Method for predicting pole changing quality based on image information and slot control data

Abstract

The invention relates to the technical field of electrolytic aluminum production pole change prediction, and discloses a method for predicting pole change quality based on image information and cell control data, and the method comprises the steps: carrying out the CV large model training through the cell control data; fuzzy classification is carried out on video information and cell control information 2 hours after pole changing, a fuzzy matrix is constructed, then self-adaptive iterative adjustment is carried out on weights by utilizing historical data, finally, the weights are converged, 18-24-hour conduction condition prediction of an anode guide rod is carried out, and the situation that the state of an electrolytic cell is influenced for a long time due to the pole changing operation problem can be reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of anode changing prediction in electrolytic aluminum production, and specifically to a method for predicting anode changing quality based on image information and cell control data. It aims at the data, image recognition and processing during the anode changing operation of the multifunctional overhead crane in the electrolysis workshop, and uses the fuzzy theory to predict the anode changing operation. Background Art

[0002] In the electrolysis workshop of an electrolytic aluminum plant, the multifunctional overhead crane undertakes multiple key production tasks including anode changing. The anode changing operation, that is, the process of replacing the anode carbon block, is a very important link in electrolytic production. Whether this link is carried out in accordance with the standard has a great impact on the voltage stability and current distribution of the electrolytic cell.

[0003] The traditional evaluation of anode changing effect mainly depends on the recovery of the anode rod current 18 - 24 hours after the change. Since the new anode has a long conduction time and it is not easy to detect problems during the process, the evaluation of the anode changing operation has a strong lag. At the same time, due to the great difficulty in quantifying the anode changing operation process and the high proportion of human factors, it is very difficult to predict the anode changing operation. With the development of intelligent and information technology, how to use more advanced methods to quantitatively evaluate the anode changing operation and reasonably predict the anode changing effect has become an urgent problem to be solved in the current electrolytic aluminum industry. Summary of the Invention

[0004] In view of the above disadvantages and deficiencies, the present invention provides a method for predicting anode changing quality based on image information and cell control data. By using data fuzzification processing, it realizes the scientific evaluation and classification of the anode changing operation quality, and finally realizes the effect prediction. The specific steps are as follows

[0005] S1. Provide the anode changing video data from multiple cameras installed above the overhead crane and the cell control data from the cell controller;

[0006] S2. Use the CV large model to train the anode changing video data to obtain the image score value x i1 and the image loss value x i2 , and perform fuzzy classification by defining the membership function to obtain the membership degrees of the image score value and the image loss value;

[0007] S3. Perform fuzzy classification on the cell voltage data, cell current data, and low - frequency noise data in the cell control data by defining the membership function to obtain the membership degrees of the cell voltage data x i3 , cell current data x i4 and low - frequency noise data x i5 ;

[0008] S4. For the data X i two hours after the anode change = {image score value x i1, the image loss value x i2 , the cell voltage data x i3 , the cell current data x i4 and the low-frequency noise data x i5}, construct the fuzzy matrix R of membership degrees i , converge the weights through the adaptive weight calculation method to obtain the final weight value W, and finally obtain the fuzzy classification B i , select the data with the highest membership degree as the classification for evaluating the quality of pole changing.

[0009] Respectively divide the image score value, the image loss value, the cell voltage data, the cell current data, and the low-frequency noise data into five evaluation levels: "poor", "relatively poor", "medium", "relatively good", and "good". Use the Gaussian function to calculate the membership degree of each type of data. The formula is:

[0010]

[0011] where, μ sj (x i1 ) is the membership degree of the i-th group of image score value data in the j-th category, c s is the optimal value of the image score value, σ s is the standard deviation of all the data of the image score value;

[0012] μ lj (x i2 ) is the membership degree of the i-th group of image loss value data in the j-th category, c l is the optimal value of the image loss value, σ l is the standard deviation of all the data of the image loss value;

[0013] μ vj (x i3 ) is the membership degree of the i-th group of cell voltage data in the j-th category, c v is the optimal value of the cell voltage data, σ v is the standard deviation of all the data of the cell voltage data;

[0014] μ Ij (x i4 ) is the membership degree of the i-th group of cell current data in the j-th category, c I is the optimal value of the cell current data, σ I is the standard deviation of all the data of the cell current data;

[0015] μ Lj (x i5 ) is the membership degree of the i-th group of cell voltage data in the j-th category, c L is the optimal value of the low-frequency noise data, σ LIt is the standard deviation of all the low-frequency noise data.

