CN109359320B - Blast furnace index prediction method based on multiple sampling rate autoregressive distribution hysteresis model - Google Patents
Blast furnace index prediction method based on multiple sampling rate autoregressive distribution hysteresis model Download PDFInfo
- Publication number
- CN109359320B CN109359320B CN201810945553.2A CN201810945553A CN109359320B CN 109359320 B CN109359320 B CN 109359320B CN 201810945553 A CN201810945553 A CN 201810945553A CN 109359320 B CN109359320 B CN 109359320B
- Authority
- CN
- China
- Prior art keywords
- sampling rate
- model
- variable
- variables
- explained
- 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
Images
Classifications
-
- 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
Abstract
The invention discloses a blast furnace production index prediction method based on an improved multiple sampling rate autoregressive distribution hysteresis model, and belongs to the field of industrial process monitoring, modeling and simulation. Simulating the continuous production process of blast furnace ironmaking, namely completing the process of molten iron production by reaction of furnace burden from top to bottom and coal gas from bottom to top in a countercurrent contact manner, predicting a low-frequency explained variable through a high-frequency explained variable in a multi-sampling rate model, proposing residence time distribution as a weight function, simultaneously utilizing an adaptive genetic algorithm for parameter estimation, finally selecting the explained variable, and determining the input and output variables of the model according to correlation analysis. The method has good fitting effect on the time sequence data, and can be widely applied to prediction and optimization of an industrial dynamic system with multi-sampling rate characteristic and hysteresis effect.
Description
Technical Field
The invention belongs to the field of industrial process monitoring, modeling and simulation, and particularly relates to a method for improving technical indexes of blast furnace ironmaking production by MIDAS-ADL prediction.
Background
In the steel manufacturing process, a large-scale iron-making system mainly comprising a large-scale blast furnace is a key process of the material flow conversion of the ferrite, is also a link with the largest energy consumption, the largest emission and the highest production cost, and respectively accounts for 65-75% of the comprehensive energy consumption of the steel, about 80% of the total emission amount of the atmospheric pollutants of the steel and 60-70% of the total cost of the steel manufacturing. In addition, the production quality and efficiency of the ironmaking system, which are the front-end key processes of the steel process, determine the steel quality and production efficiency of the entire steel manufacturing process. Therefore, the large-scale iron-making system is a frontier position for deeply saving energy, reducing consumption, reducing emission, improving quality and increasing efficiency in the steel process.
The original fuels of most large iron-making systems in China are not natural and have variable components, the operation conditions frequently fluctuate and the rules are difficult to grasp. In addition, the field coupling and nonlinear multi-parameter coupling of a large-scale ironmaking system are extremely complex, and key parameters representing the operation performance are difficult to detect on line, so that the operation regulation and control of the ironmaking system mainly based on artificial experience knowledge are not timely, the production is in a non-optimal state for a long time, and the problems of high energy consumption, low efficiency and unstable product quality are caused. Therefore, it is urgently needed to deeply fuse the iron-making expert knowledge, the operation experience and the intelligent technology on the basis of the existing iron-making production automation and informatization, realize the information deep perception, the intelligent optimization decision and the accurate coordination control, and improve the green and intelligent level of the production process of the large-scale iron-making system in China.
However, the existing mathematical statistics and intelligent prediction methods, such as bayesian network, kalman filtering, deep learning, neural network, support vector machine, random forest, etc., are basically based on data with the same sampling rate, and data with different sampling rates are also subjected to summation or interpolation to be changed into data with the same time interval, which results in information loss and artificial information increase contained in the data, and there is no way to fully utilize the existing high-sampling-rate data information.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention aims to provide a blast furnace index prediction method based on a multiple sampling rate autoregressive distribution hysteresis model.
