrial production processes, industrial furnaces have a high frequency of regulation and control and large fluctuations in raw material composition, which may lead to abnormal conditions in the furnace, posing a huge challenge to the stable operation, safety and longevity of the furnace.
[0003] The data-driven method and the expert knowledge-based method are two commonly used methods for identifying the working conditions of industrial furnaces. The former uses the input and output data generated during the operation of the furnace to perform statistical analysis and establish a data-driven model. However, the limitation of this method is that the model requires a large number of samples for training to achieve a certain prediction accuracy, but in actual situations, abnormal working condition data is often scarce, so it is difficult to train a model with high generalization performance. The expert knowledge-based method integrates professional knowledge into the model, models the principles of the smelting process, and calculates key indicators to detect the working conditions. However, the disadvantage of this method is that it requires a large number of constraints, and some ideal assumptions may not hold true in actual situations. In addition, the mechanism model usually involves complex equations, which are difficult to obtain analytical solutions, and many parameters of the model are often unable to be updated online in a timely manner, which may lead to the formation of theories that cannot effectively guide the identification of industrial furnace working conditions.
[0004] At the same time, in the process of industrial furnace working condition identification, when the information obtained and used by on-site staff is not sufficient to accurately describe objective things, cognitive uncertainty will appear. This uncertainty can be summarized as fuzziness, ambiguity and conflict of multi-source evidence. Therefore, simply considering process variables or expert knowledge and trying to directly integrate multiple contradictory evidence may lead to inaccurate working condition identification and even draw conclusions that go against intuition. Summary of the invention
[0005] The present invention provides an industrial furnace operating condition identification method and system integrating process variables and expert knowledge, so as to solve the problem of low accuracy of the industrial furnace operating condition identification method in the prior art.
[0006] In order to achieve the above object, the present invention is implemented by the following technical solutions:
[0007] In a first aspect, the present invention provides an industrial furnace operating condition identification method integrating process variables and expert knowledge, comprising:
[0008] S1: Screening key process variables of the industrial furnace process variable set, and extracting dynamic characteristics of the key process variables within a preset time window;
[0009] S2: combining the dynamic features with the NuSVC model and the probability mapping function to construct an objective evidence theory model, and obtaining an objective evidence source according to the objective evidence theory model;
[0010] S3: constructing logical rules and membership functions about key process variables based on industrial furnace expert knowledge, constructing a subjective evidence theory model based on the logical rules, membership functions and dynamic characteristics of the key process variables, and obtaining a subjective evidence source containing multiple subjective evidences according to the subjective evidence theory model;
[0011] S4: First, the multiple pieces of subjective evidence are integrated to obtain a new subjective evidence source, and then the new subjective evidence source is integrated with the objective evidence source for a second time to obtain the recognition result of the industrial furnace working condition.
[0012] In a second aspect, the present application provides an industrial furnace operating condition identification system that integrates process variables and expert knowledge, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method described in the first aspect when executing the computer program.
[0013] Beneficial effects:
[0014] The industrial furnace operating condition identification method provided by the present application, which integrates process variables and expert knowledge, extracts a set of non-redundant key process variables that contain important furnace operating condition information. In order to solve the problem of one-sidedly identifying furnace operating conditions only through data or knowledge, the present invention proposes a data-knowledge joint driven method, respectively establishing an objective evidence theoretical model based on process variables and a subjective evidence theoretical model based on expert knowledge, extracting corresponding evidence conclusions and evidence credibility, and when integrating multiple conflicting evidence sources, through the methods of conflict redistribution and secondary fusion of evidence, integrating multiple subjective evidence sources in turn, obtaining new subjective evidence, and then integrating it with objective evidence for a second time, resolving local contradictions between evidence, and obtaining accurate identification results of furnace operating conditions integrating process variables and expert knowledge.
[0015] In a further scheme, the distance from the sample to the corresponding hyperplane calculated by the NuSVC model is used as an influencing factor for the identification of furnace operating conditions and input into the furnace operating condition probability mapping function. An objective evidence theoretical model is established. In combination with the on-site investigation experience and the knowledge of industrial furnace experts, the logical rules and membership functions for the identification of furnace operating conditions are constructed. A subjective evidence theoretical model is established, and a rough identification result for judging the furnace operating conditions is obtained, namely, objective evidence and a subjective evidence source containing multiple subjective evidences. This provides a source of evidence for the subsequent fusion of subjective and objective evidence.