[0016] The higher the image score value, the higher the evaluation; the smaller the image loss value, the higher the evaluation; the cell voltage data represents the average value of the absolute value of the difference between the actual voltage value and 3.90 within 2 hours after the end of pole changing. The smaller this value, the higher the evaluation; the low-frequency noise data represents the average value of the low-frequency noise value within 2 hours of pole changing in the electrolytic cell. The smaller the value, the higher the evaluation; the average value of the absolute value of the difference between the actual current value and 500 kA. The smaller this value, the higher the evaluation.

[0017] Specifically, S4 includes the following steps:

[0018] S401. First, construct a fuzzy matrix R through the membership degrees of five categories, namely the image score value x i1 , the image loss value x i2 , the cell voltage data x i3 , the cell current data x i4 and the low-frequency noise data x i5 : i :

[0019]

[0020] S402. Set the temporary weight V of each evaluation level i , V i = {v i1 , v i2 , v i3 , v i4 , v i5}, where v i1 + v i2 + v i3 + v i4 + v i5 = 1. Set the step size s = 0.001 and fine-tune the weight value to obtain 5 groups of new temporary intermediate weights V Tij , j ∈ {1, 2, 3, 4, 5}:

[0021] V Ti1 = [v i1 + 4s, v i2 - s, v i3 - s, v i4 - s, v i5 - s]

[0022] V Ti2 = [v i1 - s, v i2 + 4s, v i3 - s, v i4 - s, v i5 - s]

[0023] VTi3 = [v i1 - s, v i2 - s, v i3 + 4s, v i4 - s, v i5 - s]

[0024] V Ti4 = [v i1 - s, v i2 - s, v i3 - s, v i4 + 4s, v i5 - s]

[0025] V Ti5 = [v i1 - s, v i2 - s, v i3 - s, v i4 - s, v i5 + 4s]

[0026] S403. Calculate the temporary fuzzy classification C of the i - th group of data Tij = R i V Tij (j ∈ {1, 2, 3, 4, 5}); S404. Introduce the historical data H of the anode rod current recovery situation in 18 - 24 hours of the anode current in actual production i ; H i = {h ij}, j ∈ {1, 2, 3, 4, 5}; Compare with H i Determine the evaluation level of the current group of data according to the current recovery rate, and set the category deviation of the evaluation level at this time to 0; And calculate the evaluation score F of the 5 - group temporary weights according to the category deviation of the current recovery value Til ;

[0027]

[0028] H max-i = arg max j∈{1，2，3，4，5} h ij

[0029] c Tijk is the membership degree of the temporary fuzzy classification evaluation; l ∈ {1, 2, 3, 4, 5};

[0030] S405. Select the group with the largest F Til evaluation score, and the corresponding weight value is the temporary weight value V Tij , and use it as the temporary weight V in the next iteration i+1 , Through multiple iterations of historical data, the final weight converges to a fixed value V i as the final weight W;

[0031] S406. Obtain the fuzzy classification B i =WR i to represent the membership degrees of the "poor", "relatively poor", "medium", "relatively good", and "good" rating levels corresponding to the X i data. Select the group with the highest membership degree as the classification for judging the pole-changing quality, realizing the fuzzy prediction of the pole-changing situation 18 - 24 hours later based on the pole-changing image data and the cell control data 2 hours after pole-changing. i

[0032] The present invention has the following beneficial effects and advantages:

[0033] The present invention trains the CV large model through the cell control data; performs fuzzy classification on the video information and the cell control information 2 hours after pole-changing to construct a fuzzy matrix, and then adaptively iteratively adjusts the weights using historical data until the weights converge to predict the conduction situation of the anode rod 18 - 24 hours later, which can reduce the long-term impact on the cell state caused by pole-changing operation problems. Specific Embodiments

[0035] The present invention will be further described below. The present invention is a method for predicting the pole-changing quality based on image information and cell control data, including the following steps: S1. Provide the pole-changing video data from multiple cameras installed above the crane and the cell control data from the cell controller;

[0036] S2. Perform image training on the pole-changing video data using the CV large model to obtain the image score value x i1 and the image loss value x i2 , and perform fuzzy classification by defining the membership function to obtain the membership degrees of the image score value and the image loss value;

[0037] S3. Perform fuzzy classification on the cell voltage data, cell current data, and low-frequency noise data in the cell control data by defining the membership function to obtain the membership degrees of the cell voltage data x i3 , cell current data x i4 and low-frequency noise data x i5 ;

[0038] Respectively divide the image score value, image loss value, cell voltage data, cell current data, and low-frequency noise data into five evaluation levels of "poor", "relatively poor", "medium", "relatively good", and "good", and calculate the membership degree of each type of data using the Gaussian function. The formula is:

[0039]

[0040] where μ sj (x​i1 ) is the membership degree that the image score value of the i-th group of image score value data belongs to the j-th category, c s is the optimal value of the image score value, σ s is the standard deviation of all data of the image score value; the higher the image score value, the higher the evaluation.

[0041] μ lj (x i2 ) is the membership degree that the image loss value of the i-th group of data of the image loss value belongs to the j-th category, c l is the optimal value of the image loss value, σ l is the standard deviation of all data of the image loss value; the smaller the image loss value, the higher the evaluation.

[0042] μ vj (x i3 ) is the membership degree that the i-th group of cell voltage data of the cell voltage data belongs to the j-th category, c v is the optimal value of the cell voltage data, σ v is the standard deviation of all data of the cell voltage data; the average value of the absolute value of the difference between the actual voltage value and 3.90, the smaller this value, the higher the evaluation.

[0043] μ Ij (x i4 ) is the membership degree that the i-th group of cell current data of the cell current data belongs to the j-th category, c I is the optimal value of the cell current data, σ I is the standard deviation of all data of the cell current data; the average value of the absolute value of the difference between the actual current value and 500 kA, the smaller this value, the higher the evaluation.

[0044] μ Lj (x i5 ) is the membership degree that the i-th group of low-frequency noise data of the cell voltage data belongs to the j-th category, c L is the optimal value of the low-frequency noise data, σ L is the standard deviation of all low-frequency noise data; the low-frequency noise data represents the average value of the low-frequency noise value within 2 hours after the anode change of the electrolytic cell, the smaller the value, the higher the evaluation.

[0045] S4. For the data X i = {image score value x i1 , image loss value x i2 , cell voltage data x i3 , cell current data x i4 and low-frequency noise data x i5} construct the fuzzy matrix R i , converge the weights through the adaptive weight calculation method to obtain the final weight value W, and finally obtain the fuzzy classification B i, select the data with the highest membership degree as the classification for evaluating the quality of pole changing.

[0046] Specifically, it includes the following steps:

[0047] S401. First, construct a fuzzy matrix R i1 using the membership degrees of five categories of the image score value x i2 , the image loss value x i3 , the cell voltage data x i4 , the cell current data x i5 and the low-frequency noise data x i :

[0048]

[0049] R i represents the fuzzy matrix corresponding to the i-th group of data.

[0050] S402. Set the temporary weight V i for each evaluation level, where V i = {v i1 , v i2 , v i3 , v i4 , v i5}, and among them, v i1 + v i2 + v i3 + v i4 + v i5 = 1. The initial temporary weight is set as V1 = {0.2, 0.2, 0.2, 0.2, 0.2}

[0051] Set the step size s = 0.001, and fine-tune the weight values to obtain 5 groups of new temporary intermediate weights

[0052] V Tij , j ∈ {1, 2, 3, 4, 5}:

[0053] V Ti1 = [v i1 + 4s, v i2 - s, v i3 - s, v i4 - s, v i5 - s]

[0054] V Ti2 = [v i1 - s, v i2 + 4s, v i3 - s, v i4 - s, v i5 - s]

[0055] V Ti3 = [vi1 -s, v i2 -s, v i3 +4s, v i4 -s, v i5 -s]

[0056] V Ti4 = [v i1 -s, v i2 -s, v i3 -s, v i4 +4s, v i5 -s]

[0057] V Ti5 = [v i1 -s, v i2 -s, v i3 -s, v i4 -s, v i5 +4s]

[0058] S403. Calculate the temporary fuzzy classification C of the i-th group of data Tij = R i V Tij (j ∈ {1, 2, 3, 4, 5}); C Tij represents the temporary fuzzy classification obtained by calculating the current i-th group of data X i and the intermediate temporary weight V Tij where C Tij = {c Tij1 , c Tij2 , c Tij3 , c Tij4 , c Tij5}, c Tijk represents the specific membership degree. For example: c Tij1 is the largest, indicating that the result calculated using the current "intermediate temporary weight" V Tij is evaluated as "good"; similarly, c Tij5 is the largest, indicating that the result calculated using the current "intermediate temporary weight" V Tij is evaluated as "bad". S404. Introduce the historical data H of the anode rod current recovery situation in 18 - 24 hours of the anode current in actual production i ; H i = {h i1 , h i2 , h i3 , h i4 , h i5}, where if the evaluation is "good", h i1 = 1, h i2 , h i3 , h i4 , h i5 are all 0.

[0059] Control H i The current recovery rate determines the evaluation level of the current group of data. Set the category deviation of the evaluation level at this time to 0, and the score is 16 points; and calculate the evaluation score F of 5 groups of temporary weights by comparing with the category deviation of the current recovery value Til ;

[0060]

[0061] H max-i = arg max j∈{1，2，3，4，5} h ij

[0062] c Tijk is the membership degree of the temporary fuzzy classification evaluation; l ∈ {1, 2, 3, 4, 5};

[0063]

[0064] Current recovery rate comparison table

[0065]

[0066] For example, in the comparison table of the category deviation of the current recovery value: if the actual current recovery measurement after 24 hours is 86%, according to the current recovery rate table, the i-group data should correspond to the "better" classification. At this time, the "better" category is the accurate value. According to the comparison table of the category deviation of the current recovery value, 16 points can be obtained. The "medium" and "good" categories differ by 1 category and get 8 points, the "worse" category differs by 2 categories and gets 4 points, and the "bad" category differs by 3 categories and gets 2 points.

[0067] S405. Select F Til The group with the highest evaluation score is found, that is, the closest temporary fuzzy classification in C is found, and the corresponding weight value is the temporary weight value V Tij , and it is used as the temporary weight V in the next iteration Tij , and through multiple iterations of historical data, the final weight converges to the fixed value V i+1 as the final weight W; i ;

[0068] S406. Obtain the fuzzy classification B through the formula B i = WR i to obtain the fuzzy classification B i , representing the membership degrees of the "bad", "worse", "medium", "better", and "good" rating levels corresponding to the X i data. Select the group with the highest membership degree as the classification for judging the quality of pole changing, and realize the fuzzy prediction of the pole changing situation 18 - 24 hours after pole changing through the pole changing image data and the 2-hour slot control data after pole changing.

[0069] The present invention conducts CV large model training through cell control data; it fuzzily classifies video information and cell control information 2 hours after pole changing to construct a fuzzy matrix, and then adaptively iteratively adjusts the weights using historical data until the weights converge, for predicting the conduction condition of anode rods within 18 - 24 hours, which can reduce the long - term impact on the cell state caused by pole - changing operation problems.

[0070] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications after learning the basic creative concepts; therefore, the claims should be construed to include the preferred embodiments as well as various modifications and variations that fall within the spirit and scope of the invention without departing from it. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and its equivalent technologies.

Claims

1. A method for predicting pole-changing quality based on image information and slot control data, comprising the following steps: S1, pole-changing video data provided by multiple sets of cameras installed above the overhead crane and slot control data provided by the slot control machine; S2. Use the CV large model to perform image training on the pole-changing video data to obtain the image score value x of the pole-changing process. i1 And the image loss value x i2 , by defining the membership function through classification, fuzzy classification is performed to obtain the membership of the image score value and the image loss value; S3, the slot voltage data, slot current data and low-frequency noise data in the slot control data are fuzzy classified by defining the membership function, and the slot voltage data x is obtained. i3 , slot current data x i4 and low frequency noise data x i5 The degree of membership; S4, data X 2 hours after pole switching i = {image score x i1 , image loss value x i2 , slot voltage data x i3 , slot current data x i4 and low frequency noise data x i5 } membership degree to construct the fuzzy matrix R i , the weights are converged by the adaptive weight calculation method, and the final weight value W is obtained, and finally the fuzzy classification B is obtained. i , select the data with the highest membership as the classification for judging the pole replacement quality.

2. The method for predicting pole changing quality based on image information and slot control data according to claim 1, characterized in that: The image score value, image loss value, slot voltage data, slot current data and low-frequency noise data are divided into five evaluation levels: "poor", "poor", "medium", "better" and "good". The membership degree of each data is calculated using the Gaussian function. The formula is: Among them, μ sj (x i1 ) is the membership degree of the image score value of the i-th group of image score value data belonging to the j-th category, c s is the optimal value of the image score, σ s is the standard deviation of all image score data; μ lj (x i2 ) is the membership degree of the image loss value of the i-th group of data image loss value belonging to the j-th category, c l is the optimal value of image loss, σ l is the standard deviation of all data of image loss values; μ vj (x i3 ) is the membership degree of the i-th group of slot voltage data belonging to the j-th category, c v is the optimal value of the slot voltage data, σ v is the standard deviation of all the slot voltage data; μ Ij (x i4 ) is the membership degree of the i-th group of slot current data belonging to the j-th category, c I is the optimal value for the slot current data, σ I is the standard deviation of all the slot current data; μ Lj (x i5 ) is the membership degree of the i-th group of low-frequency noise data of the slot voltage data to the j-th category, c L is the optimal value for low-frequency noise data, σ L is the standard deviation of all low-frequency noise data.

3. The method for predicting pole changing quality based on image information and slot control data according to claim 2, characterized in that: The higher the image scoring value, the higher the evaluation; the smaller the image loss value, the higher the evaluation; the cell voltage data represents the average value of the absolute value of the difference between the actual voltage value and 3.90 within 2 hours after the pole change, and the smaller the value, the higher the evaluation; the low-frequency noise data represents the average value of the low-frequency noise value within 2 hours after the pole change of the electrolytic cell, and the smaller the value, the higher the evaluation; the average value of the absolute value of the difference between the actual current value and 500kA, and the smaller the value, the higher the evaluation.

4. The method for predicting pole changing quality based on image information and slot control data according to claim 1, characterized in that: S4 specifically includes the following steps: S401, firstly, the image score value x i1 , image loss value x i2 , slot voltage data x i3 , slot current data x i4 and low frequency noise data x i5 The five categories of membership construct the fuzzy matrix R i : S402, set the temporary weight V of each evaluation level i , V i = {v i1 , v i2 , v i3 , v i4 , v i5 }, where v i1 +v i2 +v i3 +v i4 +v i5 =1, set the step size s = 0.001, and fine-tune the weights to obtain 5 new sets of temporary intermediate weights V Tij ,j∈{1,2,3,4,5}: V Ti1 =[v i1 +4s,v i2 -s,v i3 -s,v i4 -s,v i5 -s] V Ti2 =[v i1 -s,v i2 +4s,v i3 -s,v i4 -s,v i5 -s] V Ti3 =[v i1 -s,v i2 -s,v i3 +4s,v i4 -s,v i5 -s] V Ti4 =[v i1 -s,v i2 -s,v i3 -s,v i4 +4s,v i5 -s] V Ti5 =[v i1 -s,v i2 -s,v i3 -s,v i4 -s,v i5 +4s] S403, calculate the temporary fuzzy classification C of the i-th group of data Tij =R i V Tij (j∈{1, 2, 3, 4, 5}); S404, introduce the historical data of the anode current recovery situation of the anode guide rod in actual production for 18-24 hours i ;H i ={h ij }, j∈{1, 2, 3, 4, 5}; Compare H i The current recovery rate determines the evaluation level of the current group of data, and sets the category deviation of the evaluation level at this time to 0; and calculates the evaluation score F of the five groups of temporary weights according to the category deviation of the current recovery value Til , the formula is: H max-i =arg max j∈{1,2,3,4,5} h ij c Tijk is the membership degree of temporary fuzzy classification evaluation; l∈{1, 2, 3, 4, 5}; S405, Select F Til The group with the largest evaluation score has a corresponding weight of temporary weight value V Tij , and used as the temporary weight V in the next iteration i+1 , through multiple iterations of historical data, the final weight converges to the fixed value V i As the final weight W; S406, through formula B i =WR i Get fuzzy classification B i , indicating X i The data corresponds to the degree of membership of the rating levels of "poor", "relatively poor", "medium", "better" and "good", and the group with the highest degree of membership is selected as the classification for judging the pole changing quality. This realizes the fuzzy prediction of the pole changing situation 18-24 hours later through the pole changing image data and the slot control data 2 hours after the pole changing.

Images