CN116630748A - Rare earth electrolytic tank state multi-parameter monitoring method based on fused salt image characteristics - Google Patents
Rare earth electrolytic tank state multi-parameter monitoring method based on fused salt image characteristics Download PDFInfo
- Publication number
- CN116630748A CN116630748A CN202310711978.8A CN202310711978A CN116630748A CN 116630748 A CN116630748 A CN 116630748A CN 202310711978 A CN202310711978 A CN 202310711978A CN 116630748 A CN116630748 A CN 116630748A
- Authority
- CN
- China
- Prior art keywords
- rare earth
- molten salt
- image
- representing
- model
- 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
- 229910052761 rare earth metal Inorganic materials 0.000 title claims abstract description 91
- 150000002910 rare earth metals Chemical class 0.000 title claims abstract description 87
- 150000003839 salts Chemical class 0.000 title claims abstract description 78
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000012544 monitoring process Methods 0.000 title claims abstract description 43
- 230000003287 optical effect Effects 0.000 claims abstract description 27
- 238000006243 chemical reaction Methods 0.000 claims abstract description 24
- 238000013507 mapping Methods 0.000 claims abstract description 20
- 238000013528 artificial neural network Methods 0.000 claims abstract description 16
- 238000001914 filtration Methods 0.000 claims abstract description 12
- 238000012545 processing Methods 0.000 claims abstract description 9
- 230000009466 transformation Effects 0.000 claims abstract description 4
- 210000004027 cell Anatomy 0.000 claims description 30
- 238000006073 displacement reaction Methods 0.000 claims description 30
- 238000004364 calculation method Methods 0.000 claims description 25
- 230000006870 function Effects 0.000 claims description 21
- 238000005457 optimization Methods 0.000 claims description 17
- 238000004422 calculation algorithm Methods 0.000 claims description 13
- 230000014509 gene expression Effects 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 12
- 238000005070 sampling Methods 0.000 claims description 12
- 210000002569 neuron Anatomy 0.000 claims description 11
- 238000012549 training Methods 0.000 claims description 11
- 230000002068 genetic effect Effects 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 8
- 238000003062 neural network model Methods 0.000 claims description 7
- 238000012360 testing method Methods 0.000 claims description 6
- 238000005868 electrolysis reaction Methods 0.000 claims description 5
- 230000004044 response Effects 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 238000012937 correction Methods 0.000 claims description 3
- 230000007547 defect Effects 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 230000000007 visual effect Effects 0.000 claims description 3
- 238000013459 approach Methods 0.000 claims description 2
- 239000003086 colorant Substances 0.000 claims description 2
- 238000009826 distribution Methods 0.000 claims description 2
- 238000005286 illumination Methods 0.000 claims description 2
- 238000003384 imaging method Methods 0.000 claims description 2
- 238000005259 measurement Methods 0.000 claims description 2
- 238000010606 normalization Methods 0.000 claims description 2
- 230000011218 segmentation Effects 0.000 claims description 2
- 230000007613 environmental effect Effects 0.000 claims 1
- 238000005516 engineering process Methods 0.000 abstract description 6
- 238000013473 artificial intelligence Methods 0.000 abstract description 3
- 238000013079 data visualisation Methods 0.000 abstract description 3
- 238000002474 experimental method Methods 0.000 description 13
- 238000012706 support-vector machine Methods 0.000 description 8
- 238000004519 manufacturing process Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- -1 rare earth chloride Chemical class 0.000 description 4
- 238000001514 detection method Methods 0.000 description 3
- 238000005260 corrosion Methods 0.000 description 2
- 230000007797 corrosion Effects 0.000 description 2
- 125000004122 cyclic group Chemical group 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 229910002056 binary alloy Inorganic materials 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004737 colorimetric analysis Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000003487 electrochemical reaction Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 229910001404 rare earth metal oxide Inorganic materials 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000012552 review Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Classifications
-
- C—CHEMISTRY; METALLURGY
- C25—ELECTROLYTIC OR ELECTROPHORETIC PROCESSES; APPARATUS THEREFOR
- C25C—PROCESSES FOR THE ELECTROLYTIC PRODUCTION, RECOVERY OR REFINING OF METALS; APPARATUS THEREFOR
- C25C3/00—Electrolytic production, recovery or refining of metals by electrolysis of melts
- C25C3/34—Electrolytic production, recovery or refining of metals by electrolysis of melts of metals not provided for in groups C25C3/02 - C25C3/32
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Medical Informatics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Mathematical Optimization (AREA)
- Chemical & Material Sciences (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Pure & Applied Mathematics (AREA)
- Multimedia (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Metallurgy (AREA)
- Electrochemistry (AREA)
- Materials Engineering (AREA)
- Algebra (AREA)
- Chemical Kinetics & Catalysis (AREA)
- Organic Chemistry (AREA)
- Image Analysis (AREA)
Abstract
The application discloses a rare earth electrolytic cell state multi-parameter monitoring method based on molten salt image characteristics, which is characterized in that the state monitoring of a rare earth electrolytic cell is divided into rare earth molten salt temperature monitoring and rare earth molten salt reaction state monitoring according to the monitoring requirements of the operation state of the rare earth electrolytic cell; constructing a nonlinear mapping model of rare earth molten salt temperature through a GA-BP model, and realizing rare earth molten salt temperature monitoring; and quantifying the intensity of the rare earth molten salt reaction through dense optical flow method Farnesback, scale transformation and mean value filtering, and establishing a nonlinear mapping model between the characteristics of the rare earth molten salt motion field and the rare earth molten salt reaction state through a gate control circulation unit neural network GRU to realize the monitoring of the rare earth molten salt reaction state. The method combines image processing, artificial intelligence and data visualization technology to carry out intelligent monitoring on the running state of the rare earth electrolytic cell based on machine vision.
Description
Technical Field
The application relates to the technical field of rare earth electrolytic tanks, in particular to a rare earth electrolytic tank state multi-parameter monitoring method based on fused salt image characteristics.
Background
In the current industrial production, rare earth products are mainly obtained by a fused salt electrolysis method, and the method is mainly used for electrolyzing rare earth oxide by converting electric energy into chemical energy so as to achieve the purpose of separating and purifying rare earth metal. The rare earth electrolysis molten salt system mainly comprises a rare earth chloride molten salt system (binary system) and a rare earth fluoride molten salt system (ternary system). With the continuous development of the rare earth industry, rare earth chloride molten salt systems have been gradually eliminated, and rare earth fluoride molten salt systems have become the mainstream process of rare earth electrolysis. The rare earth electrolytic tank completes complex physical, chemical and dynamic production processes under the severe conditions of high temperature, strong light, strong corrosion and the like, and the electrolytic reaction process still belongs to a 'black box' model at present. With the continuous development of sensing technology, people continuously utilize the existing detection technology to judge and predict various parameters and reaction states in the rare earth electrolytic tank according to actual production experience. In an electrolytic environment, many detection devices are susceptible to corrosion, and first-line technicians are more faced with serious safety concerns.
In the prior art, the monitoring targets of the state of the rare earth electrolytic cell mainly comprise current, voltage, molten salt temperature and molten salt reaction state (whether the molten salt chemical reaction is sufficient or not). The current and voltage data can be directly obtained from the power input end, and the molten salt temperature and the molten salt reaction state also need to be monitored on site by technicians. Wherein, technicians often acquire rare earth molten salt temperature through an infrared thermometer and judge rare earth molten salt reaction state through expert experience. However, the two monitoring means have the defects of high cost, large potential safety hazard, non-visual monitoring information, lack of data accumulation and data traceability and the like, so that enterprises are difficult to obtain closed-loop regulation feedback, ensure that the electrolytic tank runs in a high-yield low-consumption state for a long time, and improve the production safety coefficient. Therefore, the method for monitoring the state of the rare earth electrolytic tank with high efficiency and low cost is provided with important guiding significance for optimizing the rare earth electrolytic process.
Disclosure of Invention
The application aims to provide a rare earth electrolytic cell state multi-parameter monitoring method based on fused salt image characteristics, which combines image processing, artificial intelligence and data visualization technology to carry out intelligent monitoring on the rare earth electrolytic cell running state based on machine vision, provides important closed loop feedback parameters for online control of the rare earth electrolytic cell, and provides important references for improving production safety coefficient and realizing unmanned factories.
The application aims at realizing the following technical scheme:
a rare earth electrolytic cell state multi-parameter monitoring method based on molten salt image characteristics, the method comprising:
step 1, according to the monitoring requirement of the running state of the rare earth electrolytic cell, the state monitoring of the rare earth electrolytic cell is divided into rare earth molten salt temperature monitoring and rare earth molten salt reaction state monitoring;
step 2, constructing a nonlinear mapping model of the rare earth molten salt temperature through a GA-BP model, and realizing rare earth molten salt temperature monitoring;
and 3, quantifying intensity of rare earth molten salt reaction by using a dense optical flow method Farnesback, scale transformation and mean value filtering, and establishing a nonlinear mapping model between the characteristics of the rare earth molten salt motion field and the rare earth molten salt reaction state by using a gate control circulation unit neural network GRU to realize rare earth molten salt reaction state monitoring.
According to the technical scheme provided by the application, the method is combined with image processing, artificial intelligence and data visualization technology to perform intelligent monitoring on the running state of the rare earth electrolytic cell based on machine vision, so that important closed-loop feedback parameters are provided for on-line control of the rare earth electrolytic cell, and important references are provided for improving the production safety coefficient and realizing an unmanned factory.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of a rare earth electrolytic cell state multi-parameter monitoring method based on molten salt image characteristics provided by an embodiment of the application;
FIG. 2 is a statistical histogram of displacements comparing two types of motion fields with adequate response and inadequate response according to an embodiment of the present application;
FIG. 3 is a schematic diagram of the result of the rare earth molten salt cathode rod profile fitting according to the embodiment of the application;
FIG. 4 is a graph comparing normalized velocity characteristics before and after median filtering according to an embodiment of the present application;
FIG. 5 is a schematic diagram of a model prediction result according to an embodiment of the present application.
Detailed Description
The technical solutions of the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is apparent that the described embodiments are only some embodiments of the present application, not all embodiments of the present application, and this is not limiting to the present application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to fall within the scope of the application.
Fig. 1 is a schematic flow chart of a rare earth electrolytic cell state multi-parameter monitoring method based on molten salt image characteristics, which is provided by the embodiment of the application, and includes:
step 1, according to the monitoring requirement of the running state of the rare earth electrolytic cell, the state monitoring of the rare earth electrolytic cell is divided into rare earth molten salt temperature monitoring and rare earth molten salt reaction state monitoring;
step 2, constructing a nonlinear mapping model of the rare earth molten salt temperature through a GA-BP model, and realizing rare earth molten salt temperature monitoring;
in this step, the present application is based on colorimetric features to enhance the mapping capability of the model by constructing assist features due to the disadvantage that conventional colorimetric formulas are susceptible to ambient light. Firstly, extracting colorimetric characteristics as basic variables through a colorimetric formula (1);
several classes of assist features are then initially established by extensive review of the literature: HSI (Hue, saturation and integrity), RR (Relative Red), CVA (Color Vector Angle), and YUV color features. The HSI compensates for the defect of less image information in the RGB color space by describing hue, brightness, and saturation information of the object, as shown in formula 2; the RR can effectively eliminate the influence of illumination and improve the robustness of the prediction model, as shown in a formula 5; using CVA to represent the visual difference of different colors as shown in equation 4; the YUV color model conforms to the characteristic that human vision is more sensitive to luminance information as shown in equation 3:
wherein Y represents the brightness of the pixel; u represents the difference between the red component and the luminance; v represents the difference between the blue component and the luminance; k is a correction coefficient related to camera parameters, exposure time, and emissivity of the object; after the measured object and the software and hardware parameters are determined, a mapping relation between K, R/G and T can be established theoretically, and a correction coefficient K can be obtained by experimental calibration; H. s, I the hue, brightness and saturation, respectively; r, G, B the gray scale values of the three red, green and blue channels respectively; grey represents a gray scale image; the method comprises the steps of carrying out a first treatment on the surface of the RR represents the relative red characteristic; CVA represents a color vector angle feature;
based on the limitation of a colorimetric formula, a nonlinear mapping model of rare earth molten salt temperature is constructed through a genetic algorithm GA-BP neural network model, and in the field of the nonlinear mapping model, compared with models such as polynomial regression, logistic regression, support vector machines and the like, the BP neural network has better nonlinear mapping capability, and the specific process is as follows:
the BP neural network model consists of an input layer, a hidden layer and an output layer, wherein the input layer is determined by independent variable (feature vector) dimension, the output layer is determined by dependent variable dimension, the number of nodes of the input layer of the model is 8, the number of nodes of the output layer is 6, and the number of nodes of the hidden layer refers to an empirical formula 6:
i and O respectively represent the number of input nodes and output nodes; as can be seen from the formula 6, the range of the hidden layer node N is (1, 13); the model is formed by fully connecting nodes of an input layer, a hidden layer and an output layer, the calculation flow of the model mainly comprises two parts of weight forward propagation and error reverse propagation, and the calculation formula is as follows:
wherein h is i Representing the hidden layer i-th neuron; w (w) ti Representing the connection weight of the t-th temperature input feature and the i-th hidden layer neuron; b ti Representing the bias values of the t-th temperature input feature and the i-th hidden layer neuron; w (w) it Representing the connection weight of the ith hidden layer neuron and the t output;b it representing the bias values of the ith hidden layer neuron and the t output; n represents the number of hidden layer nodes; k1 represents the number of input nodes; k2 represents the number of output nodes; i t Representing a t-th temperature input characteristic; o (O) t Representing a t-th output node; reLu and σ are nonlinear activation functions, their expressions are shown in equations 9 and 10, respectively:
in order to improve the prediction precision of the nonlinear mapping model, a genetic algorithm GA is introduced to optimize the number of hidden layer nodes, an initialization connection weight parameter and an initialization bias parameter in the BP neural network model, the GA performs optimization solution according to the optimization parameter, the parameter limit and the optimization function provided by the BP neural network model, the optimization process comprises the processes of initializing population, calculating fitness function, selecting, crossing, mutating and the like, the total number of the optimization parameters is set to be N, and the calculation method is as follows:
N＝wnum1+bnum1+wnum2+bnum2+hiddenlayersize (11)
wnum1＝inputnum*hiddenlayersize、wnum2＝hiddenlayersize*outputnum (12)
bnum1＝hiddenlayersize、bnum2＝outputnum (13)
in the above formula, wnum1 represents the number of connection weights between the input layer and the hidden layer; bnum1 represents the amount of bias weight between the input layer to the hidden layer; wnum2 represents the number of connection weights between the hidden layer and the output layer; bnum2 represents a bias weight parameter between the hidden layer and the output layer; hiddenlayersize represents the number of hidden layer nodes; inputnum represents the number of input layer nodes; outputnum represents the number of output layer nodes; the theoretical limit for each weight parameter is (-1, 1); to expand the search range, the experiment sets the limits of all weight parameters to (-3, 3).
It should be noted that, since the upper and lower limits of the number of hidden layer nodes (positive integer determined by the empirical formula) are different from the upper and lower limits of the weight parameters, the experiment refers to the greedy strategy, and after the number of hidden layer nodes is optimized, other weight parameters are optimized. Wherein, the optimization objective function of the genetic algorithm adopts a mean square error function, and the mean square error function is shown in a formula 14. Further, the recognition rate of the model was set to 0.001, and the maximum number of iterations was 3000. The population size of the optimizer is set to 30 and the maximum number of iterations is set to 100.
Wherein MES represents a mean square error; n represents the length of the sample data; y is Y i Representing the true value of the ith data tag; y is Y i ' represents the predicted value of the ith data tag;
the image brightness range of the electrolytic tank is the whole furnace table surface, and in order to ensure the consistency of data as much as possible, the experiment needs to intercept pixel blocks with the same size in the molten salt center and then take the maximum value. The step of taking the same pixel block comprises: image binarization, fused salt contour extraction, contour center point calculation, interception of two pixel blocks with the same size and merging of the pixel blocks, and for comparison with the prediction result of the classical image recognition algorithm, the size of the pixel blocks refers to the minimum size (16×16) commonly used by an image feature extraction network. In order to ensure that the data characteristics are kept unchanged while considering the image information of the left and right half outlines of the electrolytic bath, the experiment carries out merging processing on the two pixel blocks through a mean algorithm.
In a specific implementation, in order to solve the problem of large fluctuation of image sampling data, an average filter is set for filtering experimental data. Specifically, the maximum value of each channel is taken from the pixel block intercepted in the upper section to approach the peak value of the response curve of the camera, and then the maximum value sampling data (hereinafter referred to as sampling data) of the same temperature section is subjected to filtering processing. The filter width interval is 10 frames, the maximum width value is the maximum frame rate value of the color industrial camera, and the variance of the sampled data is reduced along with the increase of the filter width, which indicates that the mean value filter can effectively reduce the fluctuation value of the sampled data in the same temperature section, and the maximum frame rate of the color industrial camera is taken as the filter width in order to reduce the fluctuation of the data and the sampling time as much as possible.
Wherein, because the rare earth electrolytic cell has certain temperature inertia, namely the fluctuation of molten salt temperature can be maintained within a relatively stable range in a short time, when the experiment obtains the temperature label through the thermocouple, the temperature fluctuation of the electrolytic cell in a short time is found to be about 10 ℃, and because the normal working temperature of the electrolytic cell is about 1060-1100 ℃, the temperature soft measurement target is divided into the following temperature sections:
<1060 ℃, 1060-1070 ℃, 1070-1080 ℃, 1080-1090 ℃, 1090-1100 ℃ and >1100 ℃.
In a specific implementation, the method and the device can acquire 23490 frames of data for the 6 temperature segments, and can be seen through sampling the data: the mean filter can effectively filter out the influence of the fluctuation of the sampling data. Overall, the filtered sampled data has a more pronounced degree of differentiation. Wherein, the red channel meets the theoretical requirement that the higher the temperature is, the larger the gray response value is; the blue-green channel does not meet this rule, which may be due to: the main tone of the temperature image of the electrolytic cell is red, so that the anti-interference capability of the red channel is strong, and the other two channels are interfered by the ambient light and the complex working environment to a certain extent, which is also an important reason that the colorimetric method has larger error in practical application.
In addition, a large number of experiments show that the nonlinear mapping model constructed by the GA-BP algorithm has good prediction performance. For example, the data set is divided into a training set and a validation set in a ratio of 8:2, the input dimension of the data is 8 (basic feature+auxiliary feature), and the output dimension is 6. After training the model by training the set, predicting and verifying the verification set by using the converged model, and comparing the K Nearest Neighbor (KNN) method, the Support Vector Machine (SVM) and the optimized model results thereof, wherein:
the distance of the K nearest neighbor method is Euclidean distance, the optimization parameter is K, and the optimization range is [1,200]; the kernel function of the support vector machine adopts Gaussian kernels, the optimization parameters of the kernel function are box constraint and Kernel scale, the optimization range is [0.001,1000], the box constraint is related to the number of the support vector machines, the more the number of the support vector machines is, the better the nonlinear fitting capacity of the model is, but the time cost of the calculation model is increased, and Kernel scale is an equal-ratio array of sigma parameters in the Gaussian kernel function and is used for adjusting the scaling scale of the kernel function. The optimization model adopts a classical global optimization algorithm-genetic algorithm, the population size of the genetic algorithm is set to be 30, the maximum iteration number is 100, and the solving result of each model is shown in the following table 1:
table 1 model prediction accuracy comparison
From the results in table 1 above, it can be seen that: before optimization, the BPNN model has the best performance, and the accuracy of the training set and the accuracy of the prediction set reach the requirements of enterprise prediction accuracy (more than or equal to 90%). After genetic algorithm optimization, the performance of each model is obviously improved, wherein the GA-BP model has the best performance, and the prediction precision of a training set and the prediction precision of a test set respectively reach 98.75% and 98.41%.
And 3, quantifying intensity of rare earth molten salt reaction by using a dense optical flow method Farnesback, scale transformation and mean value filtering, and establishing a nonlinear mapping model between the characteristics of the rare earth molten salt motion field and the rare earth molten salt reaction state by using a gating circulation unit neural network GRU (Gated recurrent unit) to realize rare earth molten salt reaction state monitoring.
In the step, the optical flow method is mainly divided into a dense optical flow method and a sparse optical flow method, wherein the dense optical flow method needs to carry out optical flow solving on all pixels of an image, and the sparse optical flow method only needs to carry out optical flow solving on part of the image, so that the calculation result of the dense optical flow method is more accurate. Classical dense optical flow methods are two types, horn-Schunck and Farnesback. The Horn-Schunck optical flow method derives an optical flow iteration formula by introducing global smoothness constraint on the basis of an optical flow basic equation. The farnebback optical flow method derives a new velocity field 1 estimation equation based on polynomial expansion. Therefore, firstly, extracting the rare earth molten salt speed field characteristics by a Farnesback optical flow method, and setting the information expression of the image field as follows:
f(x)＝x T Ax+b T x+c (15)
if the domain information expression of the previous frame image is:
f 1 (x)＝x T A 1 x+b 1 T x+c 1 (16)
assuming that the next frame image generates a displacement d, based on the optical flow basic assumption (image gradient constant and local optical flow constant), the image domain information thereof is expressed as:
according to the assumption, the above polynomial coefficients are equal:
the image displacement calculation formula is obtained by the following steps:
although the above equation has been theoretically derived to calculate the displacement between two frames of images, in real solution, using a single polynomial model to approximate complex image information tends to bring about a large calculation error. To solve this problem, the local polynomial expansion is performed on the two graphs by introducing a local polynomial instead of the global polynomial in the formula (17), and expansion coefficients A1 (x), b1 (x), c1 (x) and A2 (x), b2 (x), c2 (x) of the two graphs are obtained, respectively; ideally, a2=a1 should be satisfied based on the optical flow assumption, but in practice the following approximations need to be made to reduce the error:
furthermore, Δb (x) is introduced:
at this time, the image displacement calculation formula is updated from (21) to:
A(x)d(x)＝Δb(x) (22)
since the calculation of each pixel according to the formula (24) brings about a huge amount of calculation, in order to reduce the neighborhood of the image calculation as much as possible, assuming that the image displacement changes slowly enough, d (x) needs to satisfy the minimization objective of the following function:
in the formula, w (deltax) is a weight parameter corresponding to an image pixel point, and the displacement calculation formula obtained by the least square method is as follows:
d(x)＝(∑wA T A) -1 ∑wA T Δb (24)
to improve the robustness of the model, the 8-parameter 2D motion parameterized model is expressed as:
the above formula is represented by the form of a matrix:
d＝Sp (26)
p＝(a 1 ,a 2 ,a 3 ,a 4 ,a 5 ,a 6 ,a 7 ,a 8 ) T (28)
substituting the 8-parameter 2D motion parameterized model into equation (25) yields a new weighted least squares expression as:
solving according to a least square method, and expressing a parameterized model as:
in the process of solving the parameterized model, the method comprises the following steps of respectively calculatingA i S i And->Δb i Then solving the displacement d through weighted average; so far, the model can well calculate the displacement change between two frames of images, but when the displacement is changed greatly, the model still introduces a large error. In general, the displacement variation between two frames of images of an industrial camera with a higher frame rate is relatively small. In order to overcome errors caused by large displacement, a prior displacement is introduced to update a model and an image pyramid technology to construct a multi-resolution input image;
in order to more intuitively reflect the intensity of the electrochemical reaction, the displacement of the sports field is counted, and the displacement statistical histogram of the sports field with the two types of fully reacted and insufficiently reacted is compared as shown in fig. 2, wherein the maximum value of the displacement with the fully reacted is larger than the maximum value of the displacement with the fully reacted as shown in fig. 2. This also verifies that fully reacted molten salts exhibit more strongly reacted physical characteristics and are represented by larger image shifts in figure 2. Thus, the experiment extracts the maximum value of the displacement of each motion field as a feature. Assuming that the frame rate of the industrial camera is N, the camera can capture N images in one second, and the experiment can obtain N-1 motion fields, and when the experiment is performed by comprehensively considering the time cost and the calculation cost, it is found that setting the sampling time to 1s can effectively reduce the calculation cost and the experiment time, and meanwhile, can obtain information sufficient to distinguish the two types of motion fields. Therefore, the experiment can obtain N-1 characteristic values when 1 sampling is carried out;
because the unit of the calculation result of the optical flow method is a pixel, the experiment needs to search a conversion factor to convert the image scale to the actual physical field scale, so that the fused salt motion fields obtained by different electrolytic cells are normalized to the same scale range. Because the diameter of the cathode rod in the rare earth electrolysis is fixed and vertically installed, the ratio of the diameter of the cathode rod to the diameter of the actual cathode rod in the image is calculated to be used as a normalization conversion factor, firstly, a fused salt area is separated through threshold segmentation, then the outline of the fused salt area is extracted, and finally, two parallel lines with the most distribution points in the vertical direction are fitted through the outline; fig. 3 is a schematic diagram showing a result of profile fitting of a rare earth molten salt cathode rod according to an embodiment of the present application, wherein two approximate semicircles are the profiles of molten salt regions, and two parallel vertical lines are the profiles of the cathode rod fitting;
representing the displacement of an image solved by a Farnesback optical flow method using a displacement D, D n Represents the distance between the fitting contours of the cathode bars, D t Representing the actual diameter of the cathode rod (85 mm), normalized image displacement d t The expression of (2) is:
assuming that the frame rate of an image is N fps, the time interval between two frames of images is 1/N s, so the expression of the normalized velocity v is:
so far, the normalized speed characteristics in the rare earth molten salt video stream can be extracted;
since the camera is affected by impulse noise and random noise caused by current during imaging, median filtering processing is also needed to be carried out on the data, and since the sampling time of the experiment is 1s, N-1 (N=90 fps in the text) normalized speed values can be obtained in 1s through calculation. Since the sampling time is small enough, the order of the median filter is set to N-1; as shown in fig. 4, which is a comparison chart of normalized speed characteristics before and after median filtering, according to the embodiment of the present application, as can be seen from fig. 4, the median filter can effectively distinguish two types of signals, so as to lay a foundation for the recognition of the signals by the neural network below;
the gate control circulation unit neural network GRU consists of a reset gate and an update gate, and the calculation flow is as follows:
input value X of current cell t And state h of the previous cell output t-1 The inputs of the reset gate and the update gate are composed by nonlinear combinations:
r t ＝σ(W t ·[h t-1 ,x t ]+b r ) (33)
z t ＝σ(W z ·[h t-1 ,x t ]+b z ) (34)
wherein r is t An output representing a reset gate; z t Representing the output of the update gate; w (W) t And W is z Weight parameters respectively representing reset gates and update gates; b r And b z Bias matrices representing reset gates and update gates, respectively; representing a matrix dot product; []Splicing the representation matrix; sigma denotes a nonlinear activation function sigmoid, expressed as:
the update candidate state is then:
wherein H is t Representing candidate states of the unit; w represents a weight parameter corresponding to the candidate state;a Hadamard product representing the matrix; b n Representing weight parameters corresponding to the candidate states; tanh represents a bi-tangent function, expressed as:
finally, the output of the hidden layer (cell) is:
the gate control circulation unit neural network GRU screens output information through resetting a gate and updating the gate control, so that the GRU can keep relatively stable effective information quantity in incremental time sequence information;
randomly disturbing the 106 groups of extracted features, dividing the 106 groups of features into a training set and a testing set according to the proportion of 8:2, and using the training set for training a nonlinear mapping model between the features of the rare earth molten salt sports field and the rare earth molten salt reaction state; and then, carrying out prediction verification on the test set data by using the trained nonlinear mapping model.
As shown in FIG. 5, which is a schematic diagram of the model prediction result according to the embodiment of the present application, it can be seen from FIG. 5 that the model prediction result can well hit the true value, the accuracy of model prediction is as high as 95.23%, and the requirement of the detection standard (90%) of the manufacturing enterprise can be well met.
The embodiment also compares the solving results of several classes of classical pattern recognition algorithms in the machine learning field, and a Support Vector Machine (SVM) classification kernel function adopts a Gaussian kernel with good nonlinear performance; the Back Propagation Neural Network (BPNN) adopts a classical 3-layer structure, and the number of hidden layer neurons is 30; the long and short time memory neural network (LSTM) is similar to the gating unit neural network (GRU), and the gating unit of the latter is replaced by the long and short time memory unit.
After each model is trained, the trained model is used to predict the test set data, and the comparison of model prediction results is shown in the following table 2:
table 2 model prediction results comparison
From the comparison results in Table 2, it can be seen that: the cyclic neural network has better learning and predicting capability on time sequence information, the performance of the gate-controlled cyclic unit neural network is superior to that of a long-time memory neural network, and the predicting precision can reach 95.23%.
It is noted that what is not described in detail in the embodiments of the present application belongs to the prior art known to those skilled in the art.
The foregoing is only a preferred embodiment of the present application, but the scope of the present application is not limited thereto, and any changes or substitutions easily contemplated by those skilled in the art within the scope of the present application should be included in the scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims. The information disclosed in the background section herein is only for enhancement of understanding of the general background of the application and is not to be taken as an admission or any form of suggestion that this information forms the prior art already known to those of ordinary skill in the art.
Claims (4)
1. The rare earth electrolytic cell state multi-parameter monitoring method based on molten salt image features is characterized by comprising the following steps of:
step 1, according to the monitoring requirement of the running state of the rare earth electrolytic cell, the state monitoring of the rare earth electrolytic cell is divided into rare earth molten salt temperature monitoring and rare earth molten salt reaction state monitoring;
step 2, constructing a nonlinear mapping model of the rare earth molten salt temperature through a GA-BP model, and realizing rare earth molten salt temperature monitoring;
and 3, quantifying intensity of rare earth molten salt reaction by using a dense optical flow method Farnesback, scale transformation and mean value filtering, and establishing a nonlinear mapping model between the characteristics of the rare earth molten salt motion field and the rare earth molten salt reaction state by using a gate control circulation unit neural network GRU to realize rare earth molten salt reaction state monitoring.
2. The rare earth electrolytic cell state multi-parameter monitoring method based on molten salt image characteristics according to claim 1, wherein the process of step 2 is specifically:
firstly, extracting colorimetric characteristics as basic variables through a colorimetric formula 1:
several classes of ancillary features are then established: HSI, RR, CVA and YUV color features;
the HSI compensates for the defect of less image information in the RGB color space by describing hue, brightness, and saturation information of the object, as shown in formula 2; the RR can effectively eliminate the influence of illumination and improve the robustness of the prediction model, as shown in a formula 5; using CVA to represent the visual difference of different colors as shown in equation 4; the YUV color model conforms to the characteristic that human vision is more sensitive to luminance information as shown in equation 3:
wherein Y represents the brightness of the pixel; u represents the difference between the red component and the luminance; v represents the difference between the blue component and the luminance; k is a correction coefficient, which is related to camera parameters, exposure time and emissivity of an object, and is obtained by experimental calibration; H. s, I the hue, brightness and saturation, respectively; r, G, B the gray scale values of the three red, green and blue channels respectively; grey represents a gray scale image; θ, C R 、C B All are intermediate variables; RR represents the relative red characteristic; CVA represents a color vector angle feature;
based on the limitation of a colorimetric formula, a nonlinear mapping model of rare earth fused salt temperature is constructed through a genetic algorithm GA-BP neural network model, and the specific process is as follows:
the BP neural network model consists of an input layer, a hidden layer and an output layer, wherein the input layer is determined by independent variable dimension, the output layer is determined by independent variable dimension, and the hidden layer node number refers to formula 6:
i and O respectively represent the number of input nodes and output nodes; as can be seen from the formula 8, the range of the hidden layer node N is (1, 13); the model is formed by fully connecting nodes of an input layer, a hidden layer and an output layer respectively; the calculation flow of the model consists of two parts, namely weight forward propagation and error reverse propagation, and the calculation formula is as follows:
wherein h is i Representing the hidden layer i-th neuron; w (w) ti Representing the connection weight of the t-th temperature input feature and the i-th hidden layer neuron; b ti Representing the bias values of the t-th temperature input feature and the i-th hidden layer neuron; w (w) it Representing the connection weight of the ith hidden layer neuron and the t output; b it Representing the bias values of the ith hidden layer neuron and the t output; n represents the number of hidden layer nodes; k1 represents the number of input nodes; k2 represents the number of output nodes; i t Representing a t-th temperature input characteristic; o (O) t Representing a t-th output node; reLu and σ are nonlinear activation functions, their expressions are shown in equations 11 and 12, respectively:
optimizing the number of hidden layer nodes, the initialized connection weight parameters and the initialized bias parameters in the BP neural network model by introducing a genetic algorithm GA, wherein the optimization process comprises the processes of initializing population, calculating fitness function, selecting, crossing and mutating, and the total number of the optimization parameters is set as N, and the calculation method is as follows:
N＝wnum1+bnum1+wnum2+bnum2+hiddenlayersize (11)
wnum1＝inputnum*hiddenlayersize、wnum2＝hiddenlayersize*outputnum (12)
bnum1＝hiddenlayersize、bnum2＝outputnum (13)
in the above formula, wnum1 represents the number of connection weights between the input layer and the hidden layer; bnum1 represents the amount of bias weight between the input layer to the hidden layer; wnum2 represents the number of connection weights between the hidden layer and the output layer; bnum2 represents a bias weight parameter between the hidden layer and the output layer; hiddenlayersize represents the number of hidden layer nodes; inputnum represents the number of input layer nodes; outputnum represents the number of output layer nodes; the theoretical limit for each weight parameter is (-1, 1);
wherein, the optimization objective function of the genetic algorithm is expressed as a mean square error function:
wherein MES represents a mean square error; n represents the length of the sample data; y is Y i Representing the true value of the ith data tag; y is Y i ' represents the predicted value of the ith data tag;
the image brightness range of the rare earth electrolytic tank is the whole furnace table surface, and in order to ensure the consistency of data as much as possible, the pixel blocks with the same size in the molten salt center are needed to be intercepted and then the maximum value is taken.
3. The rare earth electrolytic cell state multi-parameter monitoring method based on molten salt image characteristics according to claim 2, wherein in step 2, an average filter is further set to perform filtering processing on experimental data, specifically, the maximum value of each channel is taken from the intercepted pixel block to approach the peak value of a camera response curve, and then the maximum value sampling data of the same temperature section is subjected to filtering processing; the filter width interval is 10 frames, and the maximum width value is the maximum frame value of the color industrial camera;
wherein, because the rare earth electrolytic tank has certain temperature inertia, the temperature soft measurement target is divided into the following temperature sections:
<1060 ℃, 1060-1070 ℃, 1070-1080 ℃, 1080-1090 ℃, 1090-1100 ℃ and >1100 ℃.
4. The rare earth electrolytic cell state multi-parameter monitoring method based on molten salt image characteristics according to claim 1, wherein the process of step 3 is specifically:
firstly, extracting rare earth molten salt speed field characteristics by a Farnesback optical flow method, and setting an image field information expression as follows:
f(x)＝x T Ax+b T x+c(15)
if the domain information expression of the previous frame image is:
f 1 (x)＝x T A 1 x+b 1 T x+c 1 (16)
assuming that the next frame image generates a displacement d, based on the optical flow basic assumption, the image domain information is expressed as:
according to the assumption, the above polynomial coefficients are equal:
the image displacement calculation formula is obtained by the following steps:
the local polynomial is introduced to replace the global polynomial in the formula (17), and the local polynomial expansion is carried out on the two images to respectively obtain expansion coefficients A1 (x), b1 (x), c1 (x) and A2 (x), b2 (x) and c2 (x) of the two images; ideally, a2=a1 should be satisfied based on the optical flow assumption, but in practice the following approximations need to be made to reduce the error:
furthermore, Δb (x) is introduced:
at this time, the image displacement calculation formula is updated from (21) to:
A(x)d(x)＝Δb(x) (22)
assuming that the image displacement changes slowly enough, d (x) needs to meet the minimization objective of the following function:
in the formula, w (deltax) is a weight parameter corresponding to an image pixel point, and the displacement calculation formula obtained by the least square method is as follows:
d(x)＝(∑wA T A) -1 ∑wA T Δb (24)
to improve the robustness of the model, the 8-parameter 2D motion parameterized model is expressed as:
the above formula is represented by the form of a matrix:
d＝Sp (26)
p＝(a 1 ,a 2 ,a 3 ,a 4 ,a 5 ,a 6 ,a 7 ,a 8 ) T (28)
substituting the 8-parameter 2D motion parameterized model into equation (25) yields a new weighted least squares expression as:
solving according to a least square method, and expressing a parameterized model as:
in the process of solving the parameterized model, the method comprises the following steps of respectively calculatingAnd->Then solving the displacement d through weighted average;
because the diameter of the cathode rod in the rare earth electrolysis is fixed and vertically installed, the ratio of the diameter of the cathode rod to the diameter of the actual cathode rod in the image is calculated to be used as a normalization conversion factor, firstly, a fused salt area is separated through threshold segmentation, then the outline of the fused salt area is extracted, and finally, two parallel lines with the most distribution points in the vertical direction are fitted through the outline;
representing the displacement of an image solved by a Farnesback optical flow method using a displacement D, D n Represents the distance between the fitting contours of the cathode bars, D t Representing the actual diameter of the cathode rod (85 mm), normalized image displacement d t The expression of (2) is:
assuming that the frame rate of an image is N fps, the time interval between two frames of images is 1/N s, so the expression of the normalized velocity v is:
the normalized speed characteristics in the rare earth molten salt video stream can be extracted, the camera is influenced by impulse noise and environmental random noise caused by current when imaging, median filtering processing is needed to be carried out on the data, and the order of the median filter is set to be N-1 because the sampling time is small enough;
the gate control circulation unit neural network GRU consists of a reset gate and an update gate, and the calculation flow is as follows:
input value X of current cell t And state h of the previous cell output t-1 The inputs of the reset gate and the update gate are composed by nonlinear combinations:
r t ＝σ(W t ·[h t-1 ,x t ]+b r ) (33)
z t ＝σ(W z ·[h t-1 ,x t ]+b z ) (34)
wherein r is t An output representing a reset gate; z t Representing the output of the update gate; w (W) t And W is z Weight parameters respectively representing reset gates and update gates; b r And b z Bias matrices representing reset gates and update gates, respectively; representing a matrix dot product; []Splicing the representation matrix; sigma denotes a nonlinear activation function sigmoid, expressed as:
the update candidate state is then:
wherein H is t Representing candidate states of the unit; w represents a weight parameter corresponding to the candidate state;a Hadamard product representing the matrix; b n Representing weight parameters corresponding to the candidate states; tanh represents a bi-tangent function, expressed as:
finally, the output of the hidden layer (cell) is:
the gate control circulation unit neural network GRU screens output information through resetting a gate and updating the gate control, so that the GRU can keep relatively stable effective information quantity in incremental time sequence information;
randomly disturbing the extracted features, dividing the features into a training set and a testing set according to the proportion of 8:2, and using the training set for training a nonlinear mapping model between the features of the rare earth molten salt sports field and the rare earth molten salt reaction state; and then, carrying out prediction verification on the test set data by using the trained nonlinear mapping model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310711978.8A CN116630748A (en) | 2023-06-15 | 2023-06-15 | Rare earth electrolytic tank state multi-parameter monitoring method based on fused salt image characteristics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310711978.8A CN116630748A (en) | 2023-06-15 | 2023-06-15 | Rare earth electrolytic tank state multi-parameter monitoring method based on fused salt image characteristics |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116630748A true CN116630748A (en) | 2023-08-22 |
Family
ID=87641804
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310711978.8A Pending CN116630748A (en) | 2023-06-15 | 2023-06-15 | Rare earth electrolytic tank state multi-parameter monitoring method based on fused salt image characteristics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116630748A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116821694A (en) * | 2023-08-30 | 2023-09-29 | 中国石油大学(华东) | Soil humidity inversion method based on multi-branch neural network and segmented model |
| CN117253024A (en) * | 2023-11-17 | 2023-12-19 | 山东海晟盐业有限公司 | Industrial salt quality inspection control method and system based on machine vision |
-
2023
- 2023-06-15 CN CN202310711978.8A patent/CN116630748A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116821694A (en) * | 2023-08-30 | 2023-09-29 | 中国石油大学(华东) | Soil humidity inversion method based on multi-branch neural network and segmented model |
| CN116821694B (en) * | 2023-08-30 | 2023-12-01 | 中国石油大学(华东) | Soil humidity inversion method based on multi-branch neural network and segmented model |
| CN117253024A (en) * | 2023-11-17 | 2023-12-19 | 山东海晟盐业有限公司 | Industrial salt quality inspection control method and system based on machine vision |
| CN117253024B (en) * | 2023-11-17 | 2024-02-06 | 山东海晟盐业有限公司 | Industrial salt quality inspection control method and system based on machine vision |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Labach et al. | Survey of dropout methods for deep neural networks | |
| CN116630748A (en) | Rare earth electrolytic tank state multi-parameter monitoring method based on fused salt image characteristics | |
| CN113449864B (en) | Feedback type impulse neural network model training method for image data classification | |
| CN110580496A (en) | Deep migration learning system and method based on entropy minimization | |
| CN112819802B (en) | Method for supervising and predicting blast furnace condition abnormality based on tuyere information deep learning | |
| CN109785385A (en) | Visual target tracking method and system | |
| CN112651998B (en) | Human body tracking algorithm based on attention mechanism and double-flow multi-domain convolutional neural network | |
| CN103971386A (en) | Method for foreground detection in dynamic background scenario | |
| CN111639719A (en) | Footprint image retrieval method based on space-time motion and feature fusion | |
| CN112949828A (en) | Graph convolution neural network traffic prediction method and system based on graph learning | |
| CN113391607A (en) | Hydropower station gate control method and system based on deep learning | |
| CN110084201B (en) | Human body action recognition method based on convolutional neural network of specific target tracking in monitoring scene | |
| CN109242019B (en) | Rapid detection and tracking method for optical small target on water surface | |
| CN108460481B (en) | Unmanned aerial vehicle reconnaissance target evolution rule prediction method based on recurrent neural network | |
| CN112906813A (en) | Flotation condition identification method based on density clustering and capsule neural network | |
| CN114549955A (en) | Online intelligent identification method and system for blast furnace tapping endpoint | |
| CN112258525B (en) | Image abundance statistics and population identification algorithm based on bird high-frame frequency sequence | |
| CN111680713A (en) | Unmanned aerial vehicle ground target tracking and approaching method based on visual detection | |
| CN110363100A (en) | A kind of video object detection method based on YOLOv3 | |
| CN112233145A (en) | Multi-target shielding tracking method based on RGB-D space-time context model | |
| CN113436174A (en) | Construction method and application of human face quality evaluation model | |
| CN111222798B (en) | Complex industrial process key index soft measurement method | |
| CN115359407A (en) | Multi-vehicle tracking method in video | |
| CN113077423B (en) | Laser selective melting pool image analysis system based on convolutional neural network | |
| CN116484275B (en) | Aluminum electrolysis anode effect monitoring method based on self-adaptive space-time diagram convolutional neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |