CN112183872A - Blast furnace gas generation amount prediction method combining generation of countermeasure network and neural network - Google Patents
Blast furnace gas generation amount prediction method combining generation of countermeasure network and neural network Download PDFInfo
- Publication number
- CN112183872A CN112183872A CN202011079997.6A CN202011079997A CN112183872A CN 112183872 A CN112183872 A CN 112183872A CN 202011079997 A CN202011079997 A CN 202011079997A CN 112183872 A CN112183872 A CN 112183872A
- Authority
- CN
- China
- Prior art keywords
- network
- blast furnace
- data
- furnace gas
- gas generation
- 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
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 67
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000012549 training Methods 0.000 claims abstract description 37
- 238000004088 simulation Methods 0.000 claims abstract description 9
- 238000007781 pre-processing Methods 0.000 claims abstract description 7
- 239000007789 gas Substances 0.000 claims description 89
- 238000010606 normalization Methods 0.000 claims description 10
- 238000009826 distribution Methods 0.000 claims description 9
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 claims description 6
- 239000001301 oxygen Substances 0.000 claims description 6
- 229910052760 oxygen Inorganic materials 0.000 claims description 6
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 238000004140 cleaning Methods 0.000 claims description 3
- 238000004519 manufacturing process Methods 0.000 abstract description 22
- 229910000831 Steel Inorganic materials 0.000 abstract description 17
- 239000010959 steel Substances 0.000 abstract description 17
- 238000013473 artificial intelligence Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 13
- 238000010586 diagram Methods 0.000 description 10
- 230000000694 effects Effects 0.000 description 9
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical compound [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 description 8
- 230000008569 process Effects 0.000 description 7
- 229910052742 iron Inorganic materials 0.000 description 4
- 238000003912 environmental pollution Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 239000002699 waste material Substances 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 230000003139 buffering effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000009628 steelmaking Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 239000006227 byproduct Substances 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000004134 energy conservation Methods 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000010355 oscillation Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 239000002994 raw material Substances 0.000 description 1
- 230000009467 reduction Effects 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"
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/04—Manufacturing
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Marketing (AREA)
- Health & Medical Sciences (AREA)
- Tourism & Hospitality (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Primary Health Care (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Manufacturing & Machinery (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Evolutionary Biology (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses a blast furnace gas generation amount prediction method combining generation of a confrontation network and a neural network, which relates to the technical field of artificial intelligence and comprises the following steps: acquiring real sample data for predicting the gas generation amount of the blast furnace; preprocessing the real sample data; constructing and training to generate a confrontation network; generating simulation data for blast furnace gas generation amount prediction based on the real sample data and the generated countermeasure network; constructing a blast furnace gas generation amount prediction model based on a BP neural network, and training the blast furnace gas generation amount prediction model by using simulation data and real sample data; and predicting the blast furnace gas generation amount based on the trained blast furnace gas generation amount prediction model to obtain a blast furnace gas generation amount prediction result. The blast furnace gas generation amount prediction method based on the BP neural network combines the generation countermeasure network and the BP neural network to predict the blast furnace gas generation amount, remarkably improves the prediction accuracy, and effectively solves the problem that the BP neural network has lower prediction accuracy if the measured data is too little in the steel production process.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a blast furnace gas generation amount prediction method combining generation of a countermeasure network and a neural network.
Background
With the development of economy in China, people live better and better, and the problems of environmental pollution, low energy utilization rate and the like in China become more serious, wherein the problem is the most serious for a large energy consumption family, particularly iron and steel enterprises.
Blast furnace gas is a main secondary energy source in the steel production process, and is characterized by lower heat value and unstable self property, so that the production process of the blast furnace gas is more complicated, the output has larger fluctuation and is difficult to control. Unlike other by-products of the steel production process, the blast furnace gas often remains in large quantities after the normal production process is completed. In addition, too much or too little blast furnace gas in the production process can cause the fire of production equipment to be stopped, the steel production process to be interrupted, and potential safety hazards and raw material waste are caused. Therefore, if the amount of blast furnace gas generated cannot be accurately predicted, not only is there a safety problem in the production process, but also energy waste and environmental pollution are caused, and economic losses are caused.
