CN111292811A - Product prediction method and system for aromatic disproportionation production link - Google Patents

Product prediction method and system for aromatic disproportionation production link Download PDF

Info

Publication number
CN111292811A
CN111292811A CN202010073372.2A CN202010073372A CN111292811A CN 111292811 A CN111292811 A CN 111292811A CN 202010073372 A CN202010073372 A CN 202010073372A CN 111292811 A CN111292811 A CN 111292811A
Authority
CN
China
Prior art keywords
disproportionation
model
yield
sampling point
product
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.)
Granted
Application number
CN202010073372.2A
Other languages
Chinese (zh)
Other versions
CN111292811B (en
Inventor
杜文莉
杨明磊
钟伟民
钱锋
李智
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.)
East China University of Science and Technology
Original Assignee
East China University of Science and Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by East China University of Science and Technology filed Critical East China University of Science and Technology
Priority to CN202010073372.2A priority Critical patent/CN111292811B/en
Publication of CN111292811A publication Critical patent/CN111292811A/en
Application granted granted Critical
Publication of CN111292811B publication Critical patent/CN111292811B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
    • G16C20/10—Analysis or design of chemical reactions, syntheses or processes
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
    • G16C20/70—Machine learning, data mining or chemometrics
    • 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
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Chemical & Material Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Crystallography & Structural Chemistry (AREA)
  • Theoretical Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Databases & Information Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Analytical Chemistry (AREA)
  • Chemical Kinetics & Catalysis (AREA)
  • Organic Low-Molecular-Weight Compounds And Preparation Thereof (AREA)

Abstract

The invention discloses a product prediction method and a product prediction system in an aromatic disproportionation production link, and aims to realize a prediction technology of key product concentration information. The technical scheme is as follows: and selecting the operation parameters which have larger influence on the key products as model input, and taking the key performance indexes of the aromatic disproportionation process as output. Meanwhile, an initial sample is obtained by using a disproportionation mechanism model, a plurality of Kriging agent models are obtained by adopting a cross validation method, a large number of candidate sample points are randomly generated, a point with the maximum sum of cross validation variance and sparsity is found out to be used as a new sampling point, the real output of the point is calculated by using the mechanism model, and the point is added into the original sample set to retrain the agent model. Finally, a complete proxy model describing the disproportionation process is obtained, the simulation of the disproportionation production process is realized, and convenience is provided for the prediction of product yield and the optimization of operation conditions in the real-time production process.

Description

Product prediction method and system for aromatic disproportionation production link
Technical Field
The invention relates to a prediction technology for realizing the concentration information of key products in an aromatic hydrocarbon disproportionation production link, in particular to a technology for completely describing the original industrial process by a proxy model based on cross validation and sparsity, and modeling the aromatic hydrocarbon disproportionation production link through the proxy model so as to predict the concentration information of the key products in the aromatic hydrocarbon disproportionation production link.
Background
Xylene is a basic organic chemical raw material, is widely applied to industrial production, and has extremely important influence on the development of modern industry. Toluene disproportionation is an important step in the production of p-xylene, and toluene disproportionation converts toluene to benzene and xylene. The toluene disproportionation device is combined with the aromatic hydrocarbon device, so that the yield of high-quality benzene and p-xylene can be improved to the maximum extent, and byproducts such as low-quality toluene and heavy aromatic hydrocarbon are reduced to the minimum.
The simple flow chart of the toluene disproportionation process is shown in FIG. 1. The fresh feed is first mixed with the hydrogen-rich recycle gas, enters the furnace 2 after heat exchange with the effluent of the thermal reactor 1, is vaporized in the furnace 2, and when the reaction temperature is reached, the hot steam feed is sent to the reactor 1 and then flows down over the fixed bed catalyst. The product of reactor 1 is cooled by heat exchange with a mixed feed heat exchanger 3. And then sent to a knock out drum 5. Hydrogen is pumped out of the top of the knockout drum 5, enters the heat exchanger 3 through the compressor 4 and then returns to the reactor 1, and a small part of the recycle gas is used for purging, thus removing light and light accumulated in the recycle line. The liquid at the bottom of the separator 5 is sent to a stripping tower 6, the light hydrocarbons at the top of the stripping tower 6 are cooled to separate gas and liquid phase products, benzene and xylene products in reaction products, and unreacted toluene and C9Aromatic hydrocarbon is extracted from the bottom of the stripping tower 6 and then passes through a subsequent rectification device (comprising a benzene tower 7, a toluene tower 8, a xylene tower 9 and a C)9Aromatic tower 10) for pure benzene, circulating toluene and C in the aromatic tower8And (4) separating aromatic hydrocarbon and heavy aromatic hydrocarbon, and partially recycling.
In the actual production process, parameters such as temperature, pressure, feed flow, feed component concentration and the like in the disproportionation process fluctuate intermittently or periodically due to changes of operating conditions, certain parameters have great influence on the yield of the final key product, and improper operation easily causes great influence on the whole production process. However, the direct use of a mechanism model of the disproportionation process for simulation and optimization leads to problems of long calculation time, high calculation cost and the like. Therefore, a proxy model which conforms to the actual process and can accurately reflect key performance indexes such as product concentration and the like is established, and the method has important significance for guiding actual industrial production.
Disclosure of Invention
The following presents a simplified summary of one or more aspects in order to provide a basic understanding of such aspects. This summary is not an extensive overview of all contemplated aspects, and is intended to neither identify key or critical elements of all aspects nor delineate the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form as a prelude to the more detailed description that is presented later.
The invention aims to solve the problems and provides a product prediction method and a product prediction system for an aromatic disproportionation production link.
The technical scheme of the invention is as follows: the invention discloses a product prediction method in an aromatic disproportionation production link, which comprises the following steps:
step 1: receiving the operating conditions of the selected aromatic hydrocarbon disproportionation production link as input variables of the proxy model, receiving the product yield of the selected aromatic hydrocarbon disproportionation production link as output variables of the proxy model, setting the upper and lower limit ranges of the input variables, generating a plurality of initial sample points, obtaining the actual output response values of all the initial sample points by using a mechanism model, and simultaneously generating a test sample set for precision verification of the proxy model;
step 2: establishing a plurality of sub Kriging agent models according to the initial sample points and the actual output response values of the initial sample points;
and step 3: randomly generating a plurality of candidate sampling points in a sample space, respectively calculating output response values of the candidate sampling points on the sub Kriging proxy models, and obtaining cross validation variance of the output response values of each candidate sampling point on the sub Kriging proxy models;
and 4, step 4: calculating the sparseness of the candidate sampling points respectively, normalizing the sparseness of each candidate sampling point and the cross validation variance respectively and then adding the normalized sparseness and the cross validation variance to obtain the uncertainty weight of each candidate sampling point;
and 5: selecting a candidate sampling point with the largest uncertainty weight as a newly-added sampling point, calculating an actual output response value by using a mechanism model after inverse normalization, adding the newly-added sampling point into a training sample set of the proxy model to retrain the proxy model, and obtaining a final proxy model after multiple iterations;
step 6: the simulation of the aromatic hydrocarbon disproportionation production process is realized through the established proxy model, and the aromatic hydrocarbon disproportionation product yield is predicted.
