CN111813064B - Industrial process running state online evaluation method based on instant learning thought - Google Patents

Industrial process running state online evaluation method based on instant learning thought Download PDF

Info

Publication number
CN111813064B
CN111813064B CN202010635922.5A CN202010635922A CN111813064B CN 111813064 B CN111813064 B CN 111813064B CN 202010635922 A CN202010635922 A CN 202010635922A CN 111813064 B CN111813064 B CN 111813064B
Authority
CN
China
Prior art keywords
evaluation
ros
data
sample
level
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
Application number
CN202010635922.5A
Other languages
Chinese (zh)
Other versions
CN111813064A (en
Inventor
邹筱瑜
潘杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN202010635922.5A priority Critical patent/CN111813064B/en
Publication of CN111813064A publication Critical patent/CN111813064A/en
Application granted granted Critical
Publication of CN111813064B publication Critical patent/CN111813064B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G05B19/00—Programme-control systems
    • G05B19/02—Programme-control systems electric
    • G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS], computer integrated manufacturing [CIM]
    • G05B19/41885—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS], computer integrated manufacturing [CIM] characterised by modeling, simulation of the manufacturing system
    • 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
    • G05B2219/30—Nc systems
    • G05B2219/32—Operator till task planning
    • G05B2219/32339—Object oriented modeling, design, analysis, implementation, simulation language
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]

Abstract

The invention discloses an industrial process running state online evaluation method based on an instant learning thought, which aims at the problems of dynamic process characteristic changes and small modeling data volume of the industrial process running state evaluation; firstly, establishing an optimization target of a modeling data selection part according to online test data; then, establishing an optimization target of a state evaluation part; solving an objective function by using an evolutionary optimization algorithm, and simultaneously obtaining modeling data and an evaluation model; and finally, obtaining an operation state evaluation result by using the established evaluation model and an evaluation index based on the Bayesian posterior probability. The method simultaneously considers modeling data and evaluation model establishment, and simultaneously solves two parts of optimization targets so as to obtain a global optimal solution. The method improves the accuracy of the operation state evaluation by a more refined modeling means, can be finally applied to the actual industrial production field, and ensures the high-efficiency and economic operation of the industrial process with dynamic change and small modeling data volume.

