CN113723686B - Multitask ash box prediction method and system for energy consumption in organic silicon monomer fractionation process - Google Patents
Multitask ash box prediction method and system for energy consumption in organic silicon monomer fractionation process Download PDFInfo
- Publication number
- CN113723686B CN113723686B CN202111016250.0A CN202111016250A CN113723686B CN 113723686 B CN113723686 B CN 113723686B CN 202111016250 A CN202111016250 A CN 202111016250A CN 113723686 B CN113723686 B CN 113723686B
- Authority
- CN
- China
- Prior art keywords
- model
- energy consumption
- representing
- parameter
- test
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000005265 energy consumption Methods 0.000 title claims abstract description 105
- 238000000034 method Methods 0.000 title claims abstract description 93
- 230000008569 process Effects 0.000 title claims abstract description 51
- 238000005194 fractionation Methods 0.000 title claims abstract description 45
- 239000000178 monomer Substances 0.000 title claims abstract description 36
- XUIMIQQOPSSXEZ-UHFFFAOYSA-N Silicon Chemical compound [Si] XUIMIQQOPSSXEZ-UHFFFAOYSA-N 0.000 title claims abstract description 32
- 229910052710 silicon Inorganic materials 0.000 title claims abstract description 32
- 239000010703 silicon Substances 0.000 title claims abstract description 32
- 238000012360 testing method Methods 0.000 claims abstract description 84
- 238000012549 training Methods 0.000 claims abstract description 46
- 239000011159 matrix material Substances 0.000 claims description 28
- 238000012545 processing Methods 0.000 claims description 26
- 238000010606 normalization Methods 0.000 claims description 22
- 230000006870 function Effects 0.000 claims description 16
- 238000011156 evaluation Methods 0.000 claims description 5
- 238000000638 solvent extraction Methods 0.000 claims description 5
- 238000012843 least square support vector machine Methods 0.000 claims description 4
- 238000010992 reflux Methods 0.000 claims description 4
- 238000010845 search algorithm Methods 0.000 claims description 3
- 230000007547 defect Effects 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 11
- 238000005457 optimization Methods 0.000 description 8
- 238000004590 computer program Methods 0.000 description 7
- 229920001296 polysiloxane Polymers 0.000 description 7
- 238000004519 manufacturing process Methods 0.000 description 4
- 230000007246 mechanism Effects 0.000 description 4
- 238000012706 support-vector machine Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000004821 distillation Methods 0.000 description 3
- 239000002245 particle Substances 0.000 description 3
- 238000000926 separation method Methods 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 239000007788 liquid Substances 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 241001481710 Cerambycidae Species 0.000 description 1
- 238000009835 boiling Methods 0.000 description 1
- 238000002939 conjugate gradient method Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000012994 industrial processing Methods 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000009413 insulation Methods 0.000 description 1
- 239000007791 liquid phase Substances 0.000 description 1
- 238000002156 mixing Methods 0.000 description 1
- 239000003921 oil Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000704 physical effect Effects 0.000 description 1
- 238000011112 process operation Methods 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 238000001308 synthesis method Methods 0.000 description 1
- 239000004753 textile Substances 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
- B01D—SEPARATION
- B01D3/00—Distillation or related exchange processes in which liquids are contacted with gaseous media, e.g. stripping
- B01D3/14—Fractional distillation or use of a fractionation or rectification column
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
- B01D—SEPARATION
- B01D3/00—Distillation or related exchange processes in which liquids are contacted with gaseous media, e.g. stripping
- B01D3/14—Fractional distillation or use of a fractionation or rectification column
- B01D3/143—Fractional distillation or use of a fractionation or rectification column by two or more of a fractionation, separation or rectification step
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/04—Constraint-based CAD
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Human Resources & Organizations (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Strategic Management (AREA)
- General Engineering & Computer Science (AREA)
- Chemical & Material Sciences (AREA)
- Economics (AREA)
- Chemical Kinetics & Catalysis (AREA)
- Data Mining & Analysis (AREA)
- Game Theory and Decision Science (AREA)
- General Business, Economics & Management (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Development Economics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a multitask ash box prediction method for energy consumption in an organic silicon monomer fractionation process, which comprises the steps of determining a first principle model of a rectifying tower; acquiring operation data, and dividing the operation data into a training set and a test set; training the first principle model by using a training set to obtain a model parameter vector to be estimated, and assuming the model parameter vector wtBy a shared parameter w0And a specific parameter vtComposition, estimating model parameters at the same time; establishing an energy consumption model of the rectifying tower; and testing the energy consumption model by using the test set to obtain the energy consumption predicted value of the test sample in the test set, and comparing the energy consumption predicted value with the energy consumption actual value of the test sample to evaluate the performance of the energy consumption model. The dynamic characteristic of the rectifying tower and the similarity between different rectifying towers in the organic silicon monomer fractionation process are utilized, the model generalization performance is improved, and the defects that the modeling cost is high, the correlation among the towers is ignored and the modeling is difficult under the condition of insufficient data in the energy consumption prediction method in the organic silicon monomer fractionation process in the prior art are overcome.
Description
Technical Field
The invention relates to the technical field of process industrial production and processing, in particular to a multitask ash box prediction method and a multitask ash box prediction system for energy consumption in an organic silicon monomer fractionation process.
Background
The organosilicon material has the excellent characteristics of corrosion resistance, radiation resistance, high and low temperature resistance, electric insulation, flame retardance, good biocompatibility and the like, and is widely applied to the fields of food, textile, building, automobile, chemical industry, medicine, instruments, aerospace and the like. The organosilicon monomer produced by the direct synthesis method has more components, close boiling points, small relative volatility and difficult separation. Therefore, the organic silicon monomer fractionation process needs more substances to be separated and has high separation requirements, and the monomer is generally purified by adopting a multi-tower continuous rectification process, so that the organic silicon monomer fractionation process is a unit with higher energy consumption in the organic silicon production process. Wherein the energy consumption of each column is proportional to its separation capacity. The method has the advantages that an explanatory high-precision model of the relation between the energy consumption of each tower and the process variable in the fractionation process is established at low cost, and the method has important significance for improving the economic benefit of the organic silicon industry.
In the prior art, a data model established based on fractionation process operation data is simple in structure but lacks interpretable physical significance; the accuracy of the first principle model established based on the dynamic characteristics of rectification requires an extremely complex model structure and a large number of parameters which are difficult to calculate, and the ash box model estimates the parameters in the first principle model by using a data-driven method, so that the model complexity is reduced while the interpretability is maintained. However, modeling each rectifying column separately is not only costly, but also ignores the potential relationships and dependencies between columns. Also in situations where it is difficult to collect enough modeling data for a tower, such as sensor damage for a critical variable, it is not possible to model the tower alone.
In conclusion, the energy consumption prediction method for the organic silicon monomer fractionation process in the prior art has the problems of high modeling cost, neglect of the correlation among towers and difficulty in modeling under the condition of insufficient data.
Disclosure of Invention
Therefore, the technical problem to be solved by the invention is to overcome the defects that the energy consumption prediction method in the organic silicon monomer fractionation process in the prior art has high modeling cost, ignores the correlation among towers and is difficult to model under the condition of insufficient data.
In order to solve the technical problem, the invention provides a multitask ash box prediction method for energy consumption in an organic silicon monomer fractionation process, which comprises the following steps:
determining an ash box energy consumption prediction model structure, including determining a first principle model of each rectifying tower;
acquiring operation data of a steady-state operation condition in an organic silicon fractionation process, randomly dividing the operation data into a training set and a test set, and carrying out normalization processing on the training set and the test set;
training the first principle model of the rectifying tower by using the training set after normalization processing to obtain a model parameter vector to be estimated, and obtaining a basisCorrelation between multiple rectification columns, assuming model parameter vector wtBy a shared parameter w0And a specific parameter vtComposition in which the parameter w is shared0Capturing similarity between rectification columns, specific parameter vtCapturing differences among the rectifying towers, and simultaneously estimating model parameters by using a multi-task learning algorithm;
establishing an energy consumption model of the rectifying tower according to the model parameters;
and testing the energy consumption model of the rectifying tower by using the test set after normalization processing to obtain an energy consumption predicted value of the test sample in the test set, and comparing the energy consumption predicted value of the test sample with the energy consumption actual value of the test sample to evaluate the performance of the energy consumption model.
In one embodiment of the invention, the operational data includes feed flow, reflux flow, draw flow, and condensate flow.
In one embodiment of the present invention, randomly partitioning the operational data into training sets comprises:
randomly partitioning the operational data into training sets { (x)it,yit),i=1,2,…,ntWherein (x)it,yit) I-th training sample, n, representing the t-th tower training settRepresenting the number of training samples for that tower.
In one embodiment of the present invention, randomly partitioning the operational data into test sets comprises:
randomly partitioning the operational data into test sets { (x)jt,yjt),j=1,2,…mtWherein (x)jt,yjt) J test sample, m, representing the t tower test settRepresenting the number of test samples for that tower.
In an embodiment of the present invention, training the first principle model of the rectifying tower by using the training set after the normalization processing, and obtaining the model parameter vector to be estimated includes:
the first principle model of the tth rectifying tower is established as follows:
wherein the model parameter vector to be estimated is wt=[w1t w2t w3t w4t]TModel input matrixModel output matrixRepresenting model errors, whereinIs ntA unit column vector of dimensions.