According to an embodiment of the method for predicting the product of the aromatics disproportionation production process, the operating conditions of the aromatics disproportionation production process as input variables of the proxy model include: total amount of disproportionation feed, feed toluene flow, feed C9Flow, circulation hydrogen flow, make-up hydrogen flow, reaction temperature, reaction pressure, feed toluene content, feed methylethylbenzene content, and feed trimethylbenzene content; the product yield of the output variable selective disproportionation link comprises: disproportionation hydrogen yield, dry gas yield, light hydrocarbon yield, benzene yield, C8+The yield was found.
According to an embodiment of the method for predicting the product in the aromatic hydrocarbon disproportionation production link, the initial sample point in the step1 is generated by utilizing Latin hypercube sampling within the upper and lower limits of each input variable, and the sample set for testing is also generated by utilizing Latin hypercube sampling in the search space.
According to an embodiment of the method for predicting the product in the aromatics disproportionation production process, the initial sample point in step 2 is normalized before the multiple sub-Kriging agent models are established.
According to an embodiment of the method for predicting the product in the aromatics disproportionation production process, the uncertainty weight of each candidate sampling point in the step 4 is as follows:
weight＝δ2(x)/max(δ2(x) + sparsity (x)/max (sparsity (x)), where sparsity (x) is the sparsity of each candidate point, δ2(x) And (4) cross validation variance of output response values of each candidate sampling point on the plurality of sub Kriging agent models obtained in the step (3).
The invention also discloses a product prediction system for the aromatic disproportionation production link, which comprises the following steps:
the sample generation module receives the operating conditions of the selected aromatic hydrocarbon disproportionation production link as the input variable of the proxy model, receives the yield of the product of the selected aromatic hydrocarbon disproportionation production link as the output variable of the proxy model, sets the upper and lower limit ranges of the input variable, generates a plurality of initial sample points, obtains the actual output response values of all the initial sample points by using the mechanism model, and simultaneously generates a test sample set for the precision verification of the proxy model;
the sub-Kriging agent model establishing module is used for establishing a plurality of sub-Kriging agent models according to the initial sample points and the actual output response values of the initial sample points;
the cross validation variance calculation module randomly generates a plurality of candidate sampling points in a sample space, respectively calculates output response values of the candidate sampling points on the sub Kriging proxy models, and obtains cross validation variances of the output response values of each candidate sampling point on the sub Kriging proxy models;
the uncertainty weight acquisition module is used for respectively calculating the sparseness of the candidate sampling points, normalizing the sparseness of each candidate sampling point and the cross validation variance respectively and then adding the normalized sparseness and the cross validation variance to obtain the uncertainty weight of each candidate sampling point;
the agent model establishing module is used for selecting the candidate sampling point with the largest uncertainty weight as a new sampling point, calculating an actual output response value by using a mechanism model after inverse normalization, adding the new sampling point into a training sample set of the agent model to retrain the agent model, and obtaining a final agent model after multiple iterations;
and the model prediction module is used for simulating the aromatic hydrocarbon disproportionation production process through the established proxy model and predicting the aromatic hydrocarbon disproportionation product yield.
According to an embodiment of the product prediction system for an aromatics disproportionation production process of the present invention, the operating conditions of the aromatics disproportionation production process as input variables of the proxy model include: total amount of disproportionation feed, feed toluene flow, feed C9Flow, circulation hydrogen flow, make-up hydrogen flow, reaction temperature, reaction pressure, feed toluene content, feed methylethylbenzene content, and feed trimethylbenzene content; the product yield of the output variable selective disproportionation link comprises: disproportionation hydrogen yield, dry gas yield, light hydrocarbon yield, benzene yield, C8+The yield was found.
According to an embodiment of the product prediction system in the aromatic hydrocarbon disproportionation production link, the initial sample points in the sample generation module are generated by Latin hypercube sampling within the upper and lower limits of each input variable, and the sample set for testing is also generated by Latin hypercube sampling in the search space.
According to an embodiment of the product prediction system in the aromatic disproportionation production process, the initial sample points in the sub-Kriging agent model building module are normalized before the sub-Kriging agent models are built.
According to an embodiment of the product prediction system in the aromatics disproportionation production process of the present invention, the uncertainty weight of each candidate sampling point in the uncertainty weight acquisition module is:
weight＝δ2(x)/max(δ2(x) + sparsity (x)/max (sparsity (x)), where sparsity (x) is the sparsity of each candidate point, δ2(x) The cross validation variance of the output response values of each candidate sampling point on the plurality of sub Kriging agent models is obtained in the cross validation variance calculation module.
Compared with the prior art, the invention has the following beneficial effects:
1. the aromatic disproportionation model is constructed by utilizing the proxy model, the output of each corresponding key performance index can be obtained in a very short time by comparing the same input variable with the mechanism model, and the efficiency is higher.
2. Because an evolutionary algorithm is not used, only the point with the maximum uncertainty is searched from the randomly generated points and added into the training sample set, the calculation time is shortened.
3. The most complex region with the largest change is searched through cross validation errors, the region lacking sampling points is determined through sparsity, and new sampling points can be selected in a sparse region and a complex region in a self-adaptive mode through combination of the two regions, so that the precision of the proxy model is continuously improved along with the increase of the sampling points, and the final model can replace an original mechanism model and is used for real-time prediction and optimization.
Drawings
The above features and advantages of the present disclosure will be better understood upon reading the detailed description of embodiments of the disclosure in conjunction with the following drawings. In the drawings, components are not necessarily drawn to scale, and components having similar relative characteristics or features may have the same or similar reference numerals.
FIG. 1 shows a schematic flow diagram of an aromatics disproportionation production process.
FIG. 2 is a flow diagram illustrating an embodiment of a method for product prediction in an aromatics disproportionation production train in accordance with the present invention.
FIG. 3 shows a schematic diagram of an embodiment of the product prediction system of the aromatics disproportionation production train of the present invention.
Fig. 4a to 4e show schematic diagrams of the curves of the aromatics disproportionation proxy model RMSE.
Fig. 5a to 5e show schematic diagrams of the results of the yield predictions of the aromatics disproportionation proxy model.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments. It is noted that the aspects described below in connection with the figures and the specific embodiments are only exemplary and should not be construed as imposing any limitation on the scope of the present invention.
The principle of the invention is as follows: a proxy model is established through a mechanism model of an aromatic hydrocarbon disproportionation link, and self-adaptive sampling is carried out by utilizing cross validation errors and sparsity, so that the accuracy of the established proxy model is continuously improved, and the method is finally used for real-time prediction of product yield.
The technological process of aromatics disproportionation is shown in figure 1, and the raw materials of toluene and C9The aromatic hydrocarbon is firstly mixed with the circulating hydrogen, then exchanges heat with the material from the reactor 1, is preheated to the temperature required by the reaction by the raw material heating furnace 2, passes through the disproportionation reactor 1 from top to bottom, and contacts with the catalyst to generate the disproportionation reaction. The reaction product leaves the reactor 1 and exchanges heat with the raw material through a heat exchanger 3. Then the condensed and cooled gas enters a separator 5 for gas-liquid separation, and then passes through a stripping tower 6, a benzene tower 7, a toluene tower 8, a xylene tower 9 and a toluene tower C9The aromatics column 10 separates the products.
FIG. 2 shows a flow diagram of an embodiment of a product prediction method of the aromatics disproportionation production stage of the present invention. Referring to fig. 2, the following is a detailed description of the implementation steps of the present embodiment.
Step 1: receiving the operating conditions of the selected aromatic hydrocarbon disproportionation production link as the input variable of the proxy model, receiving the yield of the product of the selected aromatic hydrocarbon disproportionation production link as the output variable of the proxy model, setting the upper and lower limit ranges of the input variable, generating a plurality of (for example, 20) initial sample points by utilizing Latin hypercube sampling to form an initial sample set, and obtaining the actual output response values (mainly comprising the yield of disproportionation hydrogen, the yield of dry gas, the yield of light hydrocarbon, the yield of benzene and C) of all the initial sample points by utilizing a Hysys mechanism model of aromatic hydrocarbon disproportionation8+Yield) while randomly generating a test sample set for precision verification of the proxy model.
In this step, the operating conditions with large influence in the production process of disproportionation of aromatic hydrocarbon are generally selected as input variables of the proxy model, including: total amount of disproportionation feed, feed toluene flow, feed C9Flow, circulating hydrogen flow, make-up hydrogen flow, reaction temperature, reaction pressure, feed toluene content, feed methylethylbenzene content, and feed trimethylbenzene content. Selecting the product yield of the aromatic disproportionation production link as the output variable of the proxy model, wherein the product yield mainly comprises the yield of disproportionation hydrogen, the yield of dry gas, the yield of light hydrocarbon, the yield of benzene and C8+The yield was found.
At one isIn the example, the total amount of disproportionation feed, the flow rate of feed toluene and the feed C in the production link of disproportionation of aromatic hydrocarbon are selected910 operation conditions including flow, circulation hydrogen flow, make-up hydrogen flow, reaction temperature, reaction pressure, feed toluene content, feed methylethylbenzene content and feed trimethylbenzene content are used as input variables, and the upper and lower limits of the 10 input variables are set
Figure BDA0002377831720000061
Obtaining 20 initial sample points using latin hypercube sampling, X ═ X1,x2,...,xn,]TWherein
Figure BDA0002377831720000071
n represents the number of sample points, d represents the dimension of the variable, where n is 20 and d is 10. And obtaining the actual output response value of the product yield of the 20 samples by utilizing a Hysys mechanism model. These 20 samples constitute the initial sample set. In the same way, 100 more test sample sets were obtained for testing the accuracy of the final proxy model.
In the process of obtaining the initial response values of the 20 initial samples by using the Hysys mechanism model, the 20 initial samples are firstly normalized one by one to eliminate the influence of the sample dimension on the calculation:
Figure BDA0002377831720000072
(1) in the formula (I), the compound is shown in the specification,
Figure BDA0002377831720000073
represents the value after the k-dimensional normalization of the ith variable,
Figure BDA0002377831720000074
a maximum value of the k-th dimension is represented,
Figure BDA0002377831720000075
represents the minimum value of the k-th dimension,
Figure BDA0002377831720000076
expressing the kth dimension of the ith variable, and obtaining a real output y by using a Hysys mechanism model and an initial sample1,y2,...,yn]This results in an initial training sample set.
Step 2: obtaining a plurality of sub-Kriging surrogate models (the sub-Kriging surrogate models comprise sub-Kriging models) according to the initial sample points and the actual output response values of the product yield of the initial sample points.
Continuing the example above, 20 initial samples were randomly divided into 10 pieces, [ Date1,Date2,...,DateN]，N＝10。
And then establishing a Kriging agent model (Kriging is an interpolation technology based on statistics). The formula is as follows:
Figure BDA0002377831720000077
Figure BDA0002377831720000078
is the predicted response value of an input variable x, where x is a vector of dimension d, where u is defined as a constant, and z (x) is defined as a random process whose random behavior is as follows:
Figure BDA0002377831720000079
z (x) is desirably 0 and the variance is σ2Covariance of σ2R(θ,xi,xj) Wherein R (theta, x)i,xj) Representing two sample points xiAnd xjA correlation function between; theta ═ theta1,θ2,......θdIs to determine R (theta, x)i,xj) A set of parameters of the gradient. The present embodiment uses a gaussian correlation function, which is defined as follows:
R(θk,xi,xj)＝exp(-θk|xi-xj|2) (4)
for a given value of thetaPredicted value at x
Figure BDA0002377831720000081
Calculated according to the following formula:
Figure BDA0002377831720000082
where R is a n × 1 column vector representing the correlation matrix between the prediction point x and the sample point, with the i-th element being R (x, x)i) Wherein i ═ 1,2, 3.., n; r is a correlation matrix of n × n order symmetry with the (i, j) th element being R (x)i,xj) Wherein i, j ═ 1,2, 3. y is a column vector of n x 1, the i-th element of which is a point xiIs the objective function value y (x)i) (ii) a 1 is a unit column vector of n × 1.
In general, the hyper-parameter θ can be obtained by a maximum likelihood method, and the solving process is as follows:
Figure BDA0002377831720000083
the u and sigma can be obtained by taking the derivative of the above formula and making the derivative zero2Maximum likelihood estimation of
Figure BDA0002377831720000084
Wherein the content of the first and second substances,
Figure BDA0002377831720000085
and
Figure BDA0002377831720000086
all depend on the unknown parameter theta, and after theta is determined, the Kriging agent model is built.
Establishing 10 sub Kriging agent models, and selecting Date each time of construction i9 of the three sub Kriging agent models are used as training samples, and one sample set is abandoned each time, and finally 10 sub Kriging agent models F are obtainedk(x) K 1, 2.., 10, for subsequent calculation of cross-validation variance.
And step 3: randomly generating a plurality of candidate sampling points in a sample space, respectively calculating output response values of the plurality of candidate sampling points on the plurality of sub Kriging proxy models in the step 2, and obtaining the cross validation variance of the output response values of each candidate sampling point on the plurality of sub Kriging proxy models. Cross validation variance δ2(x) The formula is as follows:
Figure BDA0002377831720000087
where n represents the number of sample points and x represents the average number of samples.
Continuing with the example above, 2000 candidate sample points X are randomly generated in sample space using Latin hypercube samplingcandidateRespectively calculating the response values F of the 1000 sample points at the samplek(x) The above result, the variance δ of the response values of 10 sub Kriging agent models is calculated2(x)。
And 4, step 4: and respectively calculating the sparsity of the candidate sampling points, respectively normalizing the sparsity and the cross validation variance, and then adding the normalized sparsity and the cross validation variance to obtain the uncertainty weight of each candidate sampling point.
Continuing with the above example, the sparsity of each candidate sample point is calculated, the sparsity being defined as follows:
existing sampling point X ═ X1,x2,...,xN]TThe search upper and lower limits are UP ═ UP1,up2,...,upDAnd DOWN ═ DOWN1,down2,...,downD}. At any point x in the search spacenewSparsity of (d) is defined as follows:
step1 calculates new sampling point xnewWith existing sampling point X ═ X1,x2,...,xN]TThe Euclidean distance of (1) and sorting to obtain diatanancesort。
Step2 X2＝[x1,x2,...,xN,xnew]T
Step3 for j is 1: D (for represents loop processing, hereinafter, the processing content of the loop body)
To pair
Figure BDA0002377831720000091
Are sequenced from small to large to obtain
Figure BDA0002377831720000092
That is, the value of the ith dimension is sorted from small to large to find xnew,iAt position pos. Similarly sorting the distances dittance to obtain dittancesort。
if pos＝1
The lower limit of the ith dimension sparsity is the lower limit of a sampling space
Figure BDA0002377831720000093
The upper limit is the ratio x in the ith dimensionnew,iIn large spots, distance xnewThe value of the ith dimension of the nearest point,
Figure BDA0002377831720000094
wherein
Figure BDA0002377831720000095
else if pos＝N+1
The upper limit of the i-th dimension sparsity is the upper limit of the sampling space
Figure BDA0002377831720000096
The lower limit is the ratio x in the ith dimensionnew,iAmong small dots, distance xnewThe value of the ith dimension of the nearest point,
Figure BDA0002377831720000097
wherein
Figure BDA0002377831720000098
else
The upper limit of the i-th dimension sparsity is the ratio x in the i-th dimensionnew,iIn large spots, distance xnewIth dimension value of nearest point
Figure BDA0002377831720000099
Wherein
Figure BDA00023778317200000910
The lower limit is a sampling space lower limit is a ratio xnew,iAmong small dots, distance xnewThe value of the ith dimension of the nearest point,
Figure BDA00023778317200000911
wherein
Figure BDA00023778317200000912
end (end of cycle)
Step 4 finally obtains sparsity:
Figure BDA0002377831720000101
and then calculating the uncertainty weight of each candidate sampling point:
weight(x)＝δ2(x)/max(δ2(x))+sparsity(x)/max(sparsity(x)) (10)
and 5: and selecting the candidate sampling point with the maximum uncertainty weight as a newly-added sampling point to be added into the training sample set, calculating the actual output response value of the product yield by using the mechanism model, and retraining the proxy model, so that iteration is continuously carried out until the upper limit of the evaluation times of the mechanism model is reached, and the final Kriging proxy model is obtained.
Continuing with the above example, the candidate sample point with the greatest uncertainty weight is selected as the new sample point
Figure BDA0002377831720000102
Obtaining an actual value after inverse normalization:
Figure BDA0002377831720000103
and obtaining an actual output response value by using a Hysys mechanism model, adding the actual output response value into a training sample set, and retraining the Kriging agent model.
And repeating the process of adding the new sample points until the evaluation times of the mechanism model reach the upper limit, and finally obtaining the proxy model of the disproportionation process.
In this example, the initial sample set is selected as 20 sample points, and the total evaluation times of the mechanism model is 200, that is, the number of newly added sample points is 180. Repeating the above process of constructing the proxy model for the five yields, respectively, and calculating the Root Mean Square Error (RMSE) of the obtained proxy model each time:
Figure BDA0002377831720000104
in formula (14), yiAnd
Figure BDA0002377831720000105
the real output response value and the proxy model response value at the ith test point are respectively, and M is the number of the test points. The change curves of the proxy model RMSE after each iteration are shown in fig. 4a to 4e, and it can be found that the accuracy of the whole model is greatly improved along with the increase of the number of sampling points. The finally obtained fitting results are shown in fig. 5a to 5e, and it can be found that the agent model can predict 100 test sample points accurately, and can be used for real-time prediction of the subsequent disproportionation process.
Step 6: the simulation of the aromatic hydrocarbon disproportionation production process is realized through the established proxy model, and the aromatic hydrocarbon disproportionation product yield is predicted.
FIG. 3 illustrates the principles of an embodiment of the product prediction system of the aromatics disproportionation production train of the present invention. Referring to fig. 3, the system of the present embodiment includes: the system comprises a sample generation module, a sub Kriging agent model establishing module, a cross validation variance calculating module, an uncertainty weight obtaining module, an agent model establishing module and a model predicting module.
The sample generation module receives the operating conditions of the selected aromatic hydrocarbon disproportionation production link as the input variable of the proxy model, receives the yield of the product of the selected aromatic hydrocarbon disproportionation production link as the output variable of the proxy model, sets the upper and lower limit ranges of the input variable, generates a plurality of initial sample points, obtains the actual output response values of the product yields of all the initial sample points by using the Hysys mechanism model, and simultaneously generates a test sample set for the precision verification of the proxy model.
The operating conditions of the aromatic disproportionation production link as input variables of the proxy model comprise: total amount of disproportionation feed, feed toluene flow, feed C9Flow, circulation hydrogen flow, make-up hydrogen flow, reaction temperature, reaction pressure, feed toluene content, feed methylethylbenzene content, and feed trimethylbenzene content; the product yield included: disproportionation hydrogen yield, dry gas yield, light hydrocarbon yield, benzene yield, C8+The yield was found.
The initial sample points in the sample generation module are generated by utilizing Latin hypercube sampling in the upper and lower limit ranges of each input variable, and the sample set for testing is also generated by utilizing Latin hypercube sampling in the search space.
And the sub Kriging agent model establishing module is used for firstly carrying out normalization operation according to the initial sample point and the actual output response value of the initial sample point and then establishing a plurality of sub Kriging agent models.
And the cross validation variance calculation module randomly generates a plurality of candidate sampling points in a sample space, respectively calculates output response values of the candidate sampling points on different sub Kriging proxy models, and obtains the cross validation variance of the output response values of each candidate sampling point on the sub Kriging proxy models.
And the uncertainty weight acquisition module is used for respectively calculating the sparseness of the candidate sampling points, normalizing the sparseness of each candidate sampling point and the cross validation variance respectively and then adding the normalized sparseness and the cross validation variance to obtain the uncertainty weight of each candidate sampling point.
The uncertainty weight of each candidate sampling point in the uncertainty weight acquisition module is as follows: weight δ2(x)/max(δ2(x) + sparsity (x)/max (sparsity (x)), where sparsity (x) is the sparsity of each candidate point, δ2(x) Is the intersection of the output response values of each candidate sampling point on the multiple sub Kriging agent models obtained in the cross validation variance calculation moduleThe variance is verified.
And the proxy model establishing module is used for selecting the candidate sampling point with the maximum uncertainty weight as a new sampling point, calculating the actual output response value of the product yield by using the mechanism model after inverse normalization, adding the new sampling point into the training sample set of the proxy model to retrain the proxy model, and obtaining the final proxy model after multiple iterations.
And the model prediction module is used for simulating the aromatic hydrocarbon disproportionation production process through the established proxy model and predicting the aromatic hydrocarbon disproportionation product yield.
The previous description of the disclosure is provided to enable any person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other variations without departing from the spirit or scope of the disclosure. Thus, the disclosure is not intended to be limited to the examples and designs described herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (10)

1. A product prediction method for an aromatic disproportionation production link is characterized by comprising the following steps:
step 1: receiving the operating conditions of the selected aromatic hydrocarbon disproportionation production link as input variables of the proxy model, receiving the product yield of the selected aromatic hydrocarbon disproportionation production link as output variables of the proxy model, setting the upper and lower limit ranges of the input variables, generating a plurality of initial sample points, obtaining the actual output response values of all the initial sample points by using a mechanism model, and simultaneously generating a test sample set for precision verification of the proxy model;
step 2: establishing a plurality of sub Kriging agent models according to the initial sample points and the actual output response values of the initial sample points;
and step 3: randomly generating a plurality of candidate sampling points in a sample space, respectively calculating output response values of the candidate sampling points on the sub Kriging proxy models, and obtaining cross validation variance of the output response values of each candidate sampling point on the sub Kriging proxy models;
and 4, step 4: calculating the sparseness of the candidate sampling points respectively, normalizing the sparseness of each candidate sampling point and the cross validation variance respectively and then adding the normalized sparseness and the cross validation variance to obtain the uncertainty weight of each candidate sampling point;
and 5: selecting a candidate sampling point with the largest uncertainty weight as a newly-added sampling point, calculating an actual output response value by using a mechanism model after inverse normalization, adding the newly-added sampling point into a training sample set of the proxy model to retrain the proxy model, and obtaining a final proxy model after multiple iterations;
step 6: the simulation of the aromatic hydrocarbon disproportionation production process is realized through the established proxy model, and the aromatic hydrocarbon disproportionation product yield is predicted.
2. The method of predicting the product of an aromatics disproportionation production process as claimed in claim 1, wherein the operating conditions of the aromatics disproportionation production process as input variables of the proxy model include: total amount of disproportionation feed, feed toluene flow, feed C9Flow, circulation hydrogen flow, make-up hydrogen flow, reaction temperature, reaction pressure, feed toluene content, feed methylethylbenzene content, and feed trimethylbenzene content; the product yield of the output variable selective disproportionation link comprises: disproportionation hydrogen yield, dry gas yield, light hydrocarbon yield, benzene yield, C8+The yield was found.
3. The method for predicting the product in the disproportionation production of aromatic hydrocarbons according to claim 1, wherein the initial sample points in step1 are generated by latin hypercube sampling within the range of the upper and lower limits of each input variable, and the sample set for testing is also generated by latin hypercube sampling in the search space.
4. The method as claimed in claim 1, wherein the initial sample points in step 2 are normalized before the sub-Kriging agent models are built.
5. The method for predicting the product of the disproportionation production of aromatic hydrocarbons as claimed in claim 1, wherein the uncertainty weight of each candidate sampling point in step 4 is:
weight＝δ2(x)/max(δ2(x) + sparsity (x)/max (sparsity (x)), where sparsity (x) is the sparsity of each candidate point, δ2(x) And (4) cross validation variance of output response values of each candidate sampling point on the plurality of sub Kriging agent models obtained in the step (3).
6. A product prediction system for an aromatic disproportionation production process, comprising:
the sample generation module receives the operating conditions of the selected aromatic hydrocarbon disproportionation production link as the input variable of the proxy model, receives the yield of the product of the selected aromatic hydrocarbon disproportionation production link as the output variable of the proxy model, sets the upper and lower limit ranges of the input variable, generates a plurality of initial sample points, obtains the actual output response values of all the initial sample points by using the mechanism model, and simultaneously generates a test sample set for the precision verification of the proxy model;
the sub-Kriging agent model establishing module is used for establishing a plurality of sub-Kriging agent models according to the initial sample points and the actual output response values of the initial sample points;
the cross validation variance calculation module randomly generates a plurality of candidate sampling points in a sample space, respectively calculates output response values of the candidate sampling points on the sub Kriging proxy models, and obtains cross validation variances of the output response values of each candidate sampling point on the sub Kriging proxy models;
the uncertainty weight acquisition module is used for respectively calculating the sparseness of the candidate sampling points, normalizing the sparseness of each candidate sampling point and the cross validation variance respectively and then adding the normalized sparseness and the cross validation variance to obtain the uncertainty weight of each candidate sampling point;
the agent model establishing module is used for selecting the candidate sampling point with the largest uncertainty weight as a new sampling point, calculating an actual output response value by using a mechanism model after inverse normalization, adding the new sampling point into a training sample set of the agent model to retrain the agent model, and obtaining a final agent model after multiple iterations;
and the model prediction module is used for simulating the aromatic hydrocarbon disproportionation production process through the established proxy model and predicting the aromatic hydrocarbon disproportionation product yield.
7. The system of claim 6, wherein the operating conditions of the aromatics disproportionation production unit as input variables to the proxy model comprise: total amount of disproportionation feed, feed toluene flow, feed C9Flow, circulation hydrogen flow, make-up hydrogen flow, reaction temperature, reaction pressure, feed toluene content, feed methylethylbenzene content, and feed trimethylbenzene content; the product yield of the output variable selective disproportionation link comprises: disproportionation hydrogen yield, dry gas yield, light hydrocarbon yield, benzene yield, C8+The yield was found.
8. The aromatics disproportionation production process product prediction system of claim 6 wherein the initial sample points in the sample generation module are generated by Latin hypercube sampling within the upper and lower limits of each input variable, and the sample set for testing is also generated by Latin hypercube sampling in the search space.
9. The system of claim 6, wherein the initial sample points in the sub-Kriging agent model building module are normalized before the sub-Kriging agent models are built.
10. The system for predicting the disproportionation production of aromatics as claimed in claim 6, wherein the uncertainty weight of each candidate sampling point in the uncertainty weight obtaining module is:
weight＝δ2(x)/max(δ2(x))+sparsity(x)/max(sparsity(x) Where sparse (x) is the sparsity of each candidate point, δ2(x) The cross validation variance of the output response values of each candidate sampling point on the plurality of sub Kriging agent models is obtained in the cross validation variance calculation module.
CN202010073372.2A 2020-01-22 2020-01-22 Product prediction method and system for aromatic disproportionation production link Active CN111292811B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010073372.2A CN111292811B (en) 2020-01-22 2020-01-22 Product prediction method and system for aromatic disproportionation production link

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010073372.2A CN111292811B (en) 2020-01-22 2020-01-22 Product prediction method and system for aromatic disproportionation production link

Publications (2)

Publication Number Publication Date
CN111292811A true CN111292811A (en) 2020-06-16
CN111292811B CN111292811B (en) 2023-04-14

Family

ID=71030657

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010073372.2A Active CN111292811B (en) 2020-01-22 2020-01-22 Product prediction method and system for aromatic disproportionation production link

Country Status (1)

Country Link
CN (1) CN111292811B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1233619A (en) * 1993-06-07 1999-11-03 三井化学株式会社 Transition metallic compound, olefins polymerization using catalyst contg. same and products thereof
CN107844659A (en) * 2017-11-16 2018-03-27 华东理工大学 The agent model modeling method of coal water slurry gasification process
WO2019187372A1 (en) * 2018-03-30 2019-10-03 Ｎｅｃソリューションイノベータ株式会社 Prediction system, model generation system, method, and program

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1233619A (en) * 1993-06-07 1999-11-03 三井化学株式会社 Transition metallic compound, olefins polymerization using catalyst contg. same and products thereof
CN107844659A (en) * 2017-11-16 2018-03-27 华东理工大学 The agent model modeling method of coal water slurry gasification process
WO2019187372A1 (en) * 2018-03-30 2019-10-03 Ｎｅｃソリューションイノベータ株式会社 Prediction system, model generation system, method, and program

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
谢雨珩等: "基于自适应采样算法的芳烃异构化代理模型", 《化工学报》 *

Also Published As

Publication number Publication date
CN111292811B (en) 2023-04-14

Similar Documents

Publication Publication Date Title
Zerveas et al. A transformer-based framework for multivariate time series representation learning
Wang et al. A two‐layer ensemble learning framework for data‐driven soft sensor of the diesel attributes in an industrial hydrocracking process
CN106980648B (en) Personalized recommendation method based on probability matrix decomposition and combined with similarity
CN112949823A (en) Industrial process performance diagnosis method based on one-dimensional multi-scale depth convolution neural network
CN104657744A (en) Multi-classifier training method and classifying method based on non-deterministic active learning
Zheng et al. Prompt vision transformer for domain generalization
CN105740984A (en) Product concept performance evaluation method based on performance prediction
Carballido et al. CGD-GA: A graph-based genetic algorithm for sensor network design
CN111312338A (en) Product prediction method and system for aromatic hydrocarbon isomerization production link
CN110600085A (en) Organic matter physicochemical property prediction method based on Tree-LSTM
Badrinarayanan et al. Symmetry-invariant optimization in deep networks
CN114239400A (en) Multi-working-condition process self-adaptive soft measurement modeling method based on local double-weighted probability hidden variable regression model
Yong et al. Fault diagnosis based on fuzzy support vector machine with parameter tuning and feature selection
Shimin et al. Efficient relation-aware scoring function search for knowledge graph embedding
CN111292811B (en) Product prediction method and system for aromatic disproportionation production link
CN103605493A (en) Parallel sorting learning method and system based on graphics processing unit
Zhu et al. LMMO: a large margin approach for refining regulatory motifs
CN111950652A (en) Semi-supervised learning data classification algorithm based on similarity
Wang et al. A new input variable selection method for soft sensor based on stacked auto-encoders
CN112786120B (en) Method for synthesizing chemical material with assistance of neural network
Kou et al. Attention-guided Contrastive Hashing for Long-tailed Image Retrieval.
CN115762654A (en) Smiles-based chemical reaction yield prediction method
Si et al. A data-driven fault detection framework using mahalanobis distance based dynamic time warping
Wang et al. ConvPred: A deep learning‐based framework for predictions of potential organic reactions
CN114492988A (en) Method and device for predicting product yield in catalytic cracking process

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