Description

Industrial process running state online evaluation method based on instant learning thought
Technical Field
The invention belongs to the field of online evaluation of an industrial process running state, and particularly relates to an online evaluation method of an industrial process running state based on an instant learning thought.
Background
In order to obtain higher comprehensive economic benefit of enterprises, the evaluation of the operating state of the industrial production process is a new research subject and gradually receives attention from the academic world and the industrial world in recent years. Different from the process monitoring problem aiming at distinguishing the 'normal' or 'fault' of the process operation, the evaluation of the goodness of the process operation state refers to the further judgment of the goodness grade of the process operation state on the basis of the normal process operation, which is helpful for enterprise managers and production operators to master the operation level of the production process in real time, find out the non-excellent production operation state in time, and adjust and improve the subsequent production in time according to the analysis result of the non-excellent reasons. However, actual industrial processes are typically in dynamic variation and the amount of modeled samples is not sufficient to cover all operating conditions. If the traditional unified modeling method is adopted, the tiny characteristics can be covered, and the statistical analysis and online evaluation of the actual industrial process face challenges. Therefore, the data beneficial to the judgment of the current sample grade can be selected on line by utilizing the idea of instant learning, and an instant learning model is established to realize the evaluation of the running state. However, there is no published work to apply the instant learning idea to the evaluation of the operation state.
In order to solve the problem of evaluation of the operating state of the industrial process, Liu and the like propose an online evaluation method of the optimality of the operating state based on a Principal Component Analysis (PCA) and a multi-set Principal Component Analysis, but the method does not consider the relationship between the process variables and the evaluation indexes. The evaluation method based on index prediction avoids this problem, but requires a very large amount of data. The performance evaluation method under the probability framework can generally solve the problem of hardening component which is difficult to avoid by a multivariate statistical method. Evaluation methods based on Gaussian Mixture Model (GMM) and Bayesian theory (Bayesian theory) have been widely used in performance evaluation. The evaluation method based on probability theory needs prior knowledge to assist in determining the probability density function. Unlike the harsh requirements of the classical methods, intelligent evaluation methods, such as those based on Artificial Neural Networks (ANN), are favored by researchers because of their learning and nonlinear processing capabilities. However, such methods tend to fall into goodness and may show overfitting.
The three-phase flow process has obvious dynamic change characteristics along with the fluctuation of working conditions and the change of external conditions. Three-phase fluidization means that three phases of gas, liquid and solid exist in the device at the same time, wherein the gas and the liquid flow upwards in a cocurrent flow mode, solid particles are fluidized by the liquid to form a continuous fluidized phase, and the gas is dispersed in the fluidized phase in a bubbling ubiquitous state. The three-phase fluidized bed enhances the contact between the gas, the liquid phase and the solid particles; the bed temperature distribution can be uniform for the strong exothermic reaction. In recent years, three-phase fluidized bed reactors have been used in short-range applications such as liquid-phase catalytic synthesis of methanol, catalytic hydrogenation of petroleum, desulfurization, liquid-phase isomerization of xylene, and coal liquefaction. Due to frequent change of the set value, the three-phase flow process has obvious dynamic change characteristics, so that the evaluation of the operation state of the industrial process faces huge challenges.
And selecting data beneficial to judging the grade of the current sample on line in the instant learning, establishing an instant learning model, and realizing the evaluation of the running state. However, in the conventional instant learning method, sample selection and model establishment are performed separately, and particularly, when the sample amount is insufficient, the selected sample is not necessarily suitable for model establishment, and it is difficult to establish a high-precision evaluation model. Therefore, a more refined sample selection and model building method is needed.
Disclosure of Invention
The invention aims to provide an online evaluation method for the operation state of an industrial process based on an instant learning thought, aiming at the defects of the prior art. The method adopts an evolutionary optimization algorithm, and simultaneously completes two targets of modeling sample selection and model on-line establishment, so that the selected modeling sample has the best classification performance, and the high-efficiency and economic operation of the industrial process is ensured.
The purpose of the invention is realized by the following technical scheme: an online evaluation method for the operation state of an industrial process based on an instant learning idea comprises the following steps:
(1) acquiring J measurement variables in a dynamic industrial process, measuring each measurement variable I times, and obtaining datax jiData matrix composed of J rows and I columnsX，x jiTo representXThe element in the jth row and ith column;
(2) data matrix is processed according toXCarrying out standardization to obtain a matrix X with the mean value of 0 and the variance of 1 in each column:
Figure GDA0003027848520000021
wherein x isjiThe element representing the jth row and ith column of the matrix X,
Figure GDA0003027848520000022
is the mean of the jth variable, cjIs the standard deviation of the jth variable;
(3) the data to be evaluated collected at the time tx tCarrying out standardization processing in the step (2) to obtain data xt；
(4) Constructing an initial modeling sample set ROS0Comprises the following steps:
ROS0＝{x∈X|γ(x,xt)≥σ}
wherein X represents a sample in matrix X; the sum of the values of gamma (x,xt) Denotes x and xtThe similarity of (2); sigma is a similarity threshold;
(5) constructing a modeling sample set ROS and a grade evaluation model, comprising the following substeps:
(5.1) initial modeling sample set ROS constructed in step (4) by the following equation0Internally screening a modeling sample set ROS:
Figure GDA0003027848520000031
s.t.vi∈{0,1}
Figure GDA0003027848520000032
where w is the mapping, xiRepresents an arbitrary sample; v. ofi1 denotes selection xiAs data points in ROS, v i0 means that x is not selectediAs data points in ROS;
Figure GDA0003027848520000033
represents ROS0Inner sample point xiAfter projection with xtDivergence of (d); k is the level of the sample, there are K levels,
Figure GDA0003027848520000034
represents ROS0A set of medium-level k sample points; i is0Is a preset sample number threshold;
(5.2) establishing a grade evaluation model as follows:
Figure GDA0003027848520000035
s.t.w≠0
wherein, at a given viWithin grade divergence
Figure GDA0003027848520000036
And gradeDegree of interspersion
Figure GDA0003027848520000037
Respectively as follows:
Figure GDA0003027848520000038
Figure GDA0003027848520000039
wherein the content of the first and second substances,
Figure GDA00030278485200000310
is at a given viCenter of the k-th level, IkIs the number of samples of the k-th level,
Figure GDA00030278485200000311
is at a given viThe centers of all the levels in the case of (1) are respectively:
Figure GDA00030278485200000312
Figure GDA00030278485200000313
(6) solving the following objective function by using an evolutionary optimization algorithm to obtain v for optimizing the objective functioniAnd mapping w:
Figure GDA0003027848520000041
s.t.vi∈{0,1}
Figure GDA0003027848520000042
wherein, deltaIs a regular coefficient; for a particular viThe evolutionary optimization algorithm is equivalent to solving the maximum lagrangian multiplier λ in the following formula:
(Sw+δSd)-1Sbw＝λw
(7) for data xtAnd performing running state grade evaluation, wherein the running state grade evaluation comprises the following substeps:
(7.1) any sample X in the matrix XiScore t after w mapping obtained by step (6)iComprises the following steps:
ti＝wTxi
(7.2) sample xiIntermediate statistic s at the k-th leveli,kComprises the following steps:
Figure GDA0003027848520000043
wherein Λ is a diagonal matrix, and the elements on the diagonal are (S) in step (6)w+δSd)-1Sbw is a non-zero eigenvalue of λ w;
Figure GDA0003027848520000044
represents the center of the kth class sample in the ROS after projection, calculated as follows:
Figure GDA0003027848520000045
(7.3) obtaining an intermediate statistic s according to step (7.2)i,kObtaining the control limit G of the intermediate statistic under the k level by a kernel density estimation methodk,lim；
(7.4) calculating the test data x according to the steps (7.1) to (7.2)tScore t oftAnd the intermediate statistic st,k：
tt＝wTxt
Figure GDA0003027848520000046
Figure GDA0003027848520000047
(7.5) intermediate statistic st,kCorresponding probability density function p(s)i,k|Gk) Comprises the following steps:
Figure GDA0003027848520000048
wherein G iskRepresents the kth level;
(7.6) data xtPosterior probability Pr (G) of k-th gradek|st,k)：
Figure GDA0003027848520000051
Wherein alpha iskIs the prior probability of the rank k,
Figure GDA0003027848520000052
represents that K is from 1 to K to alphakp(st,k|Gk) Summing;
(7.7) probability Pr (G)k|st,k) The maximum level is data xtLevel oft：
Figure GDA0003027848520000053
Further, the similarity γ (x, x) in the step (4)t) Comprises the following steps:
Figure GDA0003027848520000054
wherein, | | x-xt||2Represents x and xtThe euclidean distance between, theta is a hyperparameter.
Further, the similarity threshold in the step (4) is 0 < σ < 0.5.
The invention has the beneficial effects that: the invention is the running state online evaluation of the dynamic time-varying process, so that an evaluation model can continuously adapt to new conditions, wherein the idea of instant learning is not applied in the field of the running state evaluation of the industrial process. Firstly, establishing an optimization target of a modeling data selection part according to online test data, and then establishing an optimization target of a state evaluation part; solving an objective function by using an evolutionary optimization algorithm, and simultaneously obtaining modeling data and an evaluation model; and finally, obtaining an operation state evaluation result by using the established evaluation model and an evaluation index based on the Bayesian posterior probability. The method simultaneously considers modeling data and evaluation model establishment, simultaneously solves two parts of optimization targets, obtains higher evaluation precision by a more refined modeling means, and can be finally applied to an actual industrial production field to ensure efficient and economic operation of the dynamic time-varying industrial process.
Drawings
FIG. 1 is a flow chart of a method for evaluating the condition status of a non-stationary industrial process according to the present invention;
FIG. 2 is a schematic diagram of a three-phase fluidization process in an embodiment of the present invention;
FIG. 3 is a schematic diagram showing the evaluation results of the method according to the embodiment of the present invention; wherein, (a) is a schematic diagram of a grade 1 evaluation index; (b) is a schematic diagram of a grade 2 evaluation index; (c) is a schematic diagram of a grade 3 evaluation index; (d) is a schematic diagram of the grade evaluation result;
FIG. 4 is a schematic diagram of the evaluation results of the co-integration analysis method in an embodiment of the present invention; wherein, (a) is a level 1 probability diagram; (b) is a level 2 probability diagram; (c) is a level 3 probability diagram; (d) is a schematic diagram of the grade evaluation result;
FIG. 5 is a diagram illustrating evaluation results of a co-principal component analysis method according to an embodiment of the present invention; wherein, (a) is a level 1 probability diagram; (b) is a level 2 probability diagram; (c) is a level 3 probability diagram; (d) is a schematic diagram of the grade evaluation result;
FIG. 6 is a diagram illustrating evaluation results of a neighbor search method according to an embodiment of the present invention; wherein, (a) is a level 1 probability diagram; (b) is a level 2 probability diagram; (c) is a level 3 probability diagram; (d) is a schematic diagram of the grade evaluation result; .
Detailed Description
The invention is further described with reference to the following drawings and specific embodiments.
Fig. 1 is a flowchart of the online evaluation method for the operation state of the industrial process based on the idea of learning immediately, which specifically includes the following steps:
(1) acquiring data to be analyzed: acquiring J measured variables in a dynamic industrial process, measuring each measured variable I times, and collecting data of the dynamic processx jiComposing a two-dimensional data matrixX(J×I)，x jiTo representXRow j and column i.
(2) Data matrixXThe normalization process of (1): to pairx jiThe normalization process of dividing the mean value by the standard deviation is performed to obtain a matrix X (J × I) with the mean value of 0 and the variance of 1 in each column, and the calculation formula is as follows:
Figure GDA0003027848520000061
wherein x isjiRepresenting the elements of the normalized jth row and ith column of the matrix X,
Figure GDA0003027848520000062
is the mean of the jth variable, cjIs the standard deviation of the jth variable.
(3) Collecting and processing test data: dynamic industrial process test data acquired at time tx t(JX 1) repeating the process of the step (2) to obtain the standardized test data xt。
(4) Initial modeling sample set ROS0Determination of (1): defining an initial modeling sample set ROS0Is X in combination with XtThe sample set with similarity greater than the threshold σ:
ROS0＝{x∈X|γ(x,xt)≥σ} (2)
wherein the threshold σ need not be chosen particularly large to guarantee ROS0Enough samples are available, usually 0 < σ < 0.5; gamma (x, x)t) Representing the similarity of two samples, defined as:
Figure GDA0003027848520000063
wherein, | | x-xt||2Represents x and xtThe euclidean distance between, theta is a hyperparameter.
(5) Determination of a modeled sample set ROS, comprising the sub-steps of:
(5.1) modeling the sample set ROS initially0Further screening out a data point set ROS for modeling, wherein the data points are mapped w (J multiplied by 1) and the current test data xtSufficiently similar, the screening process is expressed as the following optimization function:
Figure GDA0003027848520000071
wherein the content of the first and second substances,
Figure GDA0003027848520000072
represents ROS0The sample of (1) is selected from,
Figure GDA0003027848520000073
represents ROS0Inner sample point xiAfter projection with xtDivergence of (d); v. ofi1 denotes selection xiAs data points in ROS, v i0 means that x is not selectediAs data points in ROS; ROS0Each data point x iniAll have their own grade labels, there are a total of K grades,
Figure GDA0003027848520000074
represents ROS0A set of medium-level k sample points; i is0Is a preset sample number threshold.
(5.2) establishing a grade evaluation model: at a given viWithin grade divergence
Figure GDA0003027848520000075
And degree of inter-grade divergence
Figure GDA0003027848520000076
Respectively as follows:
Figure GDA0003027848520000077
wherein the content of the first and second substances,
Figure GDA0003027848520000078
is at a given viCenter of the k-th level, IkIs the number of samples of the k-th level,
Figure GDA0003027848520000079
is at a given viThe center of all levels in case of (1) is:
Figure GDA00030278485200000710
Figure GDA00030278485200000711
the objective function established by the finally obtained evaluation model is as follows:
Figure GDA00030278485200000712
(6) the solution of the optimization problem comprises the following substeps:
(6.1) the two-part objective function of the simultaneous formulas (4) and (8) is obtained:
Figure GDA0003027848520000081
where δ is a regular coefficient. The optimization problem comprises viTwo decision functions of (0 or 1) and w (real number) are taken, which are a mixed integer programming problem and difficult to solve; if the intelligent optimization algorithm is directly adopted for solving, great cost and time are needed, and the convergence speed is low; therefore, the optimization problem is decomposed into a 0-1 integer programming problem and a feature root decomposition problem.
(6.2) solving by using an evolutionary optimization algorithm: for a particular viThe optimization problem of equation (9) is converted into:
Figure GDA0003027848520000082
let | wT(Sw+δSd) w | | ═ 1, equation (10) further translates to:
Figure GDA0003027848520000083
the optimization problem of the formula (11) is solved through a Lagrange multiplier method, and finally the solution is solved through characteristic root decomposition, wherein a Lagrange function is as follows:
L＝wTSbw-λ(wT(Sw+δSd)w-1) (12)
wherein λ is the Lagrangian multiplier; the partial derivatives of L to w are calculated and are made to be 0:
Figure GDA0003027848520000084
then:
Sbw＝λ(Sw+δSd)w (14)
left multiplication on both sides by wTObtaining:
wTSbw＝λwT(Sw+δSd)w＝λ (15)
that is, solving the objective function maximization problem is equivalent to solving the maximum λ; according to equation (14):
(Sw+δSd)-1Sbw＝λw (16)
order Sw+δSdNot equal to 0, then the largest λ corresponds to the largest feature root in the generalized feature root decomposition problem of equation (16).
That is, for a particular viAnd combining, directly solving w through eigenvalue decomposition, and obtaining the optimal objective function value at the moment without solving w through an optimization algorithm. Therefore, instead of using w as a decision variable, only v may be usediAnd as a decision variable, the problem complexity is reduced and the solving efficiency is improved. Then, the optimization problem of equation (9) is further summarized as:
Figure GDA0003027848520000091
the evolutionary optimization algorithm is used to solve the optimization function of equation (17), such as the 0-1 encoded genetic algorithm, for each v in the solutioniThe combination of (a) and (b) all yields w that minimizes the objective function by equation (17).
(7) The operation state evaluation comprises the following substeps:
(7.1) obtaining v for optimizing the objective function by solving equation (17)i(I-1 to I) and w, and obtaining any one sample X of XiScore t after mapping by wi：
ti＝wTxi (18)
(7.2) defining sample xiIntermediate statistic s at the k-th leveli,k：
Figure GDA0003027848520000092
Wherein Λ is a diagonal matrix, and the elements on the diagonal are non-zero bits in equation (16)The characteristic value of the light-emitting diode is shown,
Figure GDA0003027848520000093
the center of the k-th rank ROS after projection is represented, and calculated as follows:
Figure GDA0003027848520000094
(7.3)si,kobtaining the control limit G of the intermediate statistic under the k level by a kernel density estimation methodk,limDefining a probability density function p(s) of statistics in the k-th leveli,k|Gk) Comprises the following steps:
Figure GDA0003027848520000095
wherein G iskIndicating the k-th level.
(7.4) obtaining test data x according to the steps (7.1) to (7.3)tScore t oftIntermediate statistic st,kAnd corresponding probability density value p(s)t,k|Gk)：
tt＝wTxt
Figure GDA0003027848520000101
Figure GDA0003027848520000102
Figure GDA0003027848520000103
Solving for x by Bayesian theorytPosterior probability Pr (G) of kth operation state gradek|st,k)：
Figure GDA0003027848520000104
Wherein alpha iskThe prior probability of the grade K is the ratio of the samples of the grade K in the modeling sample set ROS; k is the total number of grades,
Figure GDA0003027848520000105
represents that K is from 1 to K to alphakp(st,k|Gk) The summed total probabilities; x is the number oftOperating state rank of corresponding maximum Pr (G)k|st,k) Of (2), i.e. xtLevel oftComprises the following steps:
Figure GDA0003027848520000106
due to frequent fluctuations of the set point, the three-phase fluidization process is typically a dynamically operating process. Three-phase fluidization means that three phases of gas, liquid and solid exist in the device at the same time, wherein the gas and the liquid flow upwards in a cocurrent flow mode, solid particles are fluidized by the liquid to form a continuous fluidized phase, and the gas is dispersed in the fluidized phase in a bubbling ubiquitous state. The three-phase fluidized bed enhances the contact between the gas, the liquid phase and the solid particles; the bed temperature distribution can be uniform for the strong exothermic reaction. Fig. 2 is a schematic diagram of three-phase flow principle. In the step 1, the measurement variables comprise air supply pressure, riser bottom pressure, riser top pressure, top separator pressure, three-phase separator pressure, PT401 and PT408 pressure difference, VC404 pressure difference, riser top flow, top separator liquid level, top separator output flow, top riser density, top separator output density, internal water density, top riser temperature, top separator temperature, internal water temperature, VC501 valve opening degree, VC302 valve opening degree, VC101 valve opening degree and water pump current. First, the measurement data to be analyzed is standardized. According to the off-line analysis, the process operating state contains 3 levels. The state evaluation method of the invention is used for carrying out state evaluation on the three-phase flow process data, and the result is shown in figure 3. Wherein, three subgraphs (a), (b) and (c) respectively show the posterior probability that the test data belongs to 3 operation state grades, and the change from about the 45 th sample point to the operation state grade is evaluated. And the subgraph (d) shows the comparison result of the actual running state and the evaluated running state grade, the overlapped part shows that the evaluation is correct, and otherwise, the evaluation is wrong. Therefore, the accuracy of the method evaluation is more than 90%, and reliable evaluation results can be provided in the application of the related field. For comparison, the evaluation results of the co-integration analysis method, the co-integration-principal component analysis method, and the neighborhood search method are compared, as shown in fig. 4 to 6. Obviously, the evaluation accuracy of the three methods is lower than that of the method provided by the invention. The comparison results of the examples demonstrate the effectiveness and superiority of the proposed method.
Generally, the state evaluation method provided by the invention can effectively select a modeling sample, establish an evaluation model and evaluate the running state on line. The refined evaluation method is difficult to realize by the traditional dynamic process state evaluation method. The method has high precision and sensitivity, is beneficial to industrial engineers to accurately judge the process running state, and ensures the high-efficiency running of the actual production process.
The invention relates to an industrial process running state online evaluation method based on an instant learning idea, which is used for carrying out online modeling and online evaluation on a dynamically running industrial process aiming at the conditions of dynamic time variation and small modeling sample amount. Firstly, establishing an optimization target of a modeling data selection part according to online test data; then, establishing an optimization target of a state evaluation part; solving an objective function by using an evolutionary optimization algorithm, and simultaneously obtaining modeling data and an evaluation model; and finally, obtaining an operation state evaluation result by using the established evaluation model and an evaluation index based on the Bayesian posterior probability. The method simultaneously considers modeling data and evaluation model establishment, simultaneously solves two parts of optimization targets, obtains higher evaluation precision by a more refined modeling means, can be finally applied to an actual industrial production field, and ensures efficient and economic operation of the dynamic time-varying industrial process.
It should be understood that the present invention is not limited to the three-phase flow process of the above-described embodiment, and that equivalent modifications or substitutions can be made by those skilled in the art without departing from the spirit of the present invention, and the scope of the present invention is defined by the appended claims.

Claims (3)

1. An online evaluation method for the operation state of an industrial process based on an instant learning idea is characterized by comprising the following steps:
(1) acquiring J measurement variables in a dynamic industrial process, measuring each measurement variable I times, and obtaining datax jiData matrix composed of J rows and I columnsX，x jiTo representXThe element in the jth row and ith column;
(2) data matrix is processed according toXCarrying out standardization to obtain a matrix X with the mean value of 0 and the variance of 1 in each column:
Figure FDA0003059382150000011
wherein x isjiThe element representing the jth row and ith column of the matrix X,
Figure FDA0003059382150000012
is the mean of the jth variable, cjIs the standard deviation of the jth variable;
(3) the data to be evaluated collected at the time tx tCarrying out standardization processing in the step (2) to obtain data xt；
(4) Constructing an initial modeling sample set ROS0Comprises the following steps:
ROS0＝{x∈X|γ(x,xt)≥σ}
wherein X represents a sample in matrix X; gamma (x, x)t) Denotes x and xtThe similarity of (2); sigma is a similarity threshold;
(5) constructing a modeling sample set ROS and a grade evaluation model, comprising the following substeps:
(5.1) initial modeling sample set ROS constructed in step (4) by the following equation0Internally screening a modeling sample set ROS:
Figure FDA0003059382150000013
s.t.vi∈{0,1}
Figure FDA0003059382150000014
where w is the mapping, xiRepresents an arbitrary sample; v. ofi1 denotes selection xiAs data points in ROS, vi0 means that x is not selectediAs data points in ROS;
Figure FDA0003059382150000015
represents ROS0Inner sample point xiAfter projection with xtDivergence of (d); k is the level of the sample, there are K levels,
Figure FDA0003059382150000016
represents ROS0A set of medium-level k sample points; i is0Is a preset sample number threshold;
(5.2) establishing a grade evaluation model as follows:
Figure FDA0003059382150000021
s.t.w≠0
wherein, at a given viWithin grade divergence
Figure FDA0003059382150000022
And degree of inter-grade divergence
Figure FDA0003059382150000023
Respectively as follows:
Figure FDA0003059382150000024
Figure FDA0003059382150000025
wherein the content of the first and second substances,
Figure FDA0003059382150000026
is at a given viCenter of the k-th level, IkIs the number of samples of the k-th level,
Figure FDA0003059382150000027
is at a given viThe centers of all the levels in the case of (1) are respectively:
Figure FDA0003059382150000028
Figure FDA0003059382150000029
(6) solving the following objective function by using an evolutionary optimization algorithm to obtain v for optimizing the objective functioniAnd mapping w:
Figure FDA00030593821500000210
s.t.vi∈{0,1}
Figure FDA00030593821500000211
where δ is a regular coefficient; for a particular viThe evolutionary optimization algorithm is equivalent to solving the objective function maximization problemSolving for the largest lagrangian multiplier λ in:
(Sw+δSd)-1Sbw＝λw
(7) for data xtAnd performing running state grade evaluation, wherein the running state grade evaluation comprises the following substeps:
(7.1) any sample X in the matrix XiScore t after w mapping obtained by step (6)iComprises the following steps:
ti＝wTxi
(7.2) sample xiIntermediate statistic s at the k-th leveli,kComprises the following steps:
Figure FDA0003059382150000031
wherein Λ is a diagonal matrix, and the elements on the diagonal are (S) in step (6)w+δSd)-1Sbw is a non-zero eigenvalue of λ w;
Figure FDA0003059382150000032
represents the center of the kth class sample in the ROS after projection, calculated as follows:
Figure FDA0003059382150000033
(7.3) obtaining an intermediate statistic s according to step (7.2)i,kObtaining the control limit G of the intermediate statistic under the k level by a kernel density estimation methodk,lim；
(7.4) calculating the test data x according to the steps (7.1) to (7.2)tScore t oftAnd the intermediate statistic st,k：
tt＝wTxt
Figure FDA0003059382150000034
Figure FDA0003059382150000035
(7.5) intermediate statistic st,kCorresponding probability density function p(s)i,k|Gk) Comprises the following steps:
Figure FDA0003059382150000036
wherein G iskRepresents the kth level;
(7.6) data xtPosterior probability Pr (G) of k-th gradek|st,k)：
Figure FDA0003059382150000037
Wherein alpha iskIs the prior probability of the rank k,
Figure FDA0003059382150000038
denotes that K' is from 1 to K to alphakp(st,k|Gk) Summing;
(7.7) probability Pr (G)k|st,k) The maximum level is data xtLevel oft：
Figure FDA0003059382150000039
2. The on-line evaluation method for operation status of industrial process based on learning-immediately idea as claimed in claim 1, wherein the similarity γ (x, x) in the step (4)t) Comprises the following steps:
Figure FDA00030593821500000310
wherein, | | x-xt||2Represents x and xtThe euclidean distance between, theta is a hyperparameter.
3. The on-line evaluation method for the operation state of the industrial process based on the idea of learning immediately according to claim 1, wherein the similarity threshold in the step (4) is 0 < σ < 0.5.
CN202010635922.5A 2020-07-03 2020-07-03 Industrial process running state online evaluation method based on instant learning thought Active CN111813064B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010635922.5A CN111813064B (en) 2020-07-03 2020-07-03 Industrial process running state online evaluation method based on instant learning thought

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010635922.5A CN111813064B (en) 2020-07-03 2020-07-03 Industrial process running state online evaluation method based on instant learning thought

Publications (2)

Publication Number Publication Date
CN111813064A CN111813064A (en) 2020-10-23
CN111813064B true CN111813064B (en) 2021-06-25

Family

ID=72855778

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010635922.5A Active CN111813064B (en) 2020-07-03 2020-07-03 Industrial process running state online evaluation method based on instant learning thought

Country Status (1)

Country Link
CN (1) CN111813064B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105278520A (en) * 2015-06-29 2016-01-27 中国矿业大学 Complex industrial process running state evaluation method and application based on T-KPRM
CN107291654A (en) * 2016-03-31 2017-10-24 深圳光启合众科技有限公司 The intelligent decision system and method for robot
CN108732931A (en) * 2018-05-17 2018-11-02 北京化工大学 A kind of multi-modal batch process modeling method based on JIT-RVM
CN110263488A (en) * 2019-07-03 2019-09-20 昆明理工大学 A kind of industrial Mooney Viscosity of Rubber Mix flexible measurement method based on integrated instant learning

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140052425A1 (en) * 2012-08-16 2014-02-20 Sankar Selvaraj Method and apparatus for evaluating a model of an industrial plant process

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105278520A (en) * 2015-06-29 2016-01-27 中国矿业大学 Complex industrial process running state evaluation method and application based on T-KPRM
CN107291654A (en) * 2016-03-31 2017-10-24 深圳光启合众科技有限公司 The intelligent decision system and method for robot
CN108732931A (en) * 2018-05-17 2018-11-02 北京化工大学 A kind of multi-modal batch process modeling method based on JIT-RVM
CN110263488A (en) * 2019-07-03 2019-09-20 昆明理工大学 A kind of industrial Mooney Viscosity of Rubber Mix flexible measurement method based on integrated instant learning

Also Published As

Publication number Publication date
CN111813064A (en) 2020-10-23

Similar Documents

Publication Publication Date Title
Zhao et al. Soft sensor modeling of chemical process based on self-organizing recurrent interval type-2 fuzzy neural network
Chien et al. A data mining approach for analyzing semiconductor MES and FDC data to enhance overall usage effectiveness (OUE)
CN104932488A (en) Model predictive control performance evaluation and diagnosis method
CN111638707B (en) Intermittent process fault monitoring method based on SOM clustering and MPCA
CN108153267B (en) Industrial process monitoring method based on error principal component analysis model
Lin et al. A fuzzy support vector regression model for business cycle predictions
CN110046377B (en) Selective integration instant learning soft measurement modeling method based on heterogeneous similarity
CN115689353A (en) Hydropower station speed regulation system health assessment method based on GS-SVM algorithm
CN115188429A (en) Catalytic cracking unit key index modeling method integrating time sequence feature extraction
Zou et al. Deep non-crossing probabilistic wind speed forecasting with multi-scale features
CN112580693A (en) Petrochemical process fault diagnosis method based on self-help resampling neighborhood preserving embedding
Yuan et al. Spatiotemporal attention mechanism-based deep network for critical parameters prediction in chemical process
Yin et al. Dynamic real–time abnormal energy consumption detection and energy efficiency optimization analysis considering uncertainty
Lu et al. Variable-weighted FDA combined with t-SNE and multiple extreme learning machines for visual industrial process monitoring
CN111813064B (en) Industrial process running state online evaluation method based on instant learning thought
Wang et al. Data-Driven fault detection and reasoning for industrial monitoring
Sun et al. A multimode structured prediction model based on dynamic attribution graph attention network for complex industrial processes
Panjapornpon et al. Energy efficiency and savings analysis with multirate sampling for petrochemical process using convolutional neural network-based transfer learning
CN115169721B (en) Method and system for predicting single ton energy consumption in rectification process based on migration identification
CN114320266B (en) Dense oil reservoir conventional well yield prediction method based on support vector machine
CN113111588B (en) NO of gas turbine X Emission concentration prediction method and device
CN115293520A (en) Method for constructing structured multi-modal industrial process index estimation framework
CN112488411B (en) Machining stability assessment method based on approximate cycle process
Chu et al. Operating Performance Assessment of Complex Nonlinear Industrial Process Based on Kernel Locally Linear Embedding PLS
Yang et al. A new ensemble fault diagnosis method based on k-means algorithm

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