Currently, most of steel and iron enterprises in China need to monitor the gas generation amount and the gas consumption amount of each gas user in a gas transportation pipeline in real time in order to enable each production link to operate efficiently and optimize each numerical value and index in a gas system in the actual production process. However, the gas system is prone to generate various complicated and unstable conditions due to the influence of production conditions, so that the gas system has variability, and the system is prone to have a large-amplitude oscillation phenomenon, so that the system is unstable. In addition, because the gas system itself is very large, when the working condition changes, the gas cannot be scheduled according to different situations in real time, and the buffering capacity of the buffering device arranged in the gas system is limited, it is difficult to construct a proper mechanism prediction model according to the gas system itself. Nowadays, blast furnace gas scheduling of most iron and steel enterprises in the steelmaking production process still depends on the experience of a gas scheduling expert, so that the problems of untimely prediction, low prediction precision, low energy utilization rate, large subjective influence of individuals on a scheduling scheme and the like exist, and the efficiency and the safety of the iron and steel production process are seriously influenced.
The existing energy management mode cannot adjust the large fluctuation of the blast furnace gas in real time, has low prediction precision, cannot finally realize energy balance and achieves the purposes of energy conservation and emission reduction.
Disclosure of Invention
Aiming at the problems in blast furnace gas generation amount prediction, the invention provides a modeling prediction method based on combination of a generation countermeasure network and a BP neural network so as to realize accurate blast furnace gas generation amount prediction.
In order to achieve the above object, the following solutions are proposed:
a blast furnace gas generation amount prediction method combining generation of a countermeasure network and a neural network comprises the following steps:
s101, acquiring real sample data for predicting the blast furnace gas generation amount; preprocessing the real sample data;
s102, constructing and training a countermeasure network based on the preprocessed real sample data;
s103, generating simulation data for blast furnace gas generation amount prediction based on the real sample data and the generation countermeasure network;
s104, constructing a blast furnace gas generation amount prediction model based on a BP neural network, and training the blast furnace gas generation amount prediction model by using the simulation data and the real sample data;
and S105, predicting the blast furnace gas generation amount based on the trained blast furnace gas generation amount prediction model to obtain a blast furnace gas generation amount prediction result.
Further, the pre-processing comprises:
determining input and output of a blast furnace gas generation amount prediction model; the input includes: cold air flow, hot air pressure, oxygen enrichment and blast furnace gas generation amount in the previous time period; the output includes: a predicted value of blast furnace gas generation amount;
cleaning and normalizing data;
rejecting abnormal data in the real sample data;
normalizing the real sample data by adopting the following function, and normalizing the real sample data into a value on a [0,1] interval, wherein the normalization formula of the function is as follows:
wherein x is a value obtained by data normalization, xjAs initial value before data normalization, xminIs the minimum value of the initial data, xmaxIs the maximum value of the initial data, yminIs the left end point of the normalized interval, ymaxThe right end of the normalized interval.
Further, the generating a countermeasure network includes: generating a network and judging the network; the generation network and the discrimination network are respectively constructed based on a BP neural network.
Further, the training to generate the countermeasure network includes:
training a discrimination network by adopting an alternate iterative training method, taking the real sample data as input quantity, and training the discrimination network to generate the network when the output is close to 1, wherein the input quantity of the generated network is a group of randomly generated data; and when the network is generated by training, keeping the parameters of the trained discrimination network unchanged, and iteratively updating the parameters of the generated network.
Further, the loss function of the discriminant network is:
the loss function of the generated network is defined as:
wherein z is the noise data input to the generator; x is real sample data input into the discriminator; d is a discrimination network; g is a generation network; z to pz(z) shows the noise data obedience distribution pz(z)，
For the expectation of noisy data, x-pdata(x) Representing true data obedience distribution pdata(x)，
Is the expectation of real data.
