CN113283163A - Construction and application of RVM sinter FeO content soft measurement model - Google Patents
Construction and application of RVM sinter FeO content soft measurement model Download PDFInfo
- Publication number
- CN113283163A CN113283163A CN202110497744.9A CN202110497744A CN113283163A CN 113283163 A CN113283163 A CN 113283163A CN 202110497744 A CN202110497744 A CN 202110497744A CN 113283163 A CN113283163 A CN 113283163A
- Authority
- CN
- China
- Prior art keywords
- train
- data
- model
- variable
- sintering
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000010276 construction Methods 0.000 title claims abstract description 39
- 238000005259 measurement Methods 0.000 title claims abstract description 21
- 238000005245 sintering Methods 0.000 claims abstract description 72
- 238000000034 method Methods 0.000 claims abstract description 68
- 230000008569 process Effects 0.000 claims abstract description 58
- 238000005070 sampling Methods 0.000 claims description 26
- 238000012360 testing method Methods 0.000 claims description 23
- 238000002156 mixing Methods 0.000 claims description 13
- 238000012216 screening Methods 0.000 claims description 11
- 238000012549 training Methods 0.000 claims description 9
- 238000003556 assay Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 5
- 238000001914 filtration Methods 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical compound [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 description 10
- 238000004519 manufacturing process Methods 0.000 description 7
- 229910000831 Steel Inorganic materials 0.000 description 6
- 230000000694 effects Effects 0.000 description 6
- 239000010959 steel Substances 0.000 description 6
- 229910052742 iron Inorganic materials 0.000 description 5
- 239000002994 raw material Substances 0.000 description 5
- 238000012843 least square support vector machine Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 229910052500 inorganic mineral Inorganic materials 0.000 description 2
- 239000011707 mineral Substances 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 238000003723 Smelting Methods 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 239000000571 coke Substances 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 239000003344 environmental pollutant Substances 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000013213 extrapolation Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 239000004615 ingredient Substances 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000006386 neutralization reaction Methods 0.000 description 1
- 238000011056 performance test Methods 0.000 description 1
- 230000035699 permeability Effects 0.000 description 1
- 231100000719 pollutant Toxicity 0.000 description 1
- 239000000843 powder Substances 0.000 description 1
- 238000003908 quality control method Methods 0.000 description 1
- 230000008844 regulatory mechanism Effects 0.000 description 1
- 239000000779 smoke Substances 0.000 description 1
- 230000000087 stabilizing effect Effects 0.000 description 1
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
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/90—Dynamic range modification of images or parts thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/08—Thermal analysis or thermal optimisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30136—Metal
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- Medical Informatics (AREA)
- Artificial Intelligence (AREA)
- Quality & Reliability (AREA)
- Manufacture And Refinement Of Metals (AREA)
Abstract
The invention discloses construction and application of a soft measurement model for the FeO content of RVM sintering ore, wherein the construction of the model is completed by combining sensor data in the sintering process and sintering machine tail section image data, so that the data information is richer; in addition, the data is subjected to characteristic construction, so that the nonlinear and dynamic characteristics of the sintering data have stronger interpretative capacity; and constructing the RVM model on the basis, so that the RVM model has accurate online soft measurement capability on the FeO content of the sintering ore in a complex sintering process. The soft measurement result comparison of the model and the existing model proves that the model provides more accurate and stable modeling capability for the sintering process.
Description
Technical Field
The invention belongs to a soft measurement method of quality indexes in a sintering process, and particularly relates to an RVM (Relevance Vector machine) sintering ore FeO content soft measurement model constructed based on multi-source data and characteristics.
Technical Field
The steel industry is a pillar type industry for national infrastructure and national economic development, and the development level of the steel industry closely reflects the degree of national industrialization. The core of the iron and steel industry production is blast furnace ironmaking, and the sinter is the main raw material for blast furnace smelting, occupies more than 70% of the raw material entering the blast furnace, and provides good air permeability, production efficiency and ironmaking coke ratio for the blast furnace ironmaking process. Therefore, the stability of the sintering process state and the amount of the sintered minerals directly affect the blast furnace condition, the product quality and the production energy consumption.
Most of domestic iron and steel enterprises have more raw material types and wide sources, and the taste, price and pollutant element content of iron ore have obvious difference, so that the iron ore powder raw material for sintering production has regular change and adjustment. In the actual production process, the quality of the sinter is generally subjected to sinter sampling detection and sinter tail fire observation estimation every four hours by field personnel; if the quality of the sintered ore product does not meet the requirements of the process, the operator can provide corresponding adjustment of the process parameters by combining with field experience. However, the simple regulation mechanism can only solve part of abnormal conditions, and has more obvious hysteresis, thereby seriously influencing the production efficiency and the quality control of the sinter. Under the condition, an accurate soft measurement model is established based on raw materials and on-line process parameters in the sintering process, the quality of the sintered ore is effectively estimated, and a reasonable operation guidance suggestion is given, so that the method has very important practical significance for stabilizing the quality of the sintered ore and improving the quality of steel products. However, sintering production is a continuous and stable air draft sintering process, and along with a series of complex physicochemical reactions, the obvious time lag, dynamics, nonlinearity and coupling exist, so that the soft measurement of the quality of the sintered ore becomes a very challenging subject and a difficult problem which troubles the metallurgical industry for a long time.
Disclosure of Invention
The invention provides the construction and application of a soft measurement model for the FeO content of RVM sinter for solving the problem that the sinter quality index is difficult to measure in real time, and the construction of the model is completed by combining sensor data in the sintering process and sinter tail section image data, so that the data information is richer; in addition, the data is subjected to characteristic construction, so that the nonlinear and dynamic characteristics of the sintering data have stronger interpretative capacity; and constructing the RVM model on the basis, so that the RVM model has accurate online soft measurement capability on the FeO content of the sintering ore in a complex sintering process.
The invention is realized by adopting the following technical scheme:
step 1: modeling off line;
step 1.1: calculating the tail time of the sintering machine corresponding to the finished sintered ore and the sampling time of the key process variable by combining the structure of the sintering machine and the running speed of a belt based on the sampling time of the test of the finished sintered ore;
collecting process variable data by attached sensors in the sintering process, and screening according to the sampling time of key process variables to obtain process variable data Xtrain-1(n × m), the number of samples is n, and the number of variables is m;
and arranging a camera at the tail of the sintering machine to continuously shoot the section of the sintered ore, and screening out video pictures of 3 minutes before and after the moment according to the moment of the tail of the sintering machine. Because the sintering machine tail is mainly red, and the red component represents the main information of the machine tail section image, the red component of the image is used as the gray level image of the machine tail section image to complete the calculation of the brightness l. And selecting key frames of the video by adopting the red component of the tail section image and the peak value of the video brightness within 6 minutes, and calculating the average value. The luminance L of each key frame picture can be specifically expressed as:
wherein h and w represent the height and width of a picture, qijRepresenting the corresponding pixel value in the ith row and jth column of the image. Secondly, extracting image details based on the RGB image of the machine tail section by adopting a top-hat filtering method, converting the image details into a gray-scale image, searching the edge of the image details and completing filling, calculating the proportion of the image details in the image area, and calculating the mean value of all key frames to calculate the porosity k of the sintering section. Finally, the brightness and porosity of the tail section image can constitute variable data X about the tail sectiontrain-2(n×2)；