A blast furnace production index prediction method based on an improved multiple Sampling rate (MIDAS, Mixed Data Sampling) Autoregressive Distributed Lag (ADL) model predicts a low-frequency explained variable through a high-frequency explained variable in the multiple Sampling rate model, proposes residence time distribution as a weight function, simultaneously utilizes an adaptive genetic algorithm for parameter estimation, finally selects the explained variable, and determines a model input and output variable according to correlation analysis;
the expression of the multi-sampling rate model (MIDAS-ADL) is as follows:
y is an explained variable, X is an explanation variable, alpha is a constant term, beta is an overall multiplier of the explanation variable, gammajTo be a delay factor, mutAs a random disturbance term, ωi(θ) is a weighting function, p, q are the maximum hysteresis order of X, Y, and m is a multiple of the high sample rate to the low sample rate;
the expression of the weight function is as follows:
The adaptive genetic algorithm is used for parameter estimation in the MIDAS-ADL model:
the purpose of changing the population scale in real time according to the population performance in the evolution is achieved by calculating the population individual fitness function distribution in the iterative process and determining the increase and decrease of the population scale according to the skewness coefficient formula (11);
the relationship between the population size and the skewness coefficient is shown as the formula (12):
where SK is the skewness coefficient, ZnFor population scale, k is the population variation value.
The explained variables comprise permeability index, furnace top temperature, gas utilization rate, furnace belly gas index and theoretical combustion temperature, and the explained variables comprise molten iron yield and energy consumption coke ratio.
The method comprises the following steps of determining the input and output variables of the model according to the correlation analysis:
a. the principal component analysis method is utilized to simplify the interpretation variables,
b. interpolation is used to keep the sample rate of the interpreted variable and the interpreted variable consistent,
c. the correlation between the explained variable and the explained variable order is calculated by a Spearman order correlation coefficient,
d. and comprehensively considering the expert experience and the correlation analysis result to determine the input and output variables of the model.
The invention has the beneficial effects that: the designed weight function solves the problem that the conventional function lacks practical significance and is difficult to explain, only comprises an unknown variable, reduces the number of the variables and is favorable for parameter estimation. The improved adaptive genetic algorithm as an optimal parameter estimation method has good calculation efficiency and global convergence, and can simultaneously estimate the optimal value of each parameter of the model. Typical explanation variables are selected through correlation analysis and expert knowledge, and abundant information can be provided for the prediction method. As a whole, the invention can make a contribution to the prediction of the production index by utilizing various information in the multi-sampling rate system, and is beneficial to the assessment management and the guidance of the production operation of a production enterprise.
Drawings
FIG. 1 shows the variation relationship between population size and fitness function skewness;
FIG. 2 is a comparison of the predicted molten iron output and actual data;
FIG. 3 shows the comparison of the predicted result of the specific energy consumption and the actual data.
Detailed Description
The invention aims to provide a blast furnace production index prediction model based on an improved multiple sampling rate autoregressive distribution hysteresis model, which can realize the prediction of low-frequency explained variables through high-frequency explained variables in a multiple sampling rate system.
The weight function part in the multi-sampling rate (MIDAS) model provided by the invention mainly comprises two systems: design of the weighting function and estimation of the optimal parameters. The weight function is designed by analogy with the residence time distribution of materials in the reactor and is a core part in the MIDAS model. And the traditional numerical optimization algorithm is abandoned, an intelligent optimization algorithm, namely a genetic algorithm, is selected as an optimal estimation method of the weight function parameters, and the calculation efficiency and the convergence performance of the algorithm are further improved by improving the genetic algorithm. And finally, selecting the high-sampling-rate explained variable and the low-sampling-rate explained variable in the MIDAS model through mathematical analysis and expert knowledge, and completely conforming to theoretical analysis and actual conditions.
The expression of the MIDAS-ADL model is as follows:
where Y is an explained variable, X is an explanation variable, α is a constant term, β is an overall multiplier of the explanation variable, γjTo be a delay factor, mutAs a random disturbance term, ωiAnd (theta) is a weight function, p and q are maximum hysteresis orders of X, Y, and m is a multiple of a high sampling rate and a low sampling rate.