Further, the establishing of the blast furnace gas generation amount prediction model based on the BP neural network comprises:
determining hidden layer nodes of the BP neural network; the number of nodes of the hidden layer is determined by the following formula:
wherein m is the number of hidden layer nodes; n is the number of nodes of the input layer; l is the number of output layer nodes.
According to the technical scheme, in the blast furnace gas generation amount prediction method combining the generation countermeasure network and the neural network, the BP neural network is adopted to construct the prediction model for prediction, and the generation countermeasure network is constructed to increase sample data; the accuracy of the prediction is significantly improved and reduced in both mean absolute error and mean error rate. The problem that the BP neural network prediction precision is low if the measured data is too little in the steel production process is effectively solved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a schematic flow chart of a blast furnace gas generation prediction method combining generation of a countermeasure network and a neural network according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a generation countermeasure network;
fig. 3 is a schematic diagram of an actual structure of a generation network disclosed in the embodiment of the present invention;
FIG. 4 is a schematic structural diagram of a BP neural network according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of a model for predicting the gas generation of a blast furnace combining a generation countermeasure network and a neural network according to an embodiment of the present invention;
FIG. 6 shows the 5-minute prediction results of the BP neural network disclosed in the embodiment of the present invention;
FIG. 7 is a diagram illustrating the 8-minute prediction results of the BP neural network disclosed in the embodiments of the present invention;
FIG. 8 is a 15-minute prediction of a BP neural network as disclosed in embodiments of the present invention;
FIG. 9 is a 5-minute prediction curve based on a prediction model established by generating a countermeasure network and a BP neural network, disclosed in an embodiment of the present invention;
FIG. 10 is a 8-minute prediction curve based on a prediction model established by generating a countermeasure network and a BP neural network, disclosed by an embodiment of the present invention;
FIG. 11 is a 15-minute prediction curve based on a prediction model established by the generation of the countermeasure network and the BP neural network, disclosed in the embodiments of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention takes the steel industry as the background, aims at the problems of serious energy waste, large harmful gas emission amount, environmental pollution and the like of the current steel industry in China, combines the generation of a confrontation network and a BP neural network to construct a prediction model, predicts the blast furnace gas which is important energy in the steelmaking process flow based on real field data obtained by sampling in the actual production process, and realizes the accurate prediction of the blast furnace gas generation amount
Referring to fig. 1, a flow diagram of a blast furnace gas generation prediction method for generating a countermeasure network and a neural network in combination according to an embodiment of the present invention is shown, the method including the following steps:
s101, acquiring real sample data for blast furnace gas generation amount; preprocessing the real sample data;
wherein, the data preprocessing comprises:
a) determining input and output of a blast furnace gas generation amount prediction model;
inputting: the blast furnace gas generation process is complex and comprises a plurality of physical and chemical reactions, and a plurality of factors in the steel production process influence the blast furnace gas generation amount. By analyzing the production process of the steel plant, based on the mechanism analysis of blast furnace gas generation, the three variables of cold air flow, hot air pressure and oxygen enrichment have great influence on the blast furnace gas generation amount, so that the three variables are used as input variables of a prediction model.
Since the blast furnace gas generation amount is time series data and has a certain periodicity with time, the historical data affects future data, and therefore the blast furnace gas generation amount in the previous time period is also used as one input variable of the prediction model.
And (3) outputting: the predicted value of the blast furnace gas generation amount is used as the output of the prediction model.
b) Cleaning and normalizing data;
due to the complex conditions of the steel production field, data acquired from the field are easily interfered, sample data needs to be preprocessed before a blast furnace gas generation prediction model is constructed, and abnormal data are removed.
Because the blast furnace gas system is extremely large, the magnitude of data collected on site usually reaches more than ten thousand, and the convergence rate of the neural network is influenced if data normalization is not carried out, so that the prediction effect of the prediction model is reduced. And, the magnitude of each sample data used to train the neural network varies. Therefore, in order to ensure that all sample data are equal in status, the cleaned data is normalized. Normalizing all data by adopting the following function, and normalizing the data into a value on a [0,1] interval, wherein the normalization formula of the function is as follows:
wherein x is a value obtained by data normalization, xjAs initial value before data normalization, xminIs the minimum value of the initial data, xmaxIs the maximum value of the initial data, yminIs the left end point of the normalized interval, ymaxThe right end of the normalized interval.