Step 1.2: mixing Xtrain-1And Xtrain-2Input data X combined into modeltrain-oAnd the test of the FeO content of the finished sintered ore is marked as ytrain. First, the outlier is defined as the point that differs from the mean by more than three standard deviations and is replaced with the nearest non-outlier. Secondly, with Xtrain-oAnd ytrainOn the basis of the correlation of (A) to Xtrain-oCompleting variable construction to enhance the model interpretation capability on the data nonlinear characteristics, and specifically comprising the following six ways:
vnew＝ev
vnew＝vx
vnew＝v1+v2
vnew＝v1×v2
where v denotes the original variable, v1And v2Representing original, heterogeneous variables, vnewRepresenting the constructed new variable; in addition, with Xtrain-oAnd finishing variable construction by taking autocorrelation of historical sampling points as a reference to enhance the model interpretation capability of the dynamic characteristics of the data, wherein the specific expression is as follows:
vnew＝[vt vt-1 … vt-x]
in the formula, t represents a sampling time, and x represents the number of historical sampling points. Recording the input data after completing the variable construction as Xtrain。
Step 1.3: after data normalization, construct XtrainAnd ytrainThe mapping relationship between single samples of the RVM model of (3) can be expressed as:
y＝f(xtrain,w)+ε
wherein w ═ w0,w1,w2,…,wn]TIs a weight vector, e to N (0, σ)2) Representing gaussian white noise. f (x) can be expressed as:
where K (x) is a kernel function, Ψ (x)train)＝[1,K(xtrain,x1),K(xtrain,x2),…,K(xtrain，xn)]T. Based on the above conditions, the conditional probability formula for the output value y can be expressed as:
p(y|xtrain)～N(f(xtrain，w)，σ2)
then, assuming that the observed objects y are independent of each other, the likelihood function of the whole sample can be expressed as:
therein, Ψ (X)train)＝[Ψ1(xtrain)，Ψ2(xtrain)，…，Ψn(xtrain)]. On the basis, the prior probability distribution is set for the parameters to avoid the over-fitting problem, and the method is specifically represented as follows:
based on Bayes' theorem, the posterior distribution of weight w can be obtained, and parameter sigma can be obtained by maximization marginal likelihood function2And optimal selection of alpha
And alphaoAnd further, the posterior mean and the covariance of the weight vector w are obtained:
during the parameter training process, most of alphaiThe value tends to infinity, and the corresponding weight value wiThen 0, so the RVM model can be made to have good sparsity.
Step 2: carrying out online soft measurement;
step 2.1: for new finished ore in the sintering process, screening sensor data x in the sintering process corresponding to the current machine tail time according to step 1.1test-1(1 xm) and tail section variable data xtest-2(1×2)；
Step 2.2: processing outliers of the data according to the step 1.2, and obtaining a construction variable x by adopting a variable construction method selected by an offline modeltest。
Step 2.3: x is to betestPerforming normalization according to the hyperparameters in step 1.3
And alphaoCalculating the content of FeO:
wherein,
Ψ(xtest)＝[1，k(xtest，x1)，K(xtrain，x2)，…，K(xtrain，xn)]T。
the invention has the beneficial effects that:
1. the method combines sensor data and image data in the sintering process to establish a soft measurement model, and multi-source data provides richer sintered mineral content information.
2. The characteristic construction of the variables enables the nonlinearity and the dynamic property of the sintering data to be easily explained, and the prediction accuracy of the model can be improved.
3. The sparsity and probability of the RVM model provide greater robustness to modeling complex physicochemical reaction processes.
Drawings
Fig. 1 is a diagram of RVM soft measurement model construction and application based on multi-source data and feature construction.
Fig. 2 is a flow chart of a sintering process.
Fig. 3 is a diagram of image brightness peak selection.
Fig. 4 is a top-hat filter diagram of an RGB image.
Fig. 5 is an image detail grayscale.
FIG. 6 is an image detail fill-in view.
FIG. 7 shows a variable v1，newAnd an autocorrelation of the historical sample points.
FIG. 8 is a graph showing the results of soft measurement of FeO content in two different models.
Detailed Description
The invention will be described in further detail below with reference to the drawings and examples, which are intended to facilitate the understanding of the invention without limiting it in any way. Fig. 1 provides specific construction and application steps of an RVM sintered ore FeO content soft measurement model constructed based on multi-source data and characteristics, and the implementation of the following case is completed based on the specific steps in fig. 1.
(1) Introduction of sintering Process data
The experiment is directed to a sintering machine of 360 square meters of a certain iron and steel group in south China. The sintering variable data is mainly determined by the reverse extrapolation of the collection time of the tested sinter, and the test process is complicated, so that 3 to 4 test samples are collected on average each day. As shown in Table 1, the overall sinter variable process data contained 16 process variables collected by the sensors, 2 process variables derived from the tail section image and 1 FeO assay value variable. Process variable data composition Xtrain-oAnd the FeO assay value variable is denoted as ytrain. FIG. 2 shows the overall sintering process, in which each station is present during the sintering processWith a significant time difference. As can be seen from FIG. 2, the time for the ingredients to reach the primary mixing cylinder is l1/v0min, the average running time of the primary mixing cylinder is 2.8min, the average running time of the secondary mixing cylinder is 3.3min, and the time difference from the primary mixing cylinder to the secondary mixing cylinder is l2/v0min, time difference from secondary mixing cylinder to mud roller funnel is l3/v0min, the time difference between the head and the tail of the sintering machine is l/v1min, the time difference from the tail of the sintering machine to the end of the circular cooler is pi x d/v2min; wherein l1，l2And l3Respectively, the belt length of the sintering batch to the primary mixing cylinder, the belt length between the primary mixing cylinder and the secondary mixing cylinder, the belt length between the secondary mixing cylinder and the mud roller hopper, the length of the sintering machine, the diameter of the circular cooler, and the length of the sintering machine0，v1And v2Respectively representing the belt speed, the sintering trolley speed and the circular cooler machine speed. Therefore, the sampling time corresponding to the other variables can be obtained according to the sampling time t of the finished sintered ore, and the sampling time is shown in table 1.
Table 1 data set variable list
According to the sinter sampling point at a certain time t, the sampling time t-pi x d/v of the corresponding video picture of the sintering machine tail can be obtained2The image brightness of each frame is calculated by using the red component of the image according to the tail video 3 minutes before and after the moment, as shown in fig. 3. By selecting the brightness peak in fig. 3, the location of the key frame can be determined, considering that the image brightness is highest and the smoke is least at the moment the sinter falls from the tail. Specifically, the process of key frame screening in fig. 3 includes: smoothing an image brightness curve, selecting a peak point and screening key frames in a range near the peak point. After the key frame is selected, the brightness of the red components of all the key frames within 6 minutes is calculated, and the average value is obtained to obtain the brightness variable of the key frame.
As shown in fig. 4, 5, and 6, the calculating step of the porosity of the key frame includes: and (3) filtering the RGB image by top caps to obtain image details, converting the image details into a gray level image, filling the edges of the image details, and finally calculating the detail area proportion of all key frames and obtaining the porosity of the sintered section after averaging.
(2) Sintering process variable construction
After the data acquisition is completed, the outliers are defined as points that differ from the mean by more than three standard deviations and are replaced with the nearest non-outliers. Secondly, with Xtrain-oAnd ytrainTaking the correlation of (A) as a reference, and taking Xtrain-oThe variable construction is completed to enhance the model interpretation ability of the data nonlinear characteristics. After the variable construction is completed, the input variables for the model include:
v1，new＝v1×v9
v2,new＝v2×v9
v3,new＝v3
v5，new＝v5
v12，new＝v12+v18
wherein the variable v1，newShows the new variables reconstructed after multiplying the variable numbered 1 in table 1 (sintering burden neutralization ore ratio) by the variable numbered 9 (machine speed); if the new variables are not combined, the variable values before combination are used, for example: v. of3，newAnd v5，new。
Then, the dynamic characteristics of the data are verified on the basis of the combined variables, the number of historical sampling points is selected according to the autocorrelation of the variables to complete the construction of the variables, and the next step is enhancedThe model's ability to interpret data dynamics in a step. As shown in fig. 7, as the time lag increases, the variable v1，newThe autocorrelation with the historical value gradually decreases; in addition, a large standard deviation of the autocorrelation value indicates that the autocorrelation is dispersed and the overall variation is large. Thus, the variable v1,newThe dynamic characteristics of (2) are obvious, and the variable construction is completed by the standard that the autocorrelation of the variable is greater than 0.4:
v1,new＝[v1,new(t) v1,new(t-1) … v1,new(t-2)]
where t denotes the current sampling time, v1,new(t-1)Representing variable v1,newSample value at time t-1, v1,new(t-2)Representing variable v1,newThe sample value at time t-2. The dynamic extension of the rest variables is constructed by taking the dynamic extension as a standard.