Design of weight function in MIDAS model
The residence time distribution form of the MIDAS model weight function is inspired from chemical reaction engineering: in a stable continuous flow reactor, reaction materials simultaneously entering at a certain moment flow out of the reactor in sequence after different residence times to form residence time distribution, so that a brand-new MIDAS model weight function is provided.
I. Non-ideal flow in a reactor and residence time distribution produced thereby
The flow and mixing of the materials in the reactor varies from case to case. For the plug flow reactor, the residence time of all material particles is the same; for a complete mixing flow reactor, the concentration of the materials at each position in the reactor is equal and equal to the concentration of the reactants at the outlet, and the particle materials have certain residence time distribution in the reactor. In industrial processes, chemical reactions are directly related to residence time and material concentration, and therefore, the reaction results are directly influenced by flow conditions. In general, non-ideal flow is caused by two factors: the first is the movement of the reaction material particles in the reactor in different directions from the main body flow caused by the movement of stirring, molecular diffusion and the like, and the second is the inconsistent flow speed in each part of the reactor. The residence time distribution of the reaction materials is a random process, so that the residence time distribution of the materials in the flow system can be quantitatively described from the aspect of probability statistics.
The plug flow and the mixed flow are two extreme ideal models, the multi-stage mixed model formed by the series connection of the mixed flow can simulate various flow models, in fact, the series connection of an unlimited plurality of mixed flow is the plug flow model, and the material particle residence time distribution density function E (t) is deduced through the multi-stage mixed model of the mixed flow.
For N reaction kettles connected in series, the weight of each reaction kettle for the material B is calculated as follows:
c0as initial reactant concentration, c1、…、cNThe concentrations of the reactants, τ, from the 1 st to the N th reaction vessels, respectivelysThe average residence time in a single reaction kettle.
The initial boundary conditions for equation set (2) are:
t＝0,c0(0)＝1,c1(0)＝c2(0)＝L＝cN(0)＝0 (3)
solving equation set (2) yields:
further, the method can be obtained as follows:
where τ is N τsExpressing the overall average residence time in the system, wherein E (t) is a residence time distribution density function, F (t) is a residence time distribution density function, and the formula (5) is converted into a dimensionless expression, and the residence time distribution density function is:
II. Design MIDAS weight function according to residence time distribution density function
The formula (6) is a residence time distribution expression of a multi-stage mixed model, can be used for representing the change of the charging material in the blast furnace in the descending process, and can be used as a weight function of the MIDAS model through proper transformation. By means of a Gamma function:
the following properties are known from the function:
Γ(x+1)＝xΓ(x) (8)
therefore, the Gamma function has the capability of generalizing the mathematical concept from the integer domain to the real domain, and formula (8) is substituted into formula (6) to obtain:
by using
Instead of theta, i and imaxIs the current lag order and the maximum lag order of the weighting function and i is 0,1, K, imaxThe available weight function expression is:
second, optimal parameter estimation method based on improved genetic algorithm
The MIDAS constructs a model system which comprises a total mean function and a weight function, and the identification difficulty of weight parameters is increased due to imbalance and nonlinearity of data caused by multi-sampling rate. In the past, the selection of the optimal weight parameter and the optimal high-frequency interpretation variable lag order is determined by means of a least square method and maximum likelihood estimation, but the two methods face the problems of complex calculation and the possibility of being trapped in a local extreme value, the problems can be solved to a certain extent by a novel intelligent heuristic optimization algorithm, and the parameters alpha, beta, theta, gamma and mu can be estimated simultaneously. The genetic algorithm summarized from the theory of evolution, natural selection, has excellent global search capability, and can combine efficiency and accuracy through appropriate improvement. Six major factors for genetic algorithm implementation: the method comprises the steps of parameter coding, initial population setting, fitness function design, genetic operation, algorithm control parameter setting and constraint conditions. Based on the method, a group scale self-adaptive genetic algorithm (SAPGA) is designed, and the purpose of changing the population scale in real time according to the population performance in the evolution process is achieved by calculating the population individual fitness function distribution in the iteration process and determining the increase and decrease of the population scale according to the skewness coefficient (11).