S102, constructing and training a countermeasure network based on the preprocessed real sample data;
the basis of supervised learning is to have a large amount of training data as samples, but the number of training data is often small in the actual production process for some irresistible reasons.
The generative confrontation network GAN is a generative model constructed by the confrontation process. The main idea of generating the confrontation model is the idea of zero sum game in the game theory, which means that the sum of the benefits of both game parties is zero, that is, one party gets the result, and the other party is lost.
The structure of the generation countermeasure network is composed of a generator (G) and a discriminator (D), and the countermeasure in the generation countermeasure network is the countermeasure of the generator and the discriminator. The generator is essentially a sample generator of the blast furnace gas influence factors and is responsible for searching the potential regular distribution in the real blast furnace gas influence factor sample, so that the input noise is packaged into a sample similar to the real sample, namely the packaged false sample is output. The discriminator is essentially a two-classifier, preferably a 0-1 classifier, and is responsible for determining whether the input sample is a true sample or a false sample, the closer the input sample is to the true sample, the closer the output of the discriminator is to 1, and conversely, the closer the output is to 0.
Generating a structure diagram of the countermeasure network as shown in fig. 2, the input to the discriminator includes real sample data and wrapped fake sample data generated by the generator. The purpose of the discrimination model is to distinguish the authenticity of the input sample as much as possible. The input of the generated model is only random noise, and the purpose of the generated model is to pack the noise into false sample data which is close to real sample data to cheat the discriminant model. When the output of the discriminator is close to 1, i.e. the discriminator considers that the input samples are mostly true samples, it can be considered that the generator is windward at this time, and the generated false samples cheat the discriminator. Conversely, when the output of the discriminator is close to 0, i.e., the discriminator considers that the input samples are mostly false samples, the discriminator is windward at this time, and determines false samples generated by the generator. In order to obtain the final win, the discriminator and the generator need to continuously optimize themselves, improve the capability of generating the false samples and the capability of discriminating the false samples, the self-optimization process is a process of mutually fighting each other, finally, the two networks reach a dynamic balance state, and the output of the discriminator is a number close to 0.5.
S103, generating simulation data for blast furnace gas generation amount prediction based on the real sample data and the constructed generation countermeasure network;
through the game between the generator and the discriminator, a good sample authenticity discriminator or sample generator can be obtained. The aim of generating the countermeasure network is to increase the capacity of the sample data to improve the prediction effect of the BP neural network, so that the generator is expected to win in a game and can generate a false sample which is similar to the real sample data.
Whether the generator (G) or the discriminator (D) can be constructed essentially by means of a neural network. Wherein the loss function of the discrimination network is defined as:
the loss function of the generated network is defined as:
wherein z is the noise data input to the generator; x is real sample data input into the discriminator; d is a discrimination network; g is a generation network; z to pz(z) shows the noise data obedience distribution pz(z)，
For the expectation of noisy data, x-pdata(x) Representing true data obedience distribution pdata(x)，
Is the expectation of real data.
This formula is essentially a maximum and minimum optimization problem, with the arbiter and generator being optimized separately. The discriminator D is optimized first, and the first term of the loss function of the discrimination network is to make the result larger and better when the true sample x is input, that is, the result is closer to 1 and better. For the false sample z, it is desirable to obtain as small a result as possible, i.e., 1-D (G (z)) as large a result as possible. This allows the discriminator output to distinguish between true and false samples.
For the optimization of the generator G, since the input of the generating network is only a noise sample z, the loss function of the generating network has only one term compared to the discriminant network, and the generating network naturally expects that the generated false sample can be as close to 1 as possible, but for convenience in programming, 1-D (G (z)) is uniformly written.
Because the generation network and the discrimination network are two neural networks, if the two networks are trained simultaneously, the training effect is not ideal, and the training period is too long. In the embodiment of the invention, an alternate iterative training method is adopted, firstly, a discrimination network is trained, real sample data is used as input quantity, and when the output is close to 1, the discrimination network can well judge the regular distribution of input data and can discriminate the truth of the input data. The second step is to train the generation network, the input quantity of the generation network is a set of randomly generated data, in order to check the effect of the false sample generated by the generation network, the discrimination network trained before is connected in series after the generation network, and the actual structure diagram of the generation network is shown in fig. 3.
The generator inputs the processed and packaged false samples into the discriminator, and the output error of the discriminator is used for updating the network parameters. However, when the parameters are updated, the parameters of the discriminant network should be kept unchanged, only the parameters of the generator are updated, because the purpose of constructing the network is to construct a suitable generator to generate sample data, so as to increase the sample capacity, the parameters of the discriminant network after being trained should be kept unchanged, the parameters of the generated network should be continuously updated for a certain number of iterations, when the final output of the whole network is close to 1, which means that the false sample generated by the generator can already cheat the discriminant device and is considered as a real sample, and the data generated by the generator at this time can be used as a real sample.
S104, establishing a blast furnace gas prediction model by combining the generated countermeasure network and the BP neural network; the method comprises the following steps: and constructing a blast furnace gas generation amount prediction model based on the BP neural network, and training the blast furnace gas generation amount prediction model by using simulation data and real sample data. More specifically:
a) determining hidden layer nodes of the BP neural network;
the BP neural network is composed of three fully-connected neural networks, namely an input layer, a hidden layer and an output layer. The number of nodes of the hidden layer is determined by the following formula:
wherein m is the number of hidden layer nodes; n is the number of nodes of the input layer; l is the number of output layer nodes; thus, a BP neural network consisting of four input layer nodes, two hidden layer nodes and one output layer node is obtained. The structure diagram of the BP neural network constructed in the embodiment of the present invention is shown in fig. 4.
b) Establishing a blast furnace gas prediction model;
1600 data collected on site in a certain steel mill are taken as an example to establish a prediction model of the BP neural network, and as only 1600 sample data used for training are provided, the training effect is poor due to the undersize sample capacity, a discriminator and a generator are trained by utilizing the generated countermeasure network, and the sample capacity is expanded by utilizing the generator to be used for training the BP neural network.
1200 blast furnace gas generation amount, cold air flow, hot air pressure and oxygen enrichment amount in the last time period for training are selected as real data of the training discrimination network.
Inputting the false data generated by the generation network and the random noise into the discrimination network to check the effect of the false data generated by the generation network. With the progress of the training round, the gradual attenuation of the discriminant network loss function to about 1 can be found, which shows that the discriminant network can better judge the authenticity of the input data at this moment. The loss function of the generated network is reduced, which shows that the generated network can output false data which is similar to the regular distribution of the real data at the moment, and the false data can be used as the input data of the BP neural network for training. The blast furnace gas prediction model based on the generation countermeasure network and the BP neural network can be obtained by connecting the generation countermeasure network and the BP neural network in series, and the structural diagram of the prediction model is shown in fig. 5.
And S105, predicting the blast furnace gas generation amount based on the trained blast furnace gas prediction model to obtain a blast furnace gas generation amount prediction result.
In the blast furnace gas generation amount prediction method combining the generation countermeasure network and the neural network, the method for generating the countermeasure network to increase the sample data can obviously improve the accuracy of prediction and reduce the average absolute error and the average error rate. Therefore, the method for generating the countermeasure network to increase the sample data can effectively solve the problem of low prediction precision of the BP neural network if the measured data is too little in the steel production process.
The blast furnace gas generation amount prediction method of the present invention, which combines the generation of the countermeasure network and the neural network, will be described below with a specific example.