In one embodiment of the present invention, and using a multitask learning algorithm to simultaneously estimate model parameters comprises:
the objective function form of the multi-task learning algorithm is a least square support vector machine, and the objective function is as follows:
s.t.yit=xit(w0+vt)+bt+ξit,
i=1,2,…,nt,t=1,2,…T
wherein ξitRepresenting a relaxation variable representing the tolerance of the model to outliers, gamma representing a constrained relaxation variable ξitλ denotes the control-specific variable vtThe regularization parameter of (a);
introducing Lagrange multiplier alpha ═ alpha1 … αt … αT]TSolving the above minimization problem with equality constraints, each element in alpha is greater than 0, wherein,calculating a deviation to obtainSolving the system of linear equations and converting it into a matrix form:wherein,representing the output vector, b ═ b1 b2 … bt…bT]TRepresenting an error vector, Zt×tAll 0 matrices, Z, representing dimensions t x ttRepresenting a full 0-column vector of dimension t,denotes a block diagonal matrix, H ═ blockdiag (H)1,H2,…,Ht)∈n×nRepresenting a positive definite matrix with diagonal elements as a block diagonal matrix Ht=blockdiag(ht,ht,…,ht),
Obtaining alpha ═ alpha by matrix inversion1 … αt … αT]TAndobtaining an estimated parameter of the energy consumption prediction model as
In one embodiment of the invention, the regularization parameters γ and λ are selected to be solved by a hyper-parameter search algorithm.
In one embodiment of the present invention, the modeling the energy consumption of the rectifying tower according to the model parameters comprises:
the energy consumption model of the t-th rectifying tower is established as follows:
In an embodiment of the present invention, comparing the predicted energy consumption value of the test sample with the actual energy consumption value of the test sample to evaluate the performance of the energy consumption model includes:
and (3) evaluating the performance of the model by using the mean absolute error MAE:
wherein,representing the predicted value of energy consumption, y, of the test samplejtRepresenting the actual value of energy consumption for the test sample.
In addition, the invention also provides a multitask ash box prediction system for the energy consumption in the organic silicon monomer fractionation process, which comprises the following steps:
the first principle model determining module is used for determining an ash box energy consumption prediction model structure and comprises a first principle model determining module for determining each rectifying tower;
the data acquisition module is used for acquiring the operation data of the steady-state operation condition in the organic silicon fractionation process, randomly dividing the operation data into a training set and a test set, and carrying out normalization processing on the training set and the test set;
a model parameter estimation module for training the first principle model of the rectifying tower by using the training set after normalization processing to obtain a model parameter vector to be estimated, and assuming a model based on the correlation among a plurality of rectifying towersVector of parameters wtBy sharing a parameter w0And a specific parameter vtComposition in which the parameter w is shared0Capturing similarity between rectification towers, and specific parameter vtCapturing differences among the rectifying towers, and simultaneously estimating model parameters by using a multi-task learning algorithm;
the model establishing module is used for establishing an energy consumption model of the rectifying tower according to the model parameters;
and the model test evaluation module is used for testing the energy consumption model of the rectifying tower by using the test set after normalization processing to obtain an energy consumption predicted value of the test sample in the test set, and comparing the energy consumption predicted value of the test sample with the energy consumption actual value of the test sample to evaluate the performance of the energy consumption model.
Compared with the prior art, the technical scheme of the invention has the following advantages:
the invention provides an energy consumption multi-task ash box prediction model by utilizing the dynamic characteristics of a rectifying tower and the similarity between different rectifying towers in the organic silicon monomer fractionation process, the ash box model has higher precision than a simplified mechanism model and is more explanatory than a data driving model, meanwhile, the multi-task algorithm fully utilizes the correlation between different rectifying towers, realizes the sharing of knowledge between towers, improves the generalization performance of the model, and solves the defects that the energy consumption prediction method in the organic silicon monomer fractionation process in the prior art has high modeling cost, neglects the correlation between towers and is difficult to model under the condition of insufficient data.
Drawings
In order that the present disclosure may be more readily and clearly understood, reference will now be made in detail to the present disclosure, examples of which are illustrated in the accompanying drawings.
FIG. 1 is a flow chart of a production process of a fractionation process of an organosilicon monomer.
Fig. 2 is a schematic structural diagram of a continuous binary rectification column according to the first embodiment.
FIG. 3 is a schematic diagram of the first embodiment and a diagram for analyzing the correlation between towers.
Fig. 4 is an error convergence curve of the optimization algorithm of the first embodiment for selecting the regularization parameter.
FIG. 5 is a comparison chart of the evaluation indexes of the single task and multi-task models in the first embodiment.