Finally, completing variable construction to obtain input data Xtrain。
(3) Calculating hyper-parameters of RVM model
After the variable construction is completed, the data are standardized to construct XtrainAnd ytrainTraining to obtain optimal parameters
And alphaoAnd solving the posterior mean value and the covariance of the weight vector w:
(4) RVM model completion process constructed based on multi-source data and characteristics
Step 1: modeling off line;
step 1.1: calculating the tail time of the sintering machine corresponding to the finished sintered ore and the sampling time of the key process variable by combining the structure of the sintering machine and the running speed of a belt based on the sampling time of the test of the finished sintered ore;
collecting process variable data by attached sensors in the sintering process, and screening according to the sampling time of key process variables to obtain process variable data Xtrain-1(n × m), the number of samples is n, and the number of variables is m;
and arranging a camera at the tail of the sintering machine to continuously shoot the section of the sintered ore, and screening out video pictures of 3 minutes before and after the moment according to the moment of the tail of the sintering machine. Because the sintering machine tail is mainly red, and the red component represents the main information of the machine tail section image, the red component of the image is used as the gray level image of the machine tail section image to complete the calculation of the brightness l. And selecting key frames of the video by adopting the red component of the tail section image and the peak value of the video brightness within 6 minutes, and calculating the average value. The luminance L of each key frame picture can be specifically expressed as:
wherein h and w represent the height and width of a picture, qijRepresenting the corresponding pixel value in the ith row and jth column of the image. Secondly, extracting image details based on the RGB image of the machine tail section by adopting a top-hat filtering method, converting the image details into a gray-scale image, searching the edge of the image details and completing filling, calculating the proportion of the image details in the image area, and calculating the mean value of all key frames to calculate the porosity k of the sintering section. Finally, the brightness and porosity of the tail section image can constitute variable data X about the tail sectiontrain-2(n×2)；
Step 1.2: mixing Xtrain-1And Xtrain-2Input data X combined into modeltrain-oAnd the test of the FeO content of the finished sintered ore is marked as ytrain. First, the outlier is defined as the point that differs from the mean by more than three standard deviations and is replaced with the nearest non-outlier. Secondly, with Xtrain-oAnd ytrainOn the basis of the correlation of (A) to Xtrain-oCompleting variable construction to enhance model to dataThe interpretation capability of the nonlinear characteristics specifically comprises the following six ways:
vnew＝ev
vnew＝vx
vnew＝v1+v2
vnew＝v1×v2
where v denotes the original variable, v1And v2Representing original, heterogeneous variables, vnewRepresenting the constructed new variable; in addition, with Xtrain-oAnd finishing variable construction by taking autocorrelation of historical sampling points as a reference to enhance the model interpretation capability of the dynamic characteristics of the data, wherein the specific expression is as follows:
vnew＝[vt vt-1 … vt-x]
in the formula, t represents a sampling time, and x represents the number of historical sampling points. Recording the input data after completing the variable construction as Xtrain。
Step 1.3: after data normalization, construct XtrainAnd ytrainThe mapping relationship between single samples of the RVM model of (3) can be expressed as:
y＝f(xtrain,w)+ε
wherein w ═ w0,w1,w2,…,wn]TIs a weight vector, e to N (0, σ)2) Representing gaussian white noise. f (x) can be expressed as:
wherein K (x) is a nuclear componentNumber, Ψ (x)train)＝[1,K(xtrain,x1),K(xtrain,x2),…,K(xtrain,xn)]T. Based on the above conditions, the conditional probability formula for the output value y can be expressed as:
p(y|xtrain)～N(f(xtrain，w)，σ2)
then, assuming that the observed objects y are independent of each other, the likelihood function of the whole sample can be expressed as:
therein, Ψ (X)train)＝[Ψ1(xtrain),Ψ2(xtrain),…,Ψn(xtrain)]. On the basis, the prior probability distribution is set for the parameters to avoid the over-fitting problem, and the method is specifically represented as follows:
based on Bayes' theorem, the posterior distribution of weight w can be obtained, and parameter sigma can be obtained by maximization marginal likelihood function2And optimal selection of alpha
And alphaoAnd further, the posterior mean and the covariance of the weight vector w are obtained:
during the parameter training process, most of alphaiThe value tends to infinity, and the corresponding weight value wiThen it is 0, so that the RVM model can be madeGood sparsity is obtained.
Step 2: carrying out online soft measurement;
step 2.1: for new finished ore in the sintering process, screening sensor data x in the sintering process corresponding to the current machine tail time according to step 1.1test-1(1 xm) and tail section variable data xtest-2(1×2)；
Step 2.2: processing outliers of the data according to the step 1.2, and obtaining a construction variable x by adopting a variable construction method selected by an offline modeltest。
Step 2.3: x is to betestPerforming normalization according to the hyperparameters in step 1.3
And alphaoCalculating the content of FeO:
wherein,
Ψ(xtest)＝[1,K(xtest,x1),K(xt ain,x2),…,K(xtrain,xn)]T
(5) soft measurement performance test
In order to test the application effect of the RVM model constructed based on the multi-source data and the characteristics, the sintering process variables and FeO test values from 1 month and 1 day of 2020 to 1 month and 26 days of 2021 are selected to construct a soft measurement model. Due to the low assay frequency and the complex assay process, 104 assay samples are collected, the first 80 samples are used as a training set of the model, and the last 24 samples are used as a model testing set. As shown in fig. 8, on the basis of multi-source data and feature construction, a least square support vector machine (LS-SVR) and an RVM model are respectively constructed for performance comparison, wherein the front dotted line represents a training result, and the rear dotted line represents a prediction result; as can be seen from the figure, the LS-SVR model has better training effect, but the prediction effect deviates from the test value more obviously, and more obvious overfitting exists, while the training error of the RVM modelThe method is consistent with the basic value of the prediction error, and the prediction effect is obviously better than that of the basic value of the prediction error. The specific prediction results of the model can be compared with those of Table 2, and the prediction effect is determined by the Root Mean Square Error (RMSE) and the coefficient of determination (R)2) To be measured together. In addition to the two models mentioned in FIG. 8, Table 2 provides the prediction results for both LS-SVR and RVM models constructed without regard to characteristics.
On the whole, the RVM model constructed based on multi-source data and characteristics has the optimal prediction effect, the model simultaneously considers sensor information in the sintering process and image information of the tail section of a sintering machine, information acquisition is completed by combining the long-term hysteresis of the sintering process flow, and richer data information in the whole flow is provided for the model; meanwhile, the model carries out pretreatment on data by adopting a characteristic construction mode, so that the interpretation capability of the RVM model on nonlinear characteristics and dynamic characteristics in the sintering process data is improved; finally, the RVM model completes the training and prediction of the model based on the Bayesian framework, the nature of the probability model and the sparsity of the parameters provide stable modeling capability for the sintering process full of random noise, and overfitting of the model can be effectively avoided.
TABLE 2 comparison of model predictions
Claims (2)
1. A construction method of a soft measurement model of the FeO content in RVM sinter based on multi-source data and characteristics is characterized in that the acquisition mode of the multi-source data is as follows:
based on the sampling time of the finished product sinter assay, obtaining the key process variable corresponding to the finished product sinter and the sampling time of the sintering machine tail time by combining the structure of the sintering machine and the belt running speed, and respectively finishing the acquisition of sensor data and sintering machine tail section image data in the sintering process:
collecting process variable data by attached sensors in the sintering process, and screening according to sampling time of key process variables to obtainProcess variable data Xtrain-1(n × m), the number of samples is n, and the number of variables is m;
the method comprises the steps of setting a camera at the tail of a sintering machine to continuously shoot a section of a sintered ore, screening out video pictures 3 minutes before and after the moment according to the moment of the tail of the sintering machine, adopting a red component of an image as a gray scale image of the section image of the tail of the sintering machine to finish the calculation of brightness L, adopting the red component of the section image of the tail of the sintering machine to select key frames of a video at the peak value of video brightness within 6 minutes, and calculating an average value, wherein the brightness L of each key frame picture is specifically represented as:
wherein h and w represent the height and width of a picture, qijRepresenting the pixel value corresponding to the point of the ith row and the jth column in the image; extracting image details based on a machine tail section RGB image by adopting a top-hat filtering method, converting the image details into a gray-scale image, searching the edge of the image details and completing filling, calculating the proportion of the image details in the image area, calculating the mean value of all key frames, and finally, forming variable data X about the machine tail section by the brightness and the porosity of the machine tail section imagetrain-2(n×2)；
The characteristic construction mode in the sintering process is as follows:
mixing Xtrain-1And Xtrain-2Input data X combined into modeltrain-oAnd the test of the FeO content of the finished sintered ore is marked as ytrainFirst, define the outlier as a point that differs from the mean by more than three standard deviations and replace it with the nearest non-outlier, and second, with Xtrain-oAnd ytrainOn the basis of the correlation of (A) to Xtrain-oCompleting variable construction to enhance the model interpretation capability on the data nonlinear characteristics, and specifically comprising the following six ways:
vnew＝ev
vnew＝vx
vnew＝v1+v2
vnew＝v1×v2
where v denotes the original variable, v1And v2Representing original, heterogeneous variables, vnewRepresenting the constructed new variable; in addition, with Xtrain-oAnd finishing variable construction by taking autocorrelation of historical sampling points as a reference to enhance the model interpretation capability of the dynamic characteristics of the data, wherein the expression is as follows:
vnew＝[vt vt-1 … vt-x]
in the formula, t represents sampling time, X represents the number of historical sampling points, and input data after variable construction is recorded as Xtrain(ii) a The construction mode of the RVM sinter FeO content soft measurement model is as follows:
completing data standardization based on collected sintering data and constructing XtrainAnd ytrainThe mapping relationship between single samples of the RVM model is expressed as:
y＝f(xtrain，w)+ε
wherein w ═ w0，w1，w2，...，wn]TIs a weight vector, e to N (0, σ)2) Representing gaussian white noise, f (x) is represented as:
where K (x) is a kernel function, Ψ (x)train)＝[1，K(xtrain，x1)，K(xtrain，x2)，...，K(xtrain，xn)]TBased on the above conditions, the conditional probability formula for the output value y is expressed as:
p(y|xtrain)～N(f(xtrain，w)，σ2)
then, in the case where the observed objects y are independent of each other, the likelihood function of the whole sample is expressed as:
therein, Ψ (X)train)＝[Ψ1(xtrain)，Ψ2(xtrain)，...，Ψn(xtrain)]On the basis, the prior probability distribution is set for the parameters to avoid the over-fitting problem, and the expression is as follows:
based on Bayes theorem, posterior distribution of weight w is obtained, and parameter sigma is obtained through maximization marginal likelihood function2And optimal selection of alpha
And alphaoAnd further, the posterior mean and the covariance of the weight vector w are obtained:
during the parameter training process, most of alphaiThe value tends to infinity, and the corresponding weight value wiThen it is 0, so that the RVM model has good sparsity.
2. The application method of the RVM sintered ore FeO content soft measurement model constructed based on the multi-source data and the characteristics according to claim 1 is characterized in that the application process is as follows:
step 2.1: for new finished ore in the sintering process, screening sensor data x in the sintering process corresponding to the current tail timetest-1(1 xm) and tail section variable data xtest-2(1×2)；
Step 2.2: processing outliers of the data, and obtaining a construction variable x by adopting a variable construction method selected by an offline modeltest；
Step 2.3: x is to betestPerforming normalization process according to the parameters
And alphaoCalculating the content of FeO:
wherein,
Ψ(xtest)＝[1，K(xtest，x1)，K(xtrain，x2)，...，K(xtrain，xn)]T。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110497744.9A CN113283163B (en) | 2021-05-08 | 2021-05-08 | Construction and application of RVM sinter FeO content soft measurement model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110497744.9A CN113283163B (en) | 2021-05-08 | 2021-05-08 | Construction and application of RVM sinter FeO content soft measurement model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113283163A true CN113283163A (en) | 2021-08-20 |
| CN113283163B CN113283163B (en) | 2022-04-19 |
Family
ID=77278116
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110497744.9A Active CN113283163B (en) | 2021-05-08 | 2021-05-08 | Construction and application of RVM sinter FeO content soft measurement model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113283163B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115436371A (en) * | 2022-09-29 | 2022-12-06 | 浙江大学 | Multi-source information fusion-based soft measurement method for FeO content in auto former sinter |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060272447A1 (en) * | 2005-05-06 | 2006-12-07 | Peaslee Kent D | Continuous steel production and apparatus |
| KR101504280B1 (en) * | 2013-10-14 | 2015-03-19 | 동국제강주식회사 | METHODE OF FeO PREDICTION FOR MOLTEN SLAG IN ELECTRIC ARC FURNACE |
| CN108595892A (en) * | 2018-05-11 | 2018-09-28 | 南京林业大学 | Soft-measuring modeling method based on time difference model |
| CN111128313A (en) * | 2019-07-16 | 2020-05-08 | 中南大学 | Method and system for detecting FeO content of sinter |
| CN111862093A (en) * | 2020-08-06 | 2020-10-30 | 华中科技大学 | Corrosion grade information processing method and system based on image recognition |
-
2021
- 2021-05-08 CN CN202110497744.9A patent/CN113283163B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060272447A1 (en) * | 2005-05-06 | 2006-12-07 | Peaslee Kent D | Continuous steel production and apparatus |
| KR101504280B1 (en) * | 2013-10-14 | 2015-03-19 | 동국제강주식회사 | METHODE OF FeO PREDICTION FOR MOLTEN SLAG IN ELECTRIC ARC FURNACE |
| CN108595892A (en) * | 2018-05-11 | 2018-09-28 | 南京林业大学 | Soft-measuring modeling method based on time difference model |
| CN111128313A (en) * | 2019-07-16 | 2020-05-08 | 中南大学 | Method and system for detecting FeO content of sinter |
| CN111862093A (en) * | 2020-08-06 | 2020-10-30 | 华中科技大学 | Corrosion grade information processing method and system based on image recognition |
Non-Patent Citations (2)
| Title |
|---|
| HENG ZHOU 等: "Hybrid-Model-Based Intelligent Optimization of Ironmaking Process", 《IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS》 * |
| 宋宝宇 等: "烧结机尾断面图像的在线处理与分析方法研究", 《第八届（2011）中国钢铁年会论文集》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115436371A (en) * | 2022-09-29 | 2022-12-06 | 浙江大学 | Multi-source information fusion-based soft measurement method for FeO content in auto former sinter |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113283163B (en) | 2022-04-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111128313B (en) | Method and system for detecting FeO content of sinter | |
| JP7028316B2 (en) | Metallic material design support method, predictive model generation method, metallic material manufacturing method, and design support device | |
| CN114611844B (en) | Method and system for determining alloy addition amount in converter tapping process | |
| CN112001527A (en) | Industrial production process target data prediction method of multi-feature fusion deep neural network | |
| CN202865254U (en) | Converter flame temperature detection system | |
| CN111915080B (en) | Raw fuel cost optimal proportioning method based on molten iron quality constraint | |
| CN113283163B (en) | Construction and application of RVM sinter FeO content soft measurement model | |
| CN110363246A (en) | A kind of fusion method of high-spatial and temporal resolution vegetation index NDVI | |
| CN106022496A (en) | Raw material sintering burdening optimization method and system | |
| CN108875118A (en) | A kind of blast furnace molten iron silicon content prediction model accuracy estimating method and apparatus | |
| CN107610119A (en) | The accurate detection method of steel strip surface defect decomposed based on histogram | |
| CN111881864A (en) | Wisdom farming crops growth full-period full-dynamic monitoring management system | |
| CN114968997A (en) | Sintering end point forecasting method based on space-time feature extraction | |
| CN110533249B (en) | Metallurgical enterprise energy consumption prediction method based on integrated long-term and short-term memory network | |
| CN117541763A (en) | Tea fermentation degree determination method and system based on image recognition | |
| CN116955939B (en) | Meteorological element structural feature forecast error quantization expression method based on graph similarity | |
| CN110674972A (en) | Method for predicting content of raspberry kaempferol-3-O-rutinoside and ellagic acid | |
| CN115436371A (en) | Multi-source information fusion-based soft measurement method for FeO content in auto former sinter | |
| CN108845075B (en) | Compost maturity real-time prediction method based on deep learning network | |
| CN111931425A (en) | RF algorithm-based performance prediction system for hot-rolled Q355B steel | |
| CN116213095A (en) | Intelligent clean coal product ash content adjusting method and system based on dense medium separation | |
| CN110175932A (en) | The Agriculture Drought monitoring method coupled with more ten days precipitation is extracted based on remote sensing information | |
| CN115861235A (en) | Flotation tailing ash content prediction method based on multi-feature data fusion | |
| Li et al. | A Soft Sensor Model of Sintering Process Quality Index Based on Multi-Source Data Fusion | |
| CN108009343A (en) | A kind of blast furnace CO Influence factors of utilization rate analysis method and system |
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 |