And selecting data measured in a period of time in the production process of a blast furnace of a certain steel plant on the day of 9, 19 and 9 of 2017 for experimental simulation and test. The sampling period is 5s, the data (1200 samples) of the blast furnace gas generation amount, the cold air flow rate, the hot air pressure and the oxygen enrichment amount which are respectively collected for 100min in the previous time period are taken as training samples to respectively predict the blast furnace gas generation amounts of 5min, 8min and 15min, and the test sample data capacities are respectively 60, 96 and 180. The adopted experimental software is Matlab, a BP neural network is established by utilizing a newff function in the Matlab, and the parameters of the established neural network are as follows: the maximum training frequency is 1000 times, the network training target is 0.001, the network learning rate is 0.01, and the prediction results are shown in fig. 6-8, wherein fig. 6 shows the prediction result of the blast furnace gas generation amount of the BP neural network within 5 min; FIG. 7 is the blast furnace gas generation prediction result of BP neural network for 8 min; FIG. 8 is the blast furnace gas generation prediction for 15min for the BP neural network.
For more intuitively checking the prediction effect, the average absolute error and the average error rate of the BP neural network prediction are calculated, and the calculation results are shown in the following table:
TABLE 1
| t/min | MAE | MPE/％ |
| 5 | 4140 | 1.1 |
| 8 | 5523 | 1.6 |
| 15 | 7828 | 2.1 |
Wherein t represents a prediction time; mae (mean Absolute error) means mean Absolute error; mpe (mean percent error) indicates the average error rate. According to the prediction result graph and the table 1, it can be found that when the sample capacity of the sample data used for training is small, the prediction result is not ideal, the generated prediction curve is also rough compared with the real data curve, and the error is large. The analysis may be related to too little sample data for training, and in order to obtain a more ideal prediction effect, a method for increasing the sample capacity is needed.
Keeping the network parameters of the BP neural network the same, namely, the maximum training times is 1000 times, the network training target is 0.001, and the network learning rate is 0.01. Expanding the sample data capacities of the blast furnace gas generation amount, the cold air flow rate, the hot air pressure and the oxygen enrichment amount of the last time period to 10000 by using a generation network, then respectively predicting the blast furnace gas generation amounts 5min, 8min and 15min later, and predicting results as shown in FIGS. 9-11, wherein the blast furnace gas generation amount predicting result of 5min based on a prediction model established by a generation countermeasure network and a BP neural network is shown in FIG. 9; FIG. 10 is an 8min blast furnace gas generation prediction based on a prediction model built to generate a countermeasure network and a BP neural network; FIG. 11 is a 15min blast furnace gas occurrence prediction based on a predictive model built to generate the countermeasure network and the BP neural network.
In order to more intuitively verify the prediction effect, the average absolute error and the average error rate of the prediction are calculated, and the result calculated based on the prediction model established by generating the countermeasure network and the BP neural network is compared with the result calculated by using only the BP neural network, and the calculation results are as follows:
TABLE 2
| t/min | MAE | MPE/％ |
| 5 (BP only) | 4140 | 1.1 |
| 5 | 3034 | 0.8 |
| 8 (BP only) | 5523 | 1.6 |
| 8 | 4097 | 1.2 |
| 15 (BP only) | 7828 | 2.1 |
| 15 | 5242 | 1.4 |
The odd behaviors only use the prediction results of the prediction model established by the BP neural network, and the even behaviors are based on the prediction results of the prediction models established by the generation countermeasure network and the BP neural network. t represents a prediction time; mae (mean Absolute error) means mean Absolute error; mpe (mean percent error) indicates the average error rate.
Finally, it should also be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (6)
1. A blast furnace gas generation amount prediction method combining generation of a countermeasure network and a neural network is characterized by comprising the following steps:
s101, acquiring real sample data for predicting the blast furnace gas generation amount; preprocessing the real sample data;
s102, constructing and training a countermeasure network based on the preprocessed real sample data;
s103, generating simulation data for blast furnace gas generation amount prediction based on the real sample data and the generation countermeasure network;
s104, constructing a blast furnace gas generation amount prediction model based on a BP neural network, and training the blast furnace gas generation amount prediction model by using the simulation data and the real sample data;
and S105, predicting the blast furnace gas generation amount based on the trained blast furnace gas generation amount prediction model to obtain a blast furnace gas generation amount prediction result.
2. The method of claim 1, wherein the pre-processing comprises:
determining input and output of a blast furnace gas generation amount prediction model; the input includes at least: cold air flow, hot air pressure, oxygen enrichment and blast furnace gas generation amount in the previous time period; the output includes: a predicted value of blast furnace gas generation amount;
cleaning and normalizing data;
rejecting abnormal data in the real sample data;
normalizing the real sample data by adopting the following function, and normalizing the real sample data into a value on a [0,1] interval, wherein the normalization formula of the function is as follows:
wherein x is a value obtained by data normalization, xjInitial before normalization for dataValue, xminIs the minimum value of the initial data, xmaxIs the maximum value of the initial data, yminIs the left end point of the normalized interval, ymaxThe right end of the normalized interval.
3. The method of claim 1, wherein generating the countermeasure network comprises: generating a network and judging the network; the generation network and the discrimination network are respectively constructed based on a BP neural network.
4. The method of claim 3, wherein training to generate a countermeasure network comprises:
training a discrimination network by adopting an alternate iterative training method, taking the real sample data as input quantity, and training the discrimination network to generate the network when the output is close to 1, wherein the input quantity of the generated network is a group of randomly generated data; and when the network is generated by training, keeping the parameters of the trained discrimination network unchanged, and iteratively updating the parameters of the generated network.
5. The method of claim 3,
the loss function of the discrimination network is:
the loss function of the generated network is defined as:
wherein z is the noise data input to the generator; x is real sample data input into the discriminator; d is a discrimination network; g is a generation network, z-pz(z) shows the noise data obedience distribution pz(z)，
For the expectation of noisy data, x-pdata(x) Representing true data obedience distribution pdata(x)，
Is the expectation of real data.
6. The method of claim 1, wherein the establishing a blast furnace gas generation prediction model based on a BP neural network comprises:
determining hidden layer nodes of the BP neural network; the number of nodes of the hidden layer is determined by the following formula:
wherein m is the number of hidden layer nodes; n is the number of nodes of the input layer; l is the number of output layer nodes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011079997.6A CN112183872A (en) | 2020-10-10 | 2020-10-10 | Blast furnace gas generation amount prediction method combining generation of countermeasure network and neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011079997.6A CN112183872A (en) | 2020-10-10 | 2020-10-10 | Blast furnace gas generation amount prediction method combining generation of countermeasure network and neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112183872A true CN112183872A (en) | 2021-01-05 |
Family
ID=73948865
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011079997.6A Pending CN112183872A (en) | 2020-10-10 | 2020-10-10 | Blast furnace gas generation amount prediction method combining generation of countermeasure network and neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112183872A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114418177A (en) * | 2021-12-13 | 2022-04-29 | 中国科学院合肥物质科学研究院 | New product material distribution prediction method for digital twin workshop based on generation of countermeasure network |
| CN116127345A (en) * | 2022-12-23 | 2023-05-16 | 北京科技大学 | Converter steelmaking process mode design method based on deep clustering generation countermeasure network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107991876A (en) * | 2017-12-14 | 2018-05-04 | 南京航空航天大学 | Aero-engine condition monitoring data creation method based on production confrontation network |
| CN109376903A (en) * | 2018-09-10 | 2019-02-22 | 浙江工业大学 | A kind of PM2.5 concentration value prediction technique based on game neural network |
| CN111553587A (en) * | 2020-04-26 | 2020-08-18 | 中国电力科学研究院有限公司 | New energy scene generation method and system based on confrontation learning model |
| CN111555368A (en) * | 2020-05-15 | 2020-08-18 | 广西大学 | Deep generation type countermeasure network scheduling and control method of comprehensive energy system |
-
2020
- 2020-10-10 CN CN202011079997.6A patent/CN112183872A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107991876A (en) * | 2017-12-14 | 2018-05-04 | 南京航空航天大学 | Aero-engine condition monitoring data creation method based on production confrontation network |