[0016] In a further scheme, when performing secondary fusion, the Euclidean distance, Bhattacharyya distance and cosine similarity between each evidence were fully considered. Three revised evidences were calculated based on these three calculation indicators. Then, by calculating the degree of local conflict between each revised evidence, the conflict redistribution factor was obtained. Then, the revised evidence was fused to obtain a new source of evidence, which solved the problem that multiple conflicting evidence sets are difficult to effectively fuse. This provides a theoretical basis for the fusion of subjective and objective evidence in furnace operating condition identification. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 One of the flow charts of a method for identifying industrial furnace conditions by integrating process data and expert knowledge according to a preferred embodiment of the present invention;
[0018] Figure 2 This is the second flowchart of a method for identifying industrial furnace working conditions by integrating process data and expert knowledge according to a preferred embodiment of the present invention;
[0019] Figure 3 A flowchart for screening key process variables of a preferred embodiment of the present invention;
[0020] Figure 4 It is a flow chart of the conflict redistribution and evidence secondary fusion algorithm of the preferred embodiment of the present invention;
[0021] Figure 5 (a) is a test effect diagram of the objective evidence theory model of the preferred embodiment of the present invention; 5(b) is a test result diagram of the subjective evidence theory model; 5(c) is a test result diagram of the fusion model. DETAILED DESCRIPTION
[0022] The technical solution of the present invention is described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0023] Unless otherwise defined, the technical terms or scientific terms used in the present invention shall have the usual meanings understood by persons with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, words such as "one" or "one" do not indicate quantity restrictions, but indicate the existence of at least one. Words such as "connect" or "connected" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship also changes accordingly.
[0024] See also Figure 1-Figure 2 The present application provides an industrial furnace condition identification method integrating process variables and expert knowledge, including:
[0025] S1: Screening key process variables of the industrial furnace process variable set, and extracting dynamic characteristics of the key process variables within a preset time window;
[0026] S2: combining the dynamic features with the NuSVC (Nu-Support Vector Classifier, support vector classifier with adjustable parameter Nu) model and probability mapping function to construct an objective evidence theory model, and obtaining an objective evidence source according to the objective evidence theory model;
[0027] S3: constructing logical rules and membership functions about key process variables based on industrial furnace expert knowledge, constructing a subjective evidence theory model based on the logical rules, membership functions and dynamic characteristics of the key process variables, and obtaining a subjective evidence source containing multiple subjective evidences according to the subjective evidence theory model;
[0028] In other words, the subjective evidence source before fusion has multiple evidence conclusions, but after fusion there is only one evidence conclusion.
[0029] S4: First, the multiple pieces of subjective evidence are integrated to obtain a new subjective evidence source, and then the new subjective evidence source is integrated with the objective evidence source for a second time to obtain the recognition result of the industrial furnace working condition.
[0030] The above-mentioned industrial furnace condition identification method integrating process variables and expert knowledge proposes an industrial furnace condition identification method integrating process variables and expert knowledge to address the problem that the existing industrial furnace condition identification methods do not fully combine data and knowledge. This method extracts a set of key process variables that contain important furnace condition information and are non-redundant. In response to the problem of one-sidedly identifying furnace conditions only through data or knowledge, the present invention proposes a data-knowledge joint driven method, which respectively establishes an objective evidence theoretical model based on process variables and a subjective evidence theoretical model based on expert knowledge, extracts corresponding evidence conclusions and evidence credibility, and when fusing multiple conflicting evidence sources, multiple subjective evidence sources are sequentially fused through conflict redistribution and secondary evidence fusion methods to obtain new subjective evidence, which is then secondary fused with objective evidence to resolve local contradictions between evidence, and obtain accurate furnace condition identification results integrating process variables and expert knowledge.
[0031] The steps of the above method are described in detail below with a complete embodiment.