Detailed Description
The present invention is further described below in conjunction with the drawings and the embodiments so that those skilled in the art can better understand the present invention and can carry out the present invention, but the embodiments are not to be construed as limiting the present invention.
Example one
The production process of the silicone monomer fractionation process is shown in fig. 1, and the embodiment provides a multitask ash box prediction method for energy consumption of the silicone monomer fractionation process, wherein the method comprises the following steps:
the method comprises the following steps: and determining an ash box energy consumption prediction model structure, wherein the step of establishing a first principle model of each rectifying tower is included.
In the fractionation process, each distillation column only focuses on one quality index, and is considered as a binary distillation column in the modeling process, and the structure of the binary distillation column can be shown in fig. 2. Because the mechanism of the rectification process is unchanged, the structure of the established first principle model of each rectification tower is consistent. For a continuous binary rectifying tower, constant molar flow assumption, full mixing grade assumption, vapor-liquid phase equilibrium assumption and uniform temperature and pressure distribution in the tower are carried out, and heat loss is ignored.
Illustratively, the energy consumption (change of reboiler heat load in unit time) y of the T ∈ {1,2, …, T } th rectifying tower is established by using an energy dynamic balance equation, a full tower energy balance equation, a material balance equation and a heat load definitiontAnd process variable(feed flow rate F, reflux amount L, extraction amount D, condensate flow rate FC) The relationship between, i.e., the first principles model:
in the formula, btRepresenting the model error. w is at=[w1t w2t w3t w4t]TIs the t-th rectifying tower model parameter vector to be estimated, wherein:
w1t=HF+(q-1)HV'-qHL',
w2t=HV'-HL',
w3t=HV'-HL,
w4t=cC(Tco-Tci).
in the formula, HFIs the enthalpy of the feed, q is the feed thermal condition parameter, HV'To raise the enthalpy of the vapour stream in the stripping section, HL'For decreasing the enthalpy of the liquid in the stripping section, HLLowering the enthalpy of the liquid in the rectification section, cCIs the specific heat capacity of the condensed water, TcoAnd TciRespectively condensate outlet temperature and inlet temperature.
Under a single stable operation condition, the model parameter wtThe first principle model is a linear model.
Step two: the method comprises the steps of obtaining operation data of an organosilicon fractionation process under a steady-state operation condition, randomly dividing the operation data into a training set and a testing set, and carrying out normalization processing on the training set and the testing set.
Illustratively, the input variables (feed flow F, reflux L, draw D, condensate flow F) of the energy consumption prediction models of the M3 and MH columns were collected via the Plant Information System (PI) real-time databaseC) The operation data is acquired by taking a mean value every 5min, calculating the model output (the change of the heat load of the reboiler in unit time) through the physical property parameters of the heat conducting oil at corresponding moments, selecting the operation data acquired by the M3 tower and the MH tower in a steady-state operation time period by combining an operation condition record table and PCA clustering, eliminating damaged sample points in the selected operation data, and performing normalization processing to eliminate the influence of dimensions.
Illustratively, the running data is randomly partitioned into training sets { (x)it,yit),i=1,2,…,ntWherein (x)it,yit) I training sample, n, representing the t tower training settRepresenting the number of training samples for that tower. Likewise, the operational data is randomly partitioned into test sets { (x)jt,yjt),j=1,2,…mtWherein (x)jt,yjt) J test sample, m, representing the t tower test settRepresenting the number of test samples for that tower.
Step three: training the first principle model of the rectifying tower by using the training set after normalization processing to obtain a model parameter vector to be estimated, and assuming the model parameter vector w based on the correlation among a plurality of rectifying towerstBy sharing a parameter w0And a specific parameter vtComposition in which the parameter w is shared0Capturing similarity between rectification columns, specific parameter vtCapturing differences between rectification columns and simultaneously estimating model parameters (w) using a multi-task learning algorithm0,v1,v2,…,vt,…,vT)。
Step four: and establishing an energy consumption model of the rectifying tower according to the model parameters.
Step five: and testing the energy consumption model of the rectifying tower by using the test set after normalization processing to obtain an energy consumption predicted value of the test sample in the test set, and comparing the energy consumption predicted value of the test sample with the energy consumption actual value of the test sample to evaluate the performance of the energy consumption model.
In step three, according to the first principle model of the tth rectifying tower in the known fractionation process of step one, the method comprises the following steps:
wherein the model parameter vector to be estimated is wt=[w1t w2t w3t w4t]TModel input matrixModel output matrix Representing model errors, whereinIs ntA unit column vector of dimensions. Based on the correlation between the M3 column and the MH column during fractionation, the model parameter w was assumedtBy sharing a parameter w0And a specific parameter vtComposition in which the parameter w is shared0Capturing similarity between rectification columns, specific parameter vtDifferences between the rectification columns were captured as shown in fig. 3.
In step three, the model parameters (w) are estimated simultaneously using a multitask learning algorithm0,v1,v2,…,vt,…,vT) The method comprises the following steps:
1) the objective function form of the multi-task learning algorithm is a least square support vector machine, and the objective function is as follows:
s.t.yit=xit(w0+vt)+bt+ξit,
i=1,2,…,nt,t=1,2,…T
where γ is the constraint relaxation variable ξitThe relaxation variable represents the tolerance of the model to outliers, λ is the control-specific variable vtThe larger the lambda value, vtThe smaller the representative models are, the more similar the representative models are, the values of gamma and lambda are selected by a particle swarm optimization algorithm, the error convergence curve is shown in FIG. 4, the comparison model comprises a Single-task gray box (S-GB) model without using a multi-task strategy and a gray box (M-GB) directly using mixed dataMixed gray-box) models were compared. Multitask learning algorithm by solving simultaneously (w)0,v1,v2,…,vt,…,vT) And the sharing of knowledge among a plurality of towers is realized.
2) The minimization problem with equality constraints described above is solved using the lagrange multiplier method:
introducing Lagrange multiplier alpha ═ alpha1 … αt … αT]TSolving the above minimization problem with equality constraints, each element in alpha is greater than 0, wherein,calculating a deviation to obtainSolving the system of linear equations and converting it into a matrix form:wherein,representing the output vector, b ═ b1 b2 … bt… bT]TRepresenting the error vector, Zt×tAll 0 matrices, Z, representing dimensions t x ttRepresenting a full 0-column vector of dimension t,denotes a block diagonal matrix, H ═ blockdiag (H)1,H2,…,Ht)∈n×nRepresenting a positive definite matrix with diagonal elements as a block diagonal matrix Ht=blockdiag(ht,ht,…,ht),
3) Obtaining alpha ═ alpha by matrix inversion1 … αt … αT]TAndobtaining an estimated parameter of the energy consumption prediction model as
The regularization parameters γ and λ are selected and solved by a hyper-parameter search algorithm, which may be, but not limited to, an intelligent optimization algorithm such as grid search, random search, bayesian optimization, or particle swarm optimization, longicorn search, for example, as shown in fig. 4.
The objective function form of the multi-task learning algorithm may be, but is not limited to, a support vector machine and its extended form such as a least squares support vector machine, an approximate support vector machine, a sparse support vector machine, and the like.
The solution of the minimization problem with the constraint can be, but is not limited to, an intelligent optimization algorithm such as a sequence minimum optimization algorithm, a generalized sequence minimum optimization algorithm, a lagrange multiplier method, a penalty function method, a grid method, a feasible direction method, a linear approximation method, a generalized simple gradient method, a quadratic programming method or a particle swarm algorithm.
The solution of the large-scale linear equation system can be, but is not limited to, a matrix form solution, or an iterative algorithm such as an HS conjugate gradient method, Jacobi iteration, Gauss-Seidel, gradient descent and the like.
In the fourth step, the energy consumption model of the tth rectifying tower is as follows:
In step five, for the prediction model of the t-th rectifying tower, the performance of the prediction model is evaluated by using the average Absolute error mae (mean Absolute error):
wherein,to test the predicted value of energy consumption of a sample, yjtThe actual value of the energy consumption of the test sample is obtained. FIG. 5 compares MAE of the single-tasking gray box model S-GB and the multi-tasking gray box model MT-GB of the trimethyl tower and the hydrogenous tower, and it can be clearly seen from the figure that the multi-tasking gray box model simultaneously improves the accuracy of the energy consumption prediction models of the trimethyl tower and the hydrogenous tower.
The index of the model performance evaluation may be, but is not limited to, mean square error, root mean square error, mean absolute percentage error, logarithm of mean square error, and median absolute error.
According to the method, the ash box model with the serial structure is used for energy consumption prediction, namely, unknown parameters of the first principle model are estimated through a multi-task learning algorithm, the first principle model solidifies mechanism knowledge of the rectifying tower, and model parameters related to process variables are estimated through operation data and the multi-task learning algorithm.
The invention provides an energy consumption multi-task ash box prediction model by utilizing the dynamic characteristics of a rectifying tower and the similarity between different rectifying towers in the organic silicon monomer fractionation process, the ash box model has higher precision than a simplified mechanism model and is more explanatory than a data driving model, meanwhile, the multi-task algorithm fully utilizes the correlation between different rectifying towers, realizes the sharing of knowledge between towers, improves the generalization performance of the model, and solves the defects that the energy consumption prediction method in the organic silicon monomer fractionation process in the prior art has high modeling cost, neglects the correlation between towers and is difficult to model under the condition of insufficient data.
Example two
The following introduces a multitask ash box prediction system for energy consumption in the organic silicon monomer fractionation process disclosed in the second embodiment of the present invention, and a multitask ash box prediction system for energy consumption in the organic silicon monomer fractionation process described below and a multitask ash box prediction method for energy consumption in the organic silicon monomer fractionation process described above may be referred to correspondingly.
The embodiment of the invention discloses a multitask ash box prediction system for energy consumption in an organic silicon monomer fractionation process.
The first principle model determining module is used for determining an ash box energy consumption prediction model structure and comprises a first principle model for each rectifying tower;
the data acquisition module is used for acquiring operation data of a steady-state operation condition in the organic silicon fractionation process, randomly dividing the operation data into a training set and a test set, and carrying out normalization processing on the training set and the test set;
the model parameter estimation module is used for training the first principle model of the rectifying tower by utilizing the training set after normalization processing to obtain a model parameter vector to be estimated, and assuming the model parameter vector w based on the correlation among a plurality of rectifying towerstBy sharing a parameter w0And a specific parameter vtComposition in which the parameter w is shared0Capturing similarity between rectification columns, specific parameter vtCapturing differences between rectification columns and simultaneously estimating model parameters (w) using a multi-task learning algorithm0,v1,v2,…,vt,…,vT);
The model establishing module is used for establishing an energy consumption model of the rectifying tower according to the model parameters;
the model test evaluation module is used for testing the energy consumption model of the rectifying tower by using the test set after normalization processing to obtain an energy consumption predicted value of the test sample in the test set, and comparing the energy consumption predicted value of the test sample with the energy consumption actual value of the test sample to evaluate the performance of the energy consumption model.
The multitask grey box prediction system for the energy consumption of the silicone monomer fractionation process is used for implementing the multitask grey box prediction method for the energy consumption of the silicone monomer fractionation process, so that the specific implementation of the system can be seen in the above example section of the multitask grey box prediction method for the energy consumption of the silicone monomer fractionation process, and therefore, the specific implementation of the system can refer to the description of the corresponding section example, and will not be further described herein.
In addition, since the multitask ash box prediction system for the energy consumption in the silicone monomer fractionation process is used for implementing the multitask ash box prediction method for the energy consumption in the silicone monomer fractionation process, the function of the multitask ash box prediction system corresponds to the function of the method, and details are not repeated here.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
It should be understood that the above examples are only for clarity of illustration and are not intended to limit the embodiments. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. And obvious variations or modifications of the invention may be made without departing from the spirit or scope of the invention.
Claims (6)
1. A multitask ash box prediction method for energy consumption in an organic silicon monomer fractionation process is characterized by comprising the following steps:
determining an ash box energy consumption prediction model structure, including determining a first principle model of each rectifying tower;
acquiring operation data of an organosilicon fractionation process under a steady-state operation condition, randomly dividing the operation data into a training set and a test set, and carrying out normalization processing on the training set and the test set, wherein the normalization processing comprises randomly dividing the operation data into the training set { (x)it,yit),i=1,2,…,ntWherein (x)it,yit) I training sample, n, representing the t tower training settRepresenting the number of training samples of the tower, and randomly dividing the operation data into test sets { (x)jt,yjt),j=1,2,…mtWherein (x)jt,yjt) Denotes the t-th towerJ test sample of test set, mtRepresenting the number of test samples of the tower;
training the first principle model of the rectifying tower by using the training set after normalization processing to obtain a model parameter vector to be estimated, and assuming the model parameter vector w based on the correlation among a plurality of rectifying towerstBy sharing a parameter w0And a specific parameter vtComposition in which the parameter w is shared0Capturing similarity between rectification columns, specific parameter vtCapturing differences among the rectifying towers, and simultaneously estimating model parameters by using a multi-task learning algorithm, wherein the method specifically comprises the following steps:
the first principle model of the determined tth rectifying tower is as follows:
wherein the model parameter vector to be estimated is wt=[w1t w2t w3t w4t]TModel input matrixModel output matrix Representing model errors, whereinIs ntA unit column vector of dimensions;
the objective function form of the multi-task learning algorithm is a least square support vector machine, and the objective function is as follows:
s.t.yit=xit(w0+vt)+bt+ξit,
i=1,2,…,nt,t=1,2,…T
wherein ξitRepresenting a relaxation variable representing the model's tolerance to outliers, and gamma representing a constrained relaxation variable ξitλ represents the control-specific variable vtThe regularization parameter of (a);
introducing Lagrange multiplier alpha ═ alpha1 … αt… αT]TSolving the above minimization problem with equality constraints, each element in alpha is greater than 0, wherein,calculating a deviation to obtainSolving the system of linear equations and converting it into a matrix form:wherein,representing the output vector, b ═ b1 b2 … bt… bT]TRepresenting an error vector, Zt×tAll 0 matrices, Z, representing dimensions t x ttRepresenting a full 0-column vector of dimension t,denotes a block diagonal matrix, H ═ blockdiag (H)1,H2,…,Ht)∈n×nRepresenting a positive definite matrix with diagonal elements as a block diagonal matrix Ht=blockdiag(ht,ht,…,ht),
Obtaining alpha ═ alpha by matrix inversion1 … αt … αT]TAndobtaining an estimated parameter of the energy consumption prediction model as
Establishing an energy consumption model of the rectifying tower according to the model parameters;
and testing the energy consumption model of the rectifying tower by using the test set after normalization processing to obtain an energy consumption predicted value of the test sample in the test set, and comparing the energy consumption predicted value of the test sample with the energy consumption actual value of the test sample to evaluate the performance of the energy consumption model.
2. The multitask ash box prediction method for the energy consumption of the organic silicon monomer fractionation process according to claim 1, characterized in that: the operational data includes feed flow, reflux, draw and condensate flow.
3. The multitask ash box prediction method for the energy consumption of the organic silicon monomer fractionation process according to claim 1, characterized in that: and the regularization parameters gamma and lambda are selected and solved through a hyper-parameter search algorithm.
4. The multitask ash box prediction method for the energy consumption of the organic silicon monomer fractionation process according to claim 1, characterized in that: establishing the energy consumption model of the rectifying tower according to the model parameters comprises the following steps:
the energy consumption model of the t-th rectifying tower is established as follows:
5. The multitask ash box prediction method for the energy consumption of the organic silicon monomer fractionation process according to claim 1, characterized in that: comparing the energy consumption predicted value of the test sample with the energy consumption actual value of the test sample to evaluate the performance of the energy consumption model comprises:
and (3) evaluating the performance of the model by using the mean absolute error MAE:
6. A multitask ash box prediction system for energy consumption in an organic silicon monomer fractionation process is characterized by comprising the following steps:
the first principle model determining module is used for determining an ash box energy consumption prediction model structure and comprises a first principle model determining module for determining each rectifying tower;
the data acquisition module is used for acquiring operation data of the steady-state operation condition in the organic silicon fractionation process, randomly dividing the operation data into a training set and a test set, and carrying out normalization processing on the training set and the test set, wherein the normalization processing comprises the step of randomly dividing the operation data into the training set { (x)it,yit),i=1,2,…,ntWherein (x)it,yit) I training sample, n, representing the t tower training settRepresenting the number of training samples of the tower, running the towerRandom partitioning of data into test sets { (x)jt,yjt),j=1,2,…mtWherein (x)jt,yjt) J test sample, m, representing the t tower test settRepresenting the number of test samples of the tower;
a model parameter estimation module for training the first principle model of the rectifying tower by using the training set after normalization processing to obtain a model parameter vector to be estimated, and assuming a model parameter vector w based on the correlation among a plurality of rectifying towerstBy sharing a parameter w0And a specific parameter vtComposition in which the parameter w is shared0Capturing similarity between rectification towers, and specific parameter vtCapturing differences among the rectifying towers, and simultaneously estimating model parameters by using a multi-task learning algorithm, wherein the method specifically comprises the following steps:
the first principle model of the determined tth rectifying tower is as follows:
wherein the model parameter vector to be estimated is wt=[w1t w2t w3t w4t]TModel input matrixModel output matrix Representing model errors, whereinIs ntA unit column vector of dimensions;
the objective function form of the multi-task learning algorithm is a least square support vector machine, and the objective function is as follows:
s.t.yit=xit(w0+vt)+bt+ξit,
i=1,2,…,nt,t=1,2,…T
wherein ξitRepresenting a relaxation variable representing the model's tolerance to outliers, and gamma representing a constrained relaxation variable ξitλ represents the control-specific variable vtThe regularization parameter of (a);
introducing Lagrange multiplier alpha ═ alpha1…αt…αT]TSolving the above minimization problem with equality constraints, each element in alpha is greater than 0, wherein,calculating a deviation to obtainSolving the system of linear equations and converting it into a matrix form:wherein,representing the output vector, b ═ b1 b2 … bt … bT]TRepresenting an error vector, Zt×tAll 0 matrices, Z, representing dimensions t x ttRepresenting a full 0-column vector of dimension t,denotes a block diagonal matrix, H ═ blockdiag (H)1,H2,…,Ht)∈n×nMeans positive definiteMatrix, diagonal elements being block diagonal matrix Ht=blockdiag(ht,ht,…,ht),
Obtaining alpha ═ alpha by matrix inversion1 … αt … αT]TAndobtaining an estimated parameter of the energy consumption prediction model as
The model establishing module is used for establishing an energy consumption model of the rectifying tower according to the model parameters;
and the model test evaluation module is used for testing the energy consumption model of the rectifying tower by using the test set after normalization processing to obtain an energy consumption predicted value of the test sample in the test set, and comparing the energy consumption predicted value of the test sample with the energy consumption actual value of the test sample to evaluate the performance of the energy consumption model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111016250.0A CN113723686B (en) | 2021-08-31 | 2021-08-31 | Multitask ash box prediction method and system for energy consumption in organic silicon monomer fractionation process |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111016250.0A CN113723686B (en) | 2021-08-31 | 2021-08-31 | Multitask ash box prediction method and system for energy consumption in organic silicon monomer fractionation process |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113723686A CN113723686A (en) | 2021-11-30 |
CN113723686B true CN113723686B (en) | 2022-06-14 |
Family
ID=78680070
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111016250.0A Active CN113723686B (en) | 2021-08-31 | 2021-08-31 | Multitask ash box prediction method and system for energy consumption in organic silicon monomer fractionation process |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113723686B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114970375B (en) * | 2022-07-29 | 2022-11-04 | 山东飞扬化工有限公司 | Rectification process monitoring method based on real-time sampling data |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
GB2543281A (en) * | 2015-10-13 | 2017-04-19 | British Gas Trading Ltd | System for energy consumption prediction |
EP3376446A1 (en) * | 2017-03-18 | 2018-09-19 | Tata Consultancy Services Limited | Method and system for anomaly detection, missing data imputation and consumption prediction in energy data |
CN108647643B (en) * | 2018-05-11 | 2021-08-03 | 浙江工业大学 | Packed tower flooding state online identification method based on deep learning |
CN112560924B (en) * | 2020-12-07 | 2024-01-19 | 华东理工大学 | Propylene rectifying tower state monitoring method based on dynamic internal slow feature analysis |
CN112884012A (en) * | 2021-01-26 | 2021-06-01 | 山东历控能源有限公司 | Building energy consumption prediction method based on support vector machine principle |
-
2021
- 2021-08-31 CN CN202111016250.0A patent/CN113723686B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN113723686A (en) | 2021-11-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Stefatos et al. | Dynamic independent component analysis approach for fault detection and diagnosis | |
Vega et al. | Integrated design and control of chemical processes–Part I: Revision and classification | |
Bhattacharyya et al. | Simulation studies of the characteristics of a cryogenic distillation column for hydrogen isotope separation | |
CN108241901B (en) | Transformer early warning evaluation method and device based on prediction data | |
Jain et al. | Development of a soft sensor for a batch distillation column using support vector regression techniques | |
CN112989711B (en) | Aureomycin fermentation process soft measurement modeling method based on semi-supervised ensemble learning | |
CN113723686B (en) | Multitask ash box prediction method and system for energy consumption in organic silicon monomer fractionation process | |
Pan et al. | Fault detection with improved principal component pursuit method | |
CN104504288A (en) | Method for non-linear multistage intermittent process soft measurement based on multi-directional support vector cluster | |
Gao et al. | Multitask-based temporal-channelwise CNN for parameter prediction of two-phase flows | |
Zhu et al. | ConvLSTM and self-attention aided canonical correlation analysis for multioutput soft sensor modeling | |
Agustia et al. | Application of Fuzzy Support Vector Regression to Predict the Kovats Retention Indices of Flavors and Fragrances | |
CN114139446A (en) | On-line detection soft measurement method for components in special rectification process | |
Hael et al. | Visualization of rainfall data using functional data analysis | |
Kariwala et al. | Branch and bound method for regression-based controlled variable selection | |
Bulnes et al. | Comparison of plug flow and multi-node stratified tank modeling approaches regarding computational efficiency and accuracy | |
CN116644663A (en) | Transformer hot spot temperature inversion prediction method based on characteristic measurement point oil temperature | |
Sun et al. | Fault diagnosis with between mode similarity analysis reconstruction for multimode processes | |
Lashkarboluki et al. | Process modeling of 18O isotope separation using stochastic exchanges of isotopic components in a packed distillation column | |
Mjalli et al. | Modeling, simulation and control of a Scheibel liquid–liquid contactor: Part 1. Dynamic analysis and system identification | |
Shi et al. | Product quality prediction by a neural soft-sensor based on MSA and PCA | |
Zhang et al. | Phase partition and online monitoring for batch processes based on Harris hawks optimization | |
Torgashov et al. | Methyl sec-butyl ether content estimation in MTBE products via clustering-based adaptive nonlinear soft sensors | |
CN111145838A (en) | Iterative learning Kalman filtering method for penicillin fermentation process based on multidirectional data model | |
Muntean et al. | Modeling and composition estimation of an isotopic distillation column |
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 |