Electrolytic cell molecular ratio data correction method for large model training

Abstract

The invention discloses an electrolytic cell molecular ratio data correction method for large model training, which comprises the following steps: acquiring discontinuous actual experimental data of an electrolyte molecular ratio of an aluminum factory, and carrying out abnormal value preprocessing on the data; calculating the consumption of aluminum fluoride on the current day according to the aluminum yield on the current day; calculating the amount of substance of