CN116307149A - Blast furnace performance optimization method based on attention LSTM and KBNSGA - Google Patents

Blast furnace performance optimization method based on attention LSTM and KBNSGA Download PDF

Info

Publication number
CN116307149A
CN116307149A CN202310207713.4A CN202310207713A CN116307149A CN 116307149 A CN116307149 A CN 116307149A CN 202310207713 A CN202310207713 A CN 202310207713A CN 116307149 A CN116307149 A CN 116307149A
Authority
CN
China
Prior art keywords
optimization
lstm
attention
blast furnace
knowledge base
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.)
Pending
Application number
CN202310207713.4A
Other languages
Chinese (zh)
Inventor
杨春节
闫夺今
孙优贤
王文海
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN202310207713.4A priority Critical patent/CN116307149A/en
Publication of CN116307149A publication Critical patent/CN116307149A/en
Pending legal-status Critical Current

Links

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"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/12—Computing arrangements based on biological models using genetic models
    • G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • 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
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • 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
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06395—Quality analysis or management
    • 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
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/04—Manufacturing
    • 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
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services
    • 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

Abstract

The invention discloses a blast furnace performance optimization method based on attention LSTM and KBNSGA. The method aims at the collaborative optimization of the silicon content and the coke ratio of the molten iron of the blast furnace, selects a process characteristic variable, puts forward the Attention long-short-term memory network (Attention-LSTM) to predict the silicon content and the coke ratio, and takes the predicted silicon content and the coke ratio as the fitness evaluation function of a multi-target optimization model. And establishing a historical optimal operation index knowledge base, and guiding an NSGA-II genetic algorithm to initialize a population by using the knowledge base to form a KBNSGA multi-objective optimization method. And screening the approximate Pareto front solution set by adopting a TOPSIS evaluation method to obtain an optimized solution meeting the actual production requirement. The comparison of the method of the invention with other genetic optimization methods shows that the method has excellent performance, and can realize the optimization requirements of stable quality and reduced energy consumption of blast furnace ironmaking.