When the skewness coefficient SK is increased, it shows that the number of individuals with higher fitness tends to increase, so the number of individuals should be increased, and the newly added individuals are randomly generated, so as to increase the diversity of population genes and improve the possibility of searching the global optimal solution. When the skewness coefficient SK is reduced, it shows that the number of individuals with smaller fitness tends to increase, the population number should be reduced, especially the individuals with smaller fitness are removed, so the probability that the individuals with smaller fitness are removed is increased, and the method for realizing the method is a roulette method. In the betting round selection rule, the smaller fitness individuals are selected to have smaller probability of being inherited to the next generation, so that the probability of removing the small fitness individuals can be increased by reducing the size of the next generation population. In summary, the population size is enlarged to enrich gene diversity when skewness increases, and the population size is reduced to save computation time when skewness decreases, and the relationship between the population size and skewness coefficient is shown in (12):
SK is the skewness coefficient, ZnFor population scale, k is the population variation value.
The relationship between population size and fitness function skewness variation is shown in fig. 1. SK >0, with a few variable values being large; SK <0, with a few values of variables small. The process of changing from positive bias state to negative bias state, i.e. the process of decreasing the number of large units and increasing the number of small units. It is desirable to control the population fitness function distribution in the genetic process to be normal distribution so as to have optimal global search capability.
Selection method of input and output of prediction model
The blast furnace is divided into five parts, namely a furnace throat, a furnace body, a furnace waist, a furnace belly and a furnace hearth from top to bottom, and different parts in the furnace can undergo different changes in the descending process of coke, ore and flux until the bottom of the furnace hearth is reached and the coke, the ore and the flux are completely converted into molten iron and slag. Through communication with blast furnace operators, 5 indexes of permeability index, furnace top temperature, gas utilization rate, furnace belly gas index and theoretical combustion temperature are preliminarily determined as explanatory variables, the explanatory variables comprise high sampling rate variables influencing the reaction conditions in the whole furnace and upper, middle and lower parts, and the explained variables are production indexes such as molten iron yield, coke ratio, molten iron quality and the like. The expert experience only proposes the most probable factor affecting the explained variable from the actual production process, and the actual situation is further proved by detailed mathematics. Therefore, we verified the explanatory variables proposed by the blast furnace operator by means of a correlation test, the steps of which are as follows:
a. simplifying the interpretation variables by using a principal component analysis method;
b. the sampling rate of the explained variable and the explained variable is kept consistent by utilizing an interpolation method;
c. calculating the correlation between the explained variable and the explained variable order through a Spearman order correlation coefficient;
d. and comprehensively considering the expert experience and the correlation analysis result to determine the input and output variables of the model.
Fourthly, substituting industrial actual data for verification
The sampling interval of 5 variables of permeability index, gas utilization rate, furnace bosh gas index, theoretical combustion temperature and furnace top temperature is 1 hour, and the sampling time interval of yield and coke ratio is 1 day, so that the sampling frequency of the explained variable is 24 times that of the explained variable. We take the volume of a certain ironworks as 2650m3The data of day 2# blast furnace 477 was used as a sample, which contained 477 set of interpreted variables and 11448 set of interpreted variables. And (3) after the data are preprocessed, predicting the production index through the established improved MIDAS-ADL model, wherein the prediction result of the molten iron yield is shown in a figure 2, and the prediction result of the energy consumption index coke ratio is shown in a figure 3.
The prediction result shows that the model effect is good, the value of the explained variable can be accurately predicted, and the information contained in the high-sampling-rate explained variable can be fully mined.
Claims (3)
1. A blast furnace production index prediction method based on an improved multiple sampling rate autoregressive distribution lag model is characterized by comprising the following steps: predicting low-frequency explained variables through high-frequency explained variables in a multi-sampling rate model, proposing residence time distribution as a weight function, simultaneously utilizing an adaptive genetic algorithm for parameter estimation, finally selecting the explained variables, and determining the input and output variables of the model according to correlation analysis; the explained variables comprise air permeability index, furnace top temperature, gas utilization rate, furnace belly gas index and theoretical combustion temperature, and the explained variables comprise molten iron yield and energy consumption coke ratio;
the expression of the multi-sampling rate model is as follows:
y is an explained variable, X is an explanation variable, alpha is a constant term, beta is an overall multiplier of the explanation variable, gammajTo be a delay factor, mutAs a random disturbance term, ωi(theta) is a weight function, p and q are maximum hysteresis orders of X, Y, and m is a multiple of a high sampling rate to a low sampling rate;
the expression of the weight function is as follows:
2. The method of claim 1, wherein: the adaptive genetic algorithm is used for parameter estimation in a multi-sampling rate model:
the purpose of changing the population scale in real time according to the population performance in the evolution is achieved by calculating the population individual fitness function distribution in the iterative process and determining the increase and decrease of the population scale according to the skewness coefficient formula (11);
the relationship between the population size and the skewness coefficient is shown as the formula (12):
where SK is the skewness coefficient, ZnFor population scale, k is the population variation value.
3. The method of claim 1, wherein: the steps of determining the model input and output variables according to the correlation analysis are as follows:
a. the principal component analysis method is utilized to simplify the explanation variables,
b. interpolation is used to keep the sample rate of the interpreted variable and the interpreted variable consistent,
c. the correlation between the explained variable and the explained variable order is calculated by a Spearman order correlation coefficient,
d. and comprehensively considering the expert experience and the correlation analysis result to determine the input and output variables of the model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810945553.2A CN109359320B (en) | 2018-08-20 | 2018-08-20 | Blast furnace index prediction method based on multiple sampling rate autoregressive distribution hysteresis model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810945553.2A CN109359320B (en) | 2018-08-20 | 2018-08-20 | Blast furnace index prediction method based on multiple sampling rate autoregressive distribution hysteresis model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109359320A CN109359320A (en) | 2019-02-19 |
| CN109359320B true CN109359320B (en) | 2022-07-05 |
Family
ID=65350093
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810945553.2A Active CN109359320B (en) | 2018-08-20 | 2018-08-20 | Blast furnace index prediction method based on multiple sampling rate autoregressive distribution hysteresis model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109359320B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110400009A (en) * | 2019-07-09 | 2019-11-01 | 浙江大学 | Blast furnace ironmaking Multiobjective Intelligent optimization method based on self-adapted genetic algorithm |
| CN112098263B (en) * | 2020-09-14 | 2022-07-01 | 山西亚鑫新能科技有限公司 | Method for parameter comprehensive prediction of coke thermal strength model |
| CN113234874B (en) * | 2021-05-14 | 2022-01-11 | 浙江大学 | Blast furnace fault monitoring and self-healing control system based on dynamic Bayesian network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104899425A (en) * | 2015-05-07 | 2015-09-09 | 浙江大学 | Variable selection and forecast method of silicon content in molten iron of blast furnace |
| CN105020705A (en) * | 2015-03-04 | 2015-11-04 | 内蒙古瑞特优化科技股份有限公司 | Method and system for optimizing and controlling combustion performance of circulating fluidized bed boiler in real time |
| CN105886680A (en) * | 2016-05-11 | 2016-08-24 | 东北大学 | System and method for dynamical soft measurement of molten iron silicon content in blast furnace ironmaking process |
| CN107299170A (en) * | 2017-08-10 | 2017-10-27 | 东北大学 | A kind of blast-melted quality robust flexible measurement method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018049290A1 (en) * | 2016-09-09 | 2018-03-15 | Cibo Technologies, Inc. | Systems for determining agronomic outputs for a farmable region, and related methods and apparatus |
-
2018
- 2018-08-20 CN CN201810945553.2A patent/CN109359320B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105020705A (en) * | 2015-03-04 | 2015-11-04 | 内蒙古瑞特优化科技股份有限公司 | Method and system for optimizing and controlling combustion performance of circulating fluidized bed boiler in real time |
| CN104899425A (en) * | 2015-05-07 | 2015-09-09 | 浙江大学 | Variable selection and forecast method of silicon content in molten iron of blast furnace |
| CN105886680A (en) * | 2016-05-11 | 2016-08-24 | 东北大学 | System and method for dynamical soft measurement of molten iron silicon content in blast furnace ironmaking process |
| CN107299170A (en) * | 2017-08-10 | 2017-10-27 | 东北大学 | A kind of blast-melted quality robust flexible measurement method |
Non-Patent Citations (3)
| Title |
|---|
| Causality Diagram-based Scheduling Approach for Blast Furnace Gas System;Feng Jin等;《IEEE/CAA Journal of Automatica Sinica》;20180315(第02期);全文 * |
| 基于Fisher判别分析和核回归的质量监控和估计(英文);张曦等;《Chinese Journal of Chemical Engineering》;20090615(第03期);全文 * |
| 基于分类回归树的高炉铁水硅含量预测模型;于涛等;《内蒙古大学学报(自然科学版)》;20150915(第05期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109359320A (en) | 2019-02-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhou et al. | Hybrid-model-based intelligent optimization of ironmaking process | |
| CN109359320B (en) | Blast furnace index prediction method based on multiple sampling rate autoregressive distribution hysteresis model | |
| Hu et al. | A multilevel prediction model of carbon efficiency based on the differential evolution algorithm for the iron ore sintering process | |
| CN106096788B (en) | Converter steelmaking process cost control method and system based on PSO _ ELM neural network | |
| CN104651559B (en) | Blast furnace liquid iron quality online forecasting system and method based on multivariable online sequential extreme learning machine | |
| CN102540879A (en) | Multi-target evaluation optimization method based on group decision making retrieval strategy | |
| CN111260157B (en) | Smelting ingredient optimization method based on ecological niche optimization genetic algorithm | |
| CN102031319A (en) | Method for forecasting silicon content in blast-furnace hot metal | |
| Xu et al. | Reducing the fluctuation of oxygen demand in a steel plant through optimal production scheduling | |
| CN115522012B (en) | Method for controlling TSC stage oxygen blowing amount of large converter | |
| CN115034465B (en) | Coke quality prediction method based on artificial intelligence | |
| CN115620846A (en) | Preparation and control method of active manganese material | |
| CN103778466A (en) | Converter steel-making temperature modeling forecasting method based on vector error and system thereof | |
| Li et al. | Burden surface decision using MODE with TOPSIS in blast furnace ironmkaing | |
| Tang et al. | An estimation of distribution algorithm with resampling and local improvement for an operation optimization problem in steelmaking process | |
| Ji et al. | Application of the improved the ELM algorithm for prediction of blast furnace gas utilization rate | |
| CN111047080A (en) | Multi-objective decision optimization method based on satisfying preference | |
| CN110458348B (en) | Ingredient optimization method and system based on nonlinear opportunity constraint | |
| CN114420220A (en) | Method for predicting pH value in neutral leaching process of zinc hydrometallurgy | |
| Hu et al. | Dynamic Modeling Framework Based on Automatic Identification of Operating Conditions for Sintering Carbon Consumption Prediction | |
| CN103675012A (en) | Industrial melt index soft measurement instrument and method based on BP particle swarm optimization | |
| Liu et al. | End point prediction of basic oxygen furnace (BOF) steelmaking based on improved bat-neural network | |
| CN113637819B (en) | Blast furnace material distribution method and system based on deep reinforcement learning | |
| Quan et al. | Prediction of air leakage rate of sintering furnace based on BP neural network optimized by PSO | |
| CN116127345B (en) | Converter steelmaking process mode design method based on deep clustering generation countermeasure network |
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 |