[0032] 1. Screening of key process variables and extraction of their dynamic features
[0033] The key process variable extraction method based on prior classification information first roughly classifies the process variables through prior classification knowledge, and then performs correlation analysis between and within classes to extract key process variables. The process is as follows: Figure 3 The specific steps of this method are shown in Algorithm 1. Algorithm 1 realizes the extraction of key process variables based on prior classification information. Algorithm 2, as a sub-step, realizes the correlation analysis and variable deletion operations for a given set of process variables. The detailed analysis is as follows:
[0034] Algorithm 1:
[0035] Step 1: Divide the industrial furnace process variable set K into k categories based on the prior classification knowledge of process variables, namely:
[0036] K={C1,C2,...,C k};
[0037]
[0038]
[0039] In the formula, C j represents the set of variables of the jth type; m j is the number of variables in the j-th variable set; x j_u is the u-th variable in the j-th variable set; n is the number of samples contained in each variable;
[0040] Step 2: Linear correlation analysis involves traditional statistical methods, such as Pearson correlation coefficient and Kendall correlation coefficient, and nonlinear analysis is based on the mutual information coefficient (MIC). Through a variety of correlation analysis methods, the representatives of each category are screened out to form a representative set π, and the calculation formula is:
[0041]
[0042]
[0043]
[0044]
[0045] π={rep1,rep2,...,rep k};
[0046] Where, 1≤u≤m j ; rep j That is, select the variable corresponding to the column element and the largest column (the uth column) in the correlation coefficient matrix COR; (m j ,m j ) represents the matrix dimension; sgn is the sign function; x p and x q Both C j A column in; p,q∈[j_1,j_2,...,j_m j ]; x p and x q x p and x q An element in and x p and x q The i-th element in P(x p ,x q ) is the variable U and the variable V taking the value x at the same time p and x q The probability of P(x p ) and P(x q ) are variables U and V, respectively, taking values x p and x q The probability of
[0047] Step 3: Set the threshold σ1 based on domain expertise and expert experience, and execute Algorithm 2;
[0048] Step 4: Determine whether the elements in the representative set are deleted in this round. If so, go to Step 1 and select a new representative of the class to which the deleted representative belongs; if not, the remaining undeleted representatives form a set π' and go to Step 5;
[0049] Step 5: Set the threshold σ2 and execute Algorithm 2 for each class in turn. At this time, the set of variables remaining in each class that have not been deleted is V = {ν1, ν2, ... ν k In summary, the key variable set selected by the algorithm proposed in the present invention is KV = {π', ν1, ν2, ... ν k}.
[0050] Algorithm 2:
[0051] Given a set of process variables and a correlation threshold, the operation of performing correlation analysis and deleting variables is implemented. The specific steps are as follows:
[0052] Step 11: Calculate the correlation coefficient matrix of the set;
[0053] Step 12: Select a pair of variables (M, N) with the largest correlation coefficient and greater than the threshold σ;
[0054] Among them, σ1 and σ2 are the thresholds σ here, and they have different sizes.
[0055] Step 13: Calculate the average correlation coefficients of M, N and other variables respectively to obtain α1 and α2;
[0056] Step 14: If α1>α2, delete M, otherwise delete N;
[0057] Step 15: Repeat Step 2-Step 4 until all correlation coefficients are less than the threshold σ.
[0058] Compared with correlation analysis methods based on random forest and Lasso regression, this method first performs coarse classification of variables and coarse screening of inter-class variables based on prior knowledge, which helps to reduce the complexity of the key variable extraction process. Coarse classification and coarse screening provide a preliminary framework for further fine screening within the class, which is also conducive to in-depth analysis of the influencing factors of abnormal conditions before dealing with abnormal conditions. The advantage of this method is that it utilizes the expert knowledge of industrial furnaces and obtains the pre-classification results of process variables, making the model more interpretable and feasible.
[0059] Considering the time series characteristics of the process variables of complex multi-process industries, the dynamic characteristics of the data set within the time window are extracted. They can reflect the dynamic characteristics of the working conditions within the time window and reduce the impact of uncontrollable instantaneous disturbances. The mean, standard deviation, trend index, fluctuation coefficient, autocorrelation coefficient, kurtosis and skewness of the data within the time window are extracted. Among them, the mean can reflect the average level of the data within the window; the standard deviation, trend index and fluctuation coefficient can characterize the change trend and stability of the data and make forward-looking judgments; the autocorrelation coefficient can reflect the degree of correlation of the data within the window; the kurtosis and skewness describe the peak state and distribution position of the data respectively. These dynamic characteristics describe the changing characteristics of the process variables and provide a basis for the subsequent establishment of objective and subjective evidence theory models.
[0060] 2. Combine the NuSVC model and probability mapping function to build an objective evidence theory model based on process data
[0061] NuSVC is a classification model with wide application, diverse kernel techniques, flexible parameter adjustment, robustness to outliers and suitable for small sample data sets. It balances the number of support vectors and training error by adjusting the marginal error score nu∈(0,1]. At the same time, based on the classic structural risk minimization principle, it relies on limited training samples to explore the optimal match between the learning ability and complexity of the model to enhance the adaptability of the model. The optimization goal of the NuSVC model is to minimize the sum of squares of weights while trying to correctly classify as many as possible. The core formula and constraints are as follows:
[0062]
[0063] y i (w·x i +b)≥1-ξ i ,ξ i ≥0;
[0064] Where w is the weight; C is the regularization parameter used to balance the maximization of the interval and the minimization of the error; N is the number of samples; ξ i is a slack variable used to tolerate some training errors during training and balance the complexity and performance of the model; i is the classification result, which is processed in a one-to-many manner to take the value of 1 or -1; x i is the input vector; b is the bias term.
[0065] When NuSVC is applied to sample classification problems, it only outputs the deterministic conclusion that the sample belongs to a certain category, that is, 1 or 0, and does not support the possibility analysis of the classification results. Therefore, the present invention adopts an improved Sigmoid probability model to calculate the possibility of the operating condition, and takes the distance from each sample to the hyperplane in the trained NuSVC model as the influencing factor. The parameters U and V in the following operating condition probability mapping function P(dis,U,V) are determined by the cross entropy loss function:
[0066]
[0067] Among them, dis is the distance from the sample to the corresponding hyperplane. By inputting dis into the corresponding working condition probability mapping function, the probability of occurrence of various working conditions can be calculated, and the probability calculation result is the objective evidence credibility, which represents the credibility of the recognition result. The model obtains a rough recognition result of the furnace working condition judgment, that is, objective evidence, which provides an objective evidence source for subsequent evidence fusion.
[0068] 3. Construct a theoretical model of subjective evidence based on expert knowledge
[0069] The present invention systematizes and formalizes expert knowledge by constructing logical rules and membership functions, thereby establishing a set of scientific and reliable guiding principles. Among them, the membership function constructed using the characteristic values of process variables can effectively characterize the fuzzy state of industrial furnace process variables and their industrial process changes, and realize the uncertainty reasoning of the operating state of industrial furnaces. At the same time, the logical rules extracted from expert knowledge guide fuzzy expert reasoning, calculate the support of these rules for the operating state of industrial furnaces, and extract the credibility of subjective evidence.
[0070] In order to effectively explain and store expert knowledge, the present invention constructs a logical rule as shown in the following formula based on expert knowledge and field research experience, which uses the above-mentioned membership function calculation value to perform reasoning operations to obtain subjective evidence and its credibility.
[0071] IF A And B And(C Or D)And E Then Category(CF∈[0,1]);
[0072] In the formula, (A, B, C, D, E) are the premises of each rule (the influencing factors of the working condition); And, Or and Then are logical operators, including conjunction and disjunction relations; Category is the conclusion of the reasoning operation, including the recognition type and its credibility; CF∈[0,1] (Confidence Factor, CF, confidence coefficient) is the rule credibility of the logical rule. The model obtains a rough recognition result of the kiln working condition, that is, a subjective evidence source containing multiple subjective evidences, which provides a subjective evidence source for subsequent evidence fusion.
[0073] 4. Integrate subjective and objective sources of evidence
[0074] When there is a high degree of conflict between evidences, the traditional evidence fusion rule is difficult to describe the similarity measure between the evidences well, thus obtaining counter-intuitive results. Therefore, the present invention proposes a conflict redistribution and evidence secondary fusion algorithm, the calculation flow chart of which is as follows: Figure 4 As shown, the subjective evidence source containing multiple subjective evidences is firstly fused to obtain a new subjective evidence, and then the new subjective evidence is fused with the above objective evidence for a second time.
[0075] Firstly, three types of revised evidence sources are calculated based on Euclidean distance, Bhattacharyya distance and cosine similarity. Euclidean distance can intuitively depict the distance between two points in high-dimensional space, Bhattacharyya distance can measure the similarity of two discrete distributions, and cosine similarity analyzes the similarity between data from the angle between two vectors. It is particularly effective for similarity analysis of high-dimensional data such as complex industrial processes. Then, the conflict relationship between the three types of revised evidence sources is analyzed, and the evidence conflict redistribution factor is calculated. Then, the three types of revised evidence sources are fused to obtain new subjective evidence. Finally, the new subjective evidence and objective evidence are fused twice to resolve the local contradictions between the evidence and obtain the working condition recognition result of comprehensive data and knowledge.
[0076] The calculations for the three types of correction sources are as follows:
[0077] Corrected evidence source 1: Corrected evidence source based on Euclidean distance
[0078] Suppose the set of subjective evidence sources is
[0079] M={m i |i=1,2,...,n};
[0080] m i =(m i1 ,m i2 ,...,m iq );
[0081] And the mean of all elements in the jth (j=1,2,...,q) column (possible results) of M is satisfy and For each evidence source m i , calculate the elements m it contains ij To the column mean of the column in which it is located The Euclidean distance of:
[0082]
[0083] Therefore, the distance vector of the evidence source set M can be expressed as:
[0084] ED=[d1,d2,...,d n ];
[0085] Considering that the smaller the distance between evidences, the higher the similarity, Source of evidence i The credibility of i express Refers to the i-th evidence source m i In the above example, each element m ij To the column mean of the column The Euclidean distance of . Therefore, the modified evidence source 1 is:
[0086]
[0087] Then perform Softmax operation on M1(j) to obtain the corrected evidence source 1.
[0088] Revised source of evidence 2: Revised source of evidence based on Bhattacharyya distance
[0089] Suppose two evidence sources m in the evidence source set M i and m e The Bhattacharyya distance between (1≤i,e≤n) is:
[0090] BD(m i ,m e )=-ln(BC(m i ,m e ));
[0091]
[0092] In the formula, m ij represents the i-th evidence source m i The jth element of ej represents the e-th source of evidence m e The jth element of ;
[0093] Therefore, we can obtain a symmetric distance matrix BDM with dimension (n,n) and all diagonal elements are 0, that is, BD(m i ,m e )=BD(m e ,m i ).
[0094]
[0095] make
[0096]
[0097]
[0098] Among them, sum i represents the sum of the Bhattacharyya distances between the ith source of evidence and other sources of evidence. The larger the distance, the smaller the similarity and the lower the support. Therefore, sup i and sum i Negatively correlated and satisfying Then the revised evidence source 2 can be obtained as follows:
[0099]
[0100] Corrected evidence source 3: Corrected evidence source based on cosine similarity
[0101] Suppose two evidence sources m in the evidence source set M i and m e The cosine similarity between (1≤i,e≤n) is:
[0102]
[0103] make
[0104]
[0105] If a source of evidence is more similar to other sources of evidence, then it will get more support from other sources of evidence, and vice versa.
[0106]
[0107] And meet Therefore, the revised evidence source 3 is:
[0108]
[0109] M d =(M d (1),M d (2),...,M d (q)), d = 1, 2, 3;
[0110] Where M d is a revised evidence source with a length of q (the number of possible categories). Based on the three types of revised results of the original evidence, a conflict redistribution and evidence secondary fusion algorithm is proposed, which cleverly solves the problem of evidence conflict and realizes the fusion of conflicting evidence, namely:
[0111]
[0112] Among them, ε1(Q) is the conflict redistribution term, and the calculation formula is:
[0113]
[0114] In the formula, M1 represents the evidence distribution function of the first source of evidence, which is the credibility of the evidence source when its possible events occur. For example, a source of evidence contains three possible events, X, Y, and Z, and the credibility of these three possible events is 0.5, 0.3, and 0.2 respectively. Here, M1 refers to the revised evidence source 1; M2 is the revised evidence source 2; M3 is the revised evidence source 3; ε1 is the conflict redistribution factor when the second evidence is fused; M u is u,v,w∈[1,2,3], which is also the source of revised evidence. Here, it means replacing 1, 2, 3 with u, v, w, which have different combinations. v Refers to the revised evidence source v; M w Refers to the modified evidence source w; Q, E, and F all represent some possible events; where γ is the threshold for conflict redistribution judgment, which can be set based on prior knowledge, such as γ = 0.5. That is, when the difference in credibility between evidence is greater than the threshold γ, the conflict between evidence is assigned to the proposition with greater credibility; when the difference in credibility between evidence is less than the threshold γ, the conflict between evidence is borne by all evidence sources together.
[0115] Finally, normalize M(Q)1 to get this new subjective evidence.
[0116] Through the above steps, a new piece of subjective evidence can be obtained, and then it can be integrated with the objective evidence to obtain the final accurate recognition result. The integration rules are as follows:
[0117]
[0118] In the formula, M1 represents the evidence allocation function of the first source of evidence, which is the credibility of the source of evidence when its possible events occur. For example, a source of evidence contains three possible events X, Y, and Z, and the credibility of these three possible events is 0.5, 0.3, and 0.2 respectively. Here M1 refers to the subjective source of evidence. M2 represents the evidence allocation function of the second source of evidence, where M2 refers to the objective source of evidence. ε2 represents the conflict redistribution factor during the second evidence fusion. X, Y, Q: All are possible events. X∩Y=Q means that the intersection of event X and event Y is event Q; M(Q)2 is the result after fusion.
[0119] Among them, ε2(Q) is the conflict redistribution term, and the calculation formula is:
[0120]
[0121] Finally, the fusion evidence result can be obtained by normalizing M(Q)2. This method integrates the corrected multi-evidence source information, solves the problem of evidence conflict, realizes the redistribution of evidence conflict, and improves the availability and scientificity of the fusion model. By integrating the objective evidence source and subjective evidence source obtained in the above steps through the data-knowledge joint drive method, a more accurate industrial furnace working condition identification result can be obtained.
[0122] Next, take a domestic steel plant's 2650m 3 Taking the blast furnace as an example, the above steps are described in detail: Specifically, the 2650m 3 The data samples used for blast furnaces include normal operating conditions (7116), suspended materials (3534) and pipelines (1767). First, a key process variable extraction method based on prior classification information was used to screen out 15 key process variable sets from more than 50 industrial furnace process variables. Then, the dynamic characteristics of these 15 key process variables were extracted, the NuSVC model was trained, the parameters of the probability function were calculated, and an objective evidence theory model was established. The test results are as follows: Figure 5 Then, based on the dynamic characteristics of 15 key variables characterizing the working process of industrial furnaces obtained by the above experimental calculations, and integrating expert knowledge, five logical rules were created to judge the three types of working conditions: normal working conditions, suspended materials (abnormal working conditions) and pipelines (abnormal working conditions). Under the guidance and suggestions of on-site experts and scholars, the defined logical rules are expressed as follows:
[0123] Rule1:IF A And B And C Then Category(CF1=0.85,CF2=0.87,CF3=0.90);
[0124] Rule2:IF(D Or E)And F And(J Or K)Then Category(CF1=0.90, CF2=0.80, CF3=0.95);
[0125] Rule3:IF L And M And(N Or O)Then Category(CF1=0.90, CF2=0.93, CF3=0.88);
[0126] Rule4:IF(G Or H)And I And T Then Category(CF1=0.95, CF2=0.90, CF3=0.95);
[0127] Rule5:IF(P Or Q)And R And S Then Category(CF1=0.85, CF2=0.87, CF3=0.92);
[0128] Among them, Category∈{1,2,3} respectively represents normal working conditions, suspended materials and pipelines; CF1, CF2, CF3 respectively represent the credibility of the criterion when used to judge the three types of working conditions; the influencing factors A to T respectively refer to the CO mean, H2 mean, CO2 mean, top temperature standard deviation, top temperature fluctuation coefficient, top pressure mean, permeability index mean, permeability index standard deviation, total pressure difference mean, actual coal injection amount autocorrelation coefficient of this hour, actual coal injection amount kurtosis of this hour, oxygen enrichment flow mean, oxygen enrichment pressure mean, theoretical combustion temperature mean, theoretical combustion temperature standard deviation, cold air temperature trend index, cold air temperature fluctuation coefficient, actual wind speed mean, cold air flow mean, cold air flow resistance coefficient, a total of 20 influencing factors. Then, a subjective evidence theory model was constructed, and the test results are as follows. Figure 5 Finally, based on the proposed conflict redistribution and evidence secondary fusion algorithm, the subjective evidence source containing 5 pieces of subjective evidence is first fused to obtain a new subjective evidence source, and then the new subjective evidence source is fused with the objective evidence source for a second time to obtain the final fusion result. Figure 5 (c) shows the test results of the fusion model.
[0129] In summary, the test results of the fusion model are significantly better than those of the objective evidence theory model and the subjective evidence theory model. The objective evidence theory model is based on a large amount of data and statistical laws, and can provide objective and accurate prediction results. The subjective evidence theory model fully considers expert knowledge and can better adapt to complex actual scenarios. The fusion model can effectively integrate the two models. The experimental results show that the proposed method has achieved a high accuracy rate in the problem of working condition identification, which can meet the needs of on-site work and provide reliable decision-making support for on-site staff to monitor the operation status of the furnace in real time and adjust equipment parameters.
[0130] The embodiment of the present application also provides an industrial furnace working condition identification system integrating process variables and expert knowledge, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the above method when executing the computer program. The industrial furnace working condition identification system integrating process variables and expert knowledge can implement various embodiments of the above method and achieve the same beneficial effects, which will not be described in detail here.
[0131] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.
Claims
1. A method for identifying industrial furnace conditions by integrating process variables and expert knowledge, characterized in that: include: S1: Screening key process variables of the industrial furnace process variable set, and extracting dynamic characteristics of the key process variables within a preset time window; S2: combining the dynamic features with the NuSVC model and the probability mapping function to construct an objective evidence theory model, and obtaining an objective evidence source according to the objective evidence theory model; S3: constructing logical rules and membership functions about key process variables based on industrial furnace expert knowledge, constructing a subjective evidence theory model based on the logical rules, membership functions and dynamic characteristics of the key process variables, and obtaining a subjective evidence source containing multiple subjective evidences according to the subjective evidence theory model; S4: First, the multiple pieces of subjective evidence are integrated to obtain a new subjective evidence source, and then the new subjective evidence source is integrated with the objective evidence source for a second time to obtain the recognition result of the industrial furnace working condition.
2. The industrial furnace operating condition identification method integrating process variables and expert knowledge according to claim 1 is characterized in that: The S1 includes: S11: The key process variable extraction method based on prior classification information first roughly classifies the process variables through prior classification knowledge, and then performs correlation analysis between and within classes to extract key process variables; S12: Extract the dynamic features of the process data set within the time window to describe the changing characteristics of the process variables.
3. The industrial furnace operating condition identification method integrating process variables and expert knowledge according to claim 2 is characterized in that: The S11 includes: Step 1: Based on the prior classification knowledge of process variables, the industrial furnace process variable set K is divided into k categories, as shown in the following relationship: K={C1,C2,...,C k } (1); In the formula, C k represents the kth variable set of the industrial furnace process variable set, C j represents the set of variables of the jth type, m j is the number of variables in the j-th variable set, x j_u is the u-th variable in the j-th variable set, n is the number of samples contained in each variable, The j-th variable set C represents the industrial furnace process variable set j The mth j variables, The j-th class variable set C represents the industrial furnace process variable set j The u-th variable x in j_u The nth value of ; Step 2: Use a variety of correlation analysis methods to select the representatives of each category and form a representative set π. The calculation formula is as follows: π={rep1,rep2,...,rep k } (8); Where, 1≤u≤m j , rep j That is, select the variable corresponding to the column element and the largest column in the correlation coefficient matrix COR, (m j ,m j ) represents the matrix dimension size, sgn is the sign function, x p and x q Both C j A column in , p,q∈[j_1,j_2,...,j_m j ], x p and x q x p and x q An element in and x p and x q The i-th element in P(x p ,x q ) is the variable U and the variable V taking the value x at the same time p and x q The probability, P(x p ) and P(x q ) are variables U and V, respectively, taking values x p and x q The probability of, Pearson represents the Pearson correlation coefficient, Kendall represents the Kendall correlation coefficient, MIC represents the maximum information coefficient, Represents x p The average value of all elements in rep k Represents the variable set C from the kth class k The variables selected in represent; Step 3: Set the threshold σ1 as a parameter in the subsequent algorithm steps through domain expertise and expert experience, and perform the steps of correlation analysis and variable deletion on the target process variable set; Step 4: Determine whether the elements in the representative set are deleted in this round. If so, go to Step 1 and select a new representative of the class to which the deleted representative belongs. If not, the remaining representatives that have not been deleted in this round form a set π' and go to Step 5. Step 5: Set the threshold σ2, and perform the steps of correlation analysis and variable deletion on the target process variable set for each class in turn. At this time, the remaining undeleted variable set in each class is V = {ν1, ν2, ...ν k }, thereby combining π' and V, the key process variable set is screened out as follows: KV={π',ν1,ν2,...ν k }。 4. The industrial furnace operating condition identification method integrating process variables and expert knowledge according to claim 3 is characterized in that: The step of performing a correlation analysis on the target process variable set and deleting variables includes: Step 11: Calculate the correlation coefficient matrix of the set; Step 12: Select a pair of variables (M, N) with the largest correlation coefficient and greater than the threshold σ; Step 13: Calculate the average correlation coefficients of M, N and other variables respectively to obtain α1 and α2; Step 14: If α1>α2, delete M, otherwise delete N; Step 15: Repeat Step 12-14 until all correlation coefficients are less than the threshold σ.
5. The industrial furnace operating condition identification method integrating process variables and expert knowledge according to claim 1 is characterized in that: The dynamic characteristics include: mean, standard deviation, trend index, volatility coefficient, autocorrelation coefficient, kurtosis and skewness.
6. The industrial furnace operating condition identification method integrating process variables and expert knowledge according to claim 1 is characterized in that: The NuSVC model in S2 satisfies the following relationship: y i (w x i +b)≥1-ξ i ,x i ≥0(10); Among them, w is the weight, C is the regularization parameter, N is the number of samples, ξ i is the slack variable, y i is the classification result, x i is the input vector, b is the bias term; The probability mapping function is determined in the following way: The improved Sigmoid probability model is used to calculate the possibility of the operating condition, and the distance from each sample to the hyperplane in the trained NuSVC model is used as an influencing factor. The parameters U and V in the following operating condition probability mapping function P(dis,U,V) are determined by the cross entropy loss function: Among them, dis is the distance from the sample to the corresponding hyperplane. By inputting dis into the corresponding working condition probability mapping function, the possibility of occurrence of various working conditions is calculated. The probability calculation result is used as the objective evidence credibility, which represents the credibility of the recognition result.
7. The industrial furnace operating condition identification method integrating process variables and expert knowledge according to claim 1 is characterized in that: The S3 includes: S31: constructing a membership function through the dynamic characteristics of the process variables; S32: Construct the logic rules based on the industrial furnace expert knowledge, as shown in the following relation: IF A And B And(C Or D)And E Then Category(CF∈[0,1]) (12); In the formula, (A, B, C, D, E) are the premises of each rule, And, Or and Then are logical operators, Category is the conclusion of the reasoning operation, and CF∈[0,1] is the rule credibility of the logical rule; S33: Obtain a subjective source of evidence based on the logical rule and the membership function.
8. The industrial furnace operating condition identification method integrating process variables and expert knowledge according to claim 1 is characterized in that: The S4 includes: S41: Three types of revised evidence sources are calculated based on Euclidean distance, Bhattacharyya distance and cosine similarity respectively; S42: analyzing the conflict relationship between the three types of revised evidence sources, and calculating the evidence conflict redistribution factor according to the conflict relationship; Where, M1 represents the evidence allocation function of the first source of evidence, M2 refers to the modified evidence source 1; M2 is the modified evidence source 2; M3 is the modified evidence source 3; ε1 is the conflict redistribution factor when the second evidence is fused; M u is u,v,w∈[1,2,3], which is also the source of revised evidence. Here, it means replacing 1, 2, 3 with u, v, w, which have different combinations. v Refers to the revised evidence source v; M w Refers to the revised source of evidence w; X, Y, Z, Q, E, and F all represent events; S43: fusing the three types of revised evidence sources based on the fusion rule and the conflict redistribution factor to obtain a new subjective evidence source; S44: The new subjective evidence source and the objective evidence source are secondary integrated to obtain the identification result of the industrial furnace working condition based on comprehensive data and knowledge.
9. The industrial furnace operating condition identification method integrating process variables and expert knowledge according to claim 1 is characterized in that: The fusion rule for secondary fusion of the new subjective evidence source and the objective evidence source satisfies the following relationship: Where M1 represents the subjective source of evidence, M2 represents the objective source of evidence, ε2 represents the conflict redistribution factor during the second evidence fusion, and X, Y, and Q all represent events.
10. An industrial furnace condition identification system integrating process variables and expert knowledge, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method described in any one of claims 1 to 9 are implemented.
Citation Information
Patent Citations
Metallurgical enterprise converter gas scheduling method based on knowledge
CN106650944A
Rolling bearing fault diagnosis method based on DS evidence theory decision level fusion
CN117540335A
AGV Integrated Control Method
KR1020240141987A
Method for determining leakage level of gas pipe network based on improved evidence fusion algorithm
WO2021000061A1
Cited By
Industrial furnace working condition identification method and system
CN120524356A
Furnace body temperature control system
CN121115949A