Description

Blast furnace performance optimization method based on attention LSTM and KBNSGA
Technical Field
The invention belongs to the field of data-based industrial process optimization and decision making, and relates to an improved method for optimizing blast furnace performance indexes by using attention LSTM and KBNSGA.
Background
The iron and steel manufacturing industry is the basic core industry of national economy, and the blast furnace ironmaking is used as the most critical process of iron and steel smelting, and the energy consumption accounts for about seven times of the total energy consumption of iron and steel production. Although the output of the blast furnace ironmaking industry in China is ranked in the world, the economic benefit and the environmental benefit are inferior to the world advanced level, which greatly restricts the sustainable development of the iron and steel industry and reduces the competitive power of the international market. Some blast furnace ironworks cannot produce with low energy consumption and low pollution, so that finished products lack competitiveness, and the existence of enterprises is critical.
Because the iron-smelting blast furnace has the characteristics of high temperature, high pressure, continuity and sealing, the iron-smelting blast furnace belongs to a large-scale reaction vessel, and the production efficiency is influenced by the characteristics of raw materials and the heat, quality and momentum transfer states among various materials in the reaction process. Therefore, the blast furnace ironmaking process is a complex system with strong interference, dynamic change, nonlinearity and time lag. In order to realize the greenization and the intellectualization of the blast furnace ironmaking process, the blast furnace is operated under a higher-quality working condition, a prediction model of the blast furnace quality index and the energy consumption index is established, the influence of the operation index on the ironmaking performance is accurately predicted, and the method has positive guiding significance for production. Meanwhile, the quality index and the energy consumption index of the blast furnace can be optimized according to the requirements, so that the purposes of high-efficiency production, energy conservation and emission reduction are achieved, and the blast furnace is enabled to run smoothly in a stable section.
The research design of the blast furnace operation performance index optimization method based on data and knowledge is necessary to realize the collaborative optimization of the blast furnace quality index (molten iron silicon content) and the energy consumption index (coke ratio), ensures the good product quality while considering the economic benefit, and has important significance for realizing the intellectualization and informatization of the industrial production process.
Traditional mechanism models, experience knowledge models and the like, such as regression models, support vector machines, BP neural networks, simulated annealing algorithms and the like, can realize certain nonlinear feature extraction and optimization, but cannot fully utilize time sequence information and historical knowledge information, cannot accurately predict and optimize energy consumption indexes and quality indexes in the blast furnace production process, and can cause the problems of data information loss and difficulty in floor application of actual production scenes.
The invention comprises the following steps:
in order to overcome the defects of the existing method, the time sequence information and the historical knowledge information are fully utilized, and the optimization algorithm is enabled to operate efficiently under the condition of dynamic change of the production working condition of the blast furnace, and the invention provides a blast furnace performance optimization method based on attention LSTM and KBNSGA.
The blast furnace performance optimization method based on the attention LSTM and the KBNSGA comprises the following steps:
step (1), selecting 16 process characteristic variables related to the silicon content and the coke ratio of molten iron through correlation analysis and an iron-making mechanism, and inputting a prediction model;
step (2) adding a spatial feature Attention mechanism (Attention) on the basis of a long-short-term memory network (LSTM) to form an Attention-LSTM model, realizing weighting processing on different feature dimensions, adaptively focusing on the feature with large correlation with a prediction target, and predicting the molten iron silicon content and the coke ratio;
step (3), a blast furnace ironmaking historical Knowledge base (knowledgebase) is established, a K neighbor method is used for working condition division, and working condition characteristics and a historical optimization optimal solution are stored; when new sampling data enter an optimization period, carrying out working condition searching and matching in a knowledge base, and if the working condition exists in the knowledge base, carrying out NSGA-II genetic algorithm initialization by using a historical optimization optimal solution; if the working condition does not exist in the knowledge base, randomly initializing NSGA-II parameters, and carrying out global optimal search; the whole flow constitutes a KBNSGA method;
and (4) selecting an optimal scheme in a Pareto front of the KBNSGA solution by using a TOPSIS evaluation method, so that a final optimization result meets the production requirement and is close to an ideal global optimal solution.
The 16 process characteristic variables selected through the correlation analysis and the ironmaking mechanism in the step (1) are oxygen enrichment rate, air permeability index, oxygen enrichment flow, cold air flow, furnace belly gas quantity, furnace belly gas index, blast kinetic energy, top pressure, full pressure difference, hot air pressure, actual wind speed, hot air temperature, furnace top temperature, resistance coefficient, blast temperature and actual coal injection quantity.
The Attention-LSTM model calculation method is characterized in that:
an attribute mechanism layer is added in front of an LSTM layer, and the input sequence of the model at the time t is
Figure BDA0004111463500000021
Wherein n is the feature variable dimension, X t First, the LSTM hidden layer state h at the same time t-1 Inputting the weight value into an Attention mechanism layer for carrying out weight calculation among characteristic dimensions to obtain a weight value, and obtaining n-dimensional variable weight at t moment after Softmax calculation>
Figure BDA0004111463500000022
With the original input sequence X t Weighting to obtain weighted input features +.>
Figure BDA0004111463500000023
Handle->
Figure BDA0004111463500000024
Inputting LSTM network layer to obtain hidden layer state h at t moment t And predicted values of silicon content and coke ratio.
The KBNSGA method described in the step (3):
a historical data knowledge base is established in an NSGA-II genetic algorithm optimization model, firstly, the knowledge base is established by utilizing the existing historical data optimization cases, K neighbor working condition division is carried out, and working condition characteristics and the corresponding historical optimization optimal solutions are stored in the knowledge base. And carrying out similarity calculation on the newly acquired data according to the distance measurement, carrying out NSGA-II population initialization by using a history optimization optimal solution when the similarity is larger than a threshold delta, and randomly initializing the population if the similarity is smaller than the threshold. And (3) taking stable silicon content and minimized coke ratio as optimization targets, taking process characteristic variables as decision variables, and performing population selection, intersection and mutation operations after initialization until algorithm termination conditions are met, so as to obtain an optimized Pareto front.
The TOPSIS evaluation method described in the step (4):
firstly, the Pareto front data are normalized positively, the maximum and minimum values of each characteristic dimension are taken to form an optimal scheme and a worst scheme, the similarity degree of each scheme to be evaluated and the optimal worst scheme is obtained through the distance between the scheme to be evaluated and the optimal worst scheme, the similarity degree is used as the basis for evaluating the quality and the like, the method is applied to the multi-objective optimization problem of the blast furnace, the quality degree of each solution in the optimization scheme is evaluated, and the optimal scheme in the optimized approximate Pareto front is selected.
The beneficial results of the invention are:
1. the designed blast furnace ironmaking working condition knowledge base solves the problem that the conventional optimization method needs to be continuously updated and maintained due to environmental interference and working condition dynamic change, and can simultaneously store working condition and optimization result information, so that an optimization algorithm can be quickly converged to a target domain by utilizing historical information.
2. Based on the long-term and short-term memory neural network of the attention mechanism, the weighting processing on different feature dimensions is realized, time sequence information and spatial feature information are focused at the same time, and the prediction effect is improved.
3. The best scheme in the approximate Pareto front after optimization can be selected by using the TOPSIS method to effectively evaluate the quality degree of each solution in the optimization scheme.
4. The invention can effectively optimize the quality index and the energy consumption index of the blast furnace, thereby achieving the purposes of high-efficiency production, energy conservation and emission reduction.
Drawings
Fig. 1 shows the structure of an Attention unit.
FIG. 2 shows the overall structure of the Attention-LSTM prediction model.
FIG. 3 shows the comparison of the predicted results and actual data of the different methods of molten iron silicon content.
FIG. 4 shows the KBNSGA process flow.
Fig. 5 shows the result of optimizing the silicon content of molten iron.
Fig. 6 shows the results of the coke ratio optimization.
Detailed Description
The invention is further illustrated in the following figures and examples.
The invention aims to provide a blast furnace performance optimization method based on attention LSTM and KBNSGA, which can be used for predicting the molten iron silicon content and the coke ratio according to process operation parameters in the blast furnace ironmaking process, optimizing performance indexes and providing a feasible optimization scheme for production site operators.
The invention mainly comprises: the production index prediction model and the performance index optimization model are designed based on data and knowledge, and the two model algorithms comprise a soft measurement model based on the Attention-LSTM and a multi-objective optimization model based on the KBNSGA. The functions of blast furnace ironmaking mechanism and historical data are comprehensively considered, and the genetic algorithm searching area is guided to be concentrated in a reasonable range through the self-built historical optimization case knowledge base, so that the convergence speed and the calculation efficiency of the algorithm are improved. The TOPSIS evaluation mechanism is introduced to realize the selection of the optimal solution, and a scheme conforming to the actual situation is provided for the blast furnace production field.
1. Attention-LSTM soft measurement model structure:
LSTM forward calculation method formulation:
f t ＝σ(W xf x t +W hf h t-1 +W cf c t-1 +b f )#(1)
i t ＝σ(W xi x t +W hi h t-1 +W ci c t-1 +b i )#(2)
c t ＝f t c t-1 +i t tanh(W xc x t +W hc h t-1 +b c )#(3)
o t ＝σ(W xo x t +W ho h t-1 +W co c t +b o )#(4)
h t ＝o t tanh(c t )#(5)
the LSTM memory cell comprises three gate units, namely a forgetting gate f, an input gate i and an output gate o, which commonly affect memory finenessCell state c. Wherein sigma is an activation function sigmoid; w is a variable weight coefficient, b represents a bias term; tanh is a hyperbolic tangent function; c t-1 Memory cell state at time t-1; c t Representing the updated memory cell state; h is a t Is the output of LSTM at time t. In order to effectively distribute computing power, a prediction result is more accurate, and a Attention mechanism, namely an Attention layer, is added on the basis of an LSTM layer.
The calculation formula of the Attention mechanism is as follows:
Figure BDA0004111463500000041
Figure BDA0004111463500000042
Figure BDA0004111463500000043
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure BDA0004111463500000044
an input sequence representing a time t; v, W, U, b is a parameter that the model needs to learn; h is a t-1 The LSTM unit output is the last moment, namely (t-1); alpha ki Inputting attention weights of hidden layer states for histories; />
Figure BDA0004111463500000045
The original input sequence at time t is weighted by attention and then output, and +.>
Figure BDA0004111463500000046
Instead of X t Is input into the LSTM network. The Attention cell structure is shown in figure 1.
The original data sequence is preprocessed and then input into the Attention-LSTM network model, as shown in figure 2. And adding two full-connection layers after the method improves network performance, and finally outputting the predicted target, namely the silicon content and the coke ratio. The predicted silicon content results are shown in FIG. 3.
2. Improved KBNSGA optimization method
The optimization objective of the method is to minimize the focal ratio y CR Silicon content y in molten iron HMSC At the expected value
Figure BDA0004111463500000047
The vicinity is stable. The decision variable is a selected operation process variable, and the constraint condition is satisfied. NSGA-II is a genetic algorithm for solving the multi-objective optimization problem, the initial population size N is determined by a designer, and the parent population at the beginning of each generation of iteration is P t Calculating fitness function, and then performing selection, crossover and mutation operations to obtain newly generated population Q t . Combining the original population and the newly generated population into R t In the pair R t And performing non-dominant hierarchical ordering, wherein the critical layer selects individuals to enter the next generation according to a rule that the larger the congestion distance is, the more preferential, and iterating until a termination condition is met.
Because blast furnace ironmaking is an extremely complex production scene, environmental interference and dynamic change of working conditions occur sometimes, and the soft measurement model and the multi-objective optimization model are required to be updated and maintained continuously. In a new optimization period, effective application history information is needed to enable the algorithm to run efficiently, a history data knowledge base is established by using a knowledge expression-based method, working condition division is carried out, and working condition characteristics and a corresponding history optimization optimal solution are stored in the knowledge base. When a new optimization cycle starts, the similarity of the current production condition and the current production condition is judged by using working condition data stored in a knowledge base, and a historical optimal decision variable is used when a genetic algorithm is initialized, so that the current production condition and the current production condition can be quickly converged to a target search domain.
The attribute characteristic value description method is used as a main stream method in knowledge expression, has the characteristic of clear expression relation, and is suitable for establishing a blast furnace ironmaking history working condition knowledge base. Specifically, firstly, a knowledge base is established by using the existing data set to optimize cases and working conditions are divided, when the ith sampling data enter an optimization period, working condition searching and matching are carried out in the knowledge base, new cases are generated and added into the knowledge base, and the ith cases are expressed as follows:
E i ：<X i ；S i >，i＝1，2，...，n#(9)
wherein E is i Refers to the attribute feature overall expression of the ith batch of data, X i Is the characteristic variable of the ith batch data, namely the operation variable, S i Is X i And (5) optimizing the scheme corresponding to the operating variable under the working condition.
The working condition is divided by using a K nearest neighbor method, a blast furnace data set is marked by working condition labels according to expert experience, and the similarity of newly collected data is calculated according to distance measurement:
Figure BDA0004111463500000051
Figure BDA0004111463500000052
wherein sim (x, x k ) Representing newly added data set X and case X in knowledge base k The similarity of (2) ranges from [0,1 ]]The closer to 1 the higher the similarity. d (x, x) k ) X and X are represented by k Euclidean distance between them. If the condition is satisfied:
maxsim(x，x k )≥δk＝0，1，2，...，m#(12)
x and X k Can be regarded as the same working condition, wherein delta is the threshold value for judging the similarity of the working conditions, and [0.5,0.8 ] is generally taken]If the threshold value is smaller than the threshold value, a random initialized genetic algorithm is used for optimization, and the genetic algorithm is added into a knowledge base to serve as a new working condition. The improved genetic algorithm for initializing populations using Knowledge base historic optimal solutions is called KbNSGA (knowledgebase-based Non-dominated Sorting Genetic Algorithm) and the algorithm flow is shown in fig. 4.
3. TOPSIS-based optimized result evaluation screening mechanism
The TOPSIS method (Technique for Order Preference by Similarity to Ideal Solution) is an adaptive comprehensive evaluation method. The basic flow is that first, original data is normalized positively, the maximum and minimum values of each characteristic dimension are taken to form an optimal scheme and a worst scheme, and the similarity degree of each scheme to be evaluated and the optimal worst scheme is obtained through the distance between the schemes to be evaluated and the best worst scheme, so that the similarity degree is used as the basis for evaluating the quality. The TOPSIS method is applied to the blast furnace multi-objective optimization problem, the quality degree of each solution in the optimization scheme is evaluated, and the optimal scheme in the optimized approximate Pareto front is selected. The following steps are performed:
Figure BDA0004111463500000053
Figure BDA0004111463500000061
and selecting a solution with the priority of score ranking from the approximate Pareto front according to the steps, and taking the solution as a final optimized solution for practical operation.
4. Optimization result verification
In order to verify the effectiveness of the KBNSGA method, the optimization effect evaluation indexes of the random initialization population and the initialization population according to the historical optimal solution are compared. The KBNSGA and common multi-objective optimization algorithms including NSGA-II, NSGA-III and reference vector guided multi-objective evolutionary algorithm (RVEA) are respectively subjected to example tests, and the same iteration parameters are set. The optimization effect is evaluated according to the distribution index, and the more uniform distribution among individuals with similar Pareto fronts in the target space shows that the better the algorithm performance is, and the two indexes are a hyper volume index (HV) and a Spacing index. The results are shown in the following table, and compared with the optimization results, the solution time of the KBNSGA and the HV are dominant, so that the convergence rate of the KBNSGA is high, and the overall comprehensive performance is good.
Optimization algorithm Solving time/s HV (super volume index) Spacing (measuring distance)
NSGA-II 227.4 0.099 0.487
NSGA-III 225.7 0.112 0.411
RVEA 234.8 0.095 0.158
KbNSGA 218.6 0.115 0.191
As shown in fig. 5 and 6, stability of long-time optimized operation results proves effectiveness and robustness of the method, after the optimized model is practically applied, silicon content change is averagely reduced by 3%, coke ratio is averagely reduced by 8.4kg/t, optimization of quality index and energy consumption index is realized, and purposes of energy conservation, emission reduction and efficient production are achieved.
The above examples merely represent a few embodiments of the present invention, which are described in more detail and are not to be construed as limiting the scope of the invention. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the invention, which are all within the scope of the invention. Accordingly, the scope of the invention should be assessed as that of the appended claims.