| CN109376903A (en) * | 2018-09-10 | 2019-02-22 | 浙江工业大学 | A kind of PM2.5 concentration value prediction technique based on game neural network |
| CN111553587A (en) * | 2020-04-26 | 2020-08-18 | 中国电力科学研究院有限公司 | New energy scene generation method and system based on confrontation learning model |
| CN111555368A (en) * | 2020-05-15 | 2020-08-18 | 广西大学 | Deep generation type countermeasure network scheduling and control method of comprehensive energy system |
Non-Patent Citations (2)
| Title |
|---|
| 张克明 等: "基于生成对抗网络的航天异常事件检测方法", 《北京航空航天大学学报》, vol. 45, no. 7, pages 1329 - 1336 * |
| 王军 等: "基于神经网络的高炉煤气消耗预测模型研究", 《科技与企业》, no. 09, pages 108 - 109 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114418177A (en) * | 2021-12-13 | 2022-04-29 | 中国科学院合肥物质科学研究院 | New product material distribution prediction method for digital twin workshop based on generation of countermeasure network |
| CN114418177B (en) * | 2021-12-13 | 2024-03-29 | 中国科学院合肥物质科学研究院 | New product material distribution prediction method based on digital twin workshops for generating countermeasure network |
| CN116127345A (en) * | 2022-12-23 | 2023-05-16 | 北京科技大学 | Converter steelmaking process mode design method based on deep clustering generation countermeasure network |
| CN116127345B (en) * | 2022-12-23 | 2023-11-14 | 北京科技大学 | Converter steelmaking process mode design method based on deep clustering generation countermeasure network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11755976B2 (en) | Method for predicting oxygen load in iron and steel enterprises based on production plan | |
| Li et al. | Forecasting building energy consumption with hybrid genetic algorithm-hierarchical adaptive network-based fuzzy inference system | |
| Moeini et al. | Fuzzy rule-based model for hydropower reservoirs operation | |
| Wang et al. | A hybrid optimization-based recurrent neural network for real-time data prediction | |
| Choudhury et al. | Artificial Neural Network application for predicting in-flight particle characteristics of an atmospheric plasma spray process | |
| Shen et al. | An uncertain single machine scheduling problem with periodic maintenance | |
| Wang et al. | Re-scheduling the unit commitment problem in fuzzy environment | |
| CN112183872A (en) | Blast furnace gas generation amount prediction method combining generation of countermeasure network and neural network | |
| CN106529732A (en) | Carbon emission efficiency prediction method based on neural network and random frontier analysis | |
| CN106845012A (en) | A kind of blast furnace gas system model membership function based on multiple target Density Clustering determines method | |
| CN113094988A (en) | Data-driven slurry circulating pump operation optimization method and system | |
| Nolting et al. | Can energy system modeling benefit from artificial neural networks? Application of two-stage metamodels to reduce computation of security of supply assessments | |
| Uzun Araz et al. | A reactive scheduling approach based on fuzzy inference for hybrid flowshop systems | |
| CN115879369A (en) | Coal mill fault early warning method based on optimized LightGBM algorithm | |
| Xu et al. | A novel online combustion optimization method for boiler combining dynamic modeling, multi-objective optimization and improved case-based reasoning | |
| WO2022143021A1 (en) | Neural network generating method and apparatus, power forecasting method and apparatus, and storage medium | |
| CN115062534A (en) | Method and device for calculating gas supply reliability of natural gas pipeline system | |
| Shen et al. | A Deep Learning-Based Data Management Scheme for Intelligent Control of Wastewater Treatment Processes Under Resource-Constrained IoT Systems | |
| Shojaee et al. | Applying SVSSI sampling scheme to np-chart to decrease the time of detecting shifts using Markov chain approach and Monte Carlo simulation | |
| Zhai et al. | Associative reasoning-based interpretable continuous decision making in industrial production process | |
| Sharma | Fuzzy reliability analysis of repairable industrial systems using soft-computing based hybridized techniques | |
| AU2021100592A4 (en) | Quantum seeded hybrid evolutionary computational process for constrained optimization | |
| Shahraki et al. | A new approach for forecasting enrollments using harmony search algorithm | |
| Ruan et al. | Multi-criteria group decision support with linguistic variables in long-term scenarios for belgian energy policy | |
| CN113723660A (en) | Specific behavior type prediction method and system based on DNN-LSTM fusion model |
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 |