Claims (5)

1. A blast furnace performance optimization method based on attention LSTM and KBNSGA is characterized by comprising the following steps:
step (1), selecting process characteristic variables related to the silicon content and the coke ratio of molten iron through correlation analysis and an iron-making mechanism, and inputting a prediction model;
step (2) adding a spatial feature Attention mechanism (Attention) on the basis of a long-short-term memory network (LSTM) to form an Attention-LSTM model, realizing weighting processing on different feature dimensions, adaptively focusing on the feature with large correlation with a prediction target, and predicting the molten iron silicon content and the coke ratio;
step (3), a blast furnace ironmaking historical Knowledge base (knowledgebase) is established, a K neighbor method is used for working condition division, and working condition characteristics and a historical optimization optimal solution are stored; when new sampling data enter an optimization period, carrying out working condition searching and matching in a knowledge base, and if the working condition exists in the knowledge base, carrying out NSGA-II genetic algorithm initialization by using a historical optimization optimal solution; if the working condition does not exist in the knowledge base, randomly initializing NSGA-II parameters, and carrying out global optimal search; the whole flow constitutes a KBNSGA method;
and (4) selecting an optimal scheme in a Pareto front of the KBNSGA solution by using a TOPSIS evaluation method, so that a final optimization result meets the production requirement and is close to an ideal global optimal solution.
2. The method of claim 1, wherein the process characteristic variables selected by correlation analysis and ironmaking mechanism in step (1) are oxygen enrichment rate, air permeability index, oxygen enrichment flow, cold air flow, belly gas index, blast kinetic energy, top pressure, full pressure difference, hot air pressure, actual wind speed, hot air temperature, furnace top temperature, drag coefficient, blast temperature, actual coal injection amount.
3. The method of claim 1, wherein the Attention-LSTM model calculation method in step (2) is as follows:
an attribute mechanism layer is added in front of an LSTM layer, and the input sequence of the model at the time t is
Figure FDA0004111463490000011
Wherein n is the feature variable dimension, X t First, the LSTM hidden layer state h at the same time t-1 Inputting the weight value into an Attention mechanism layer for carrying out weight calculation among characteristic dimensions to obtain a weight value, and obtaining n-dimensional variable weight at t moment after Softmax calculation>
Figure FDA0004111463490000012
With the original input sequence X t Weighting to obtain weighted input features +.>
Figure FDA0004111463490000013
Handle->
Figure FDA0004111463490000014
Inputting LSTM network layer to obtain hidden layer state h at t moment t And predicted values of silicon content and coke ratio.
4. The method according to claim 1, wherein the KbNSGA method of step (3) is as follows:
establishing a historical data knowledge base in an NSGA-II genetic algorithm optimization model, firstly establishing the knowledge base by utilizing the existing historical data optimization cases, dividing the working conditions of K neighbors, and storing the working condition characteristics and the corresponding historical optimization optimal solutions in the knowledge base; calculating the similarity of the newly acquired data according to the distance measurement, initializing NSGA-II population by using a history optimization optimal solution when the similarity is larger than a threshold delta, and randomly initializing the population if the similarity is smaller than the threshold delta; and (3) taking stable silicon content and minimized coke ratio as optimization targets, taking process characteristic variables as decision variables, and performing population selection, intersection and mutation operations after initialization until algorithm termination conditions are met, so as to obtain an optimized Pareto front.
5. The method of claim 1, wherein the TOPSIS evaluation method of step (4) is as follows:
firstly, the Pareto front data are normalized positively, the maximum and minimum values of each characteristic dimension are taken to form an optimal scheme and a worst scheme, the similarity degree of each scheme to be evaluated and the optimal worst scheme is obtained through the distance between the scheme to be evaluated and the optimal worst scheme, the similarity degree is used as the basis for evaluating the quality and the like, the method is applied to the multi-objective optimization problem of the blast furnace, the quality degree of each solution in the optimization scheme is evaluated, and the optimal scheme in the optimized approximate Pareto front is selected.
CN202310207713.4A 2023-03-07 2023-03-07 Blast furnace performance optimization method based on attention LSTM and KBNSGA Pending CN116307149A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310207713.4A CN116307149A (en) 2023-03-07 2023-03-07 Blast furnace performance optimization method based on attention LSTM and KBNSGA

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310207713.4A CN116307149A (en) 2023-03-07 2023-03-07 Blast furnace performance optimization method based on attention LSTM and KBNSGA

Publications (1)

Publication Number Publication Date
CN116307149A true CN116307149A (en) 2023-06-23

Family

ID=86818001

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310207713.4A Pending CN116307149A (en) 2023-03-07 2023-03-07 Blast furnace performance optimization method based on attention LSTM and KBNSGA

Country Status (1)

Country Link
CN (1) CN116307149A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116837422A (en) * 2023-07-24 2023-10-03 扬中凯悦铜材有限公司 Production process and system of high-purity oxygen-free copper material
CN117313554A (en) * 2023-11-28 2023-12-29 中国科学技术大学 Multi-section combined multi-objective optimization method, system, equipment and medium for coking production

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116837422A (en) * 2023-07-24 2023-10-03 扬中凯悦铜材有限公司 Production process and system of high-purity oxygen-free copper material
CN116837422B (en) * 2023-07-24 2024-01-26 扬中凯悦铜材有限公司 Production process and system of high-purity oxygen-free copper material
CN117313554A (en) * 2023-11-28 2023-12-29 中国科学技术大学 Multi-section combined multi-objective optimization method, system, equipment and medium for coking production
CN117313554B (en) * 2023-11-28 2024-03-29 中国科学技术大学 Multi-section combined multi-objective optimization method, system, equipment and medium for coking production

Similar Documents

Publication Publication Date Title
Lu et al. Carbon trading volume and price forecasting in China using multiple machine learning models
Wang et al. A multiobjective evolutionary nonlinear ensemble learning with evolutionary feature selection for silicon prediction in blast furnace
CN116307149A (en) Blast furnace performance optimization method based on attention LSTM and KBNSGA
CN108846526A (en) A kind of CO2 emissions prediction technique
CN106845012A (en) A kind of blast furnace gas system model membership function based on multiple target Density Clustering determines method
CN109242188B (en) Long-term interval prediction and structure learning method for steel gas system
Zeng et al. Prediction of fluctuation loads based on GARCH family-CatBoost-CNNLSTM
CN105404926A (en) Aluminum electrolytic production technology optimization method based on BP neural network and MBFO algorithm
Wang et al. Strip hardness prediction in continuous annealing using multiobjective sparse nonlinear ensemble learning with evolutionary feature selection
CN114239400A (en) Multi-working-condition process self-adaptive soft measurement modeling method based on local double-weighted probability hidden variable regression model
Xu et al. Multi-cluster based equilibrium optimizer algorithm with compact approach for power system network
Yin et al. Dynamic real–time abnormal energy consumption detection and energy efficiency optimization analysis considering uncertainty
CN110849149A (en) Energy perception-based sintering batching scheme cascade optimization obtaining method and device
Li et al. Deep learning for predictive mechanical properties of hot-rolled strip in complex manufacturing systems
CN111192158A (en) Transformer substation daily load curve similarity matching method based on deep learning
CN105426959B (en) Aluminium electroloysis energy-saving and emission-reduction method based on BP neural network Yu adaptive M BFO algorithms
Wang et al. Multi-objective optimal scheduling of laminar cooling water supply system for hot rolling mills driven by digital twin for energy-saving
CN105302976A (en) Aluminum electrolysis production optimization method based on SPEA2 algorithm
CN117454765A (en) Copper smelting furnace spray gun service life prediction method based on IPSO-BP neural network
CN116702613A (en) Multi-objective optimization method and system for product performance and sintering energy consumption of ternary positive electrode material
CN113900379A (en) Neural network-based predictive control algorithm
Yan et al. Multi-objective optimization method of blast furnace operation performance indicators based on data and knowledge
CN105426960A (en) BP neural network and MBFO algorithm-based aluminum electrolysis energy conservation and emission reduction control method
Xudong et al. Prediction of building energy consumption based on IPSO-CLSTM neural network
CN110135624A (en) A kind of data predication method of the combination LSTM model based on 2-D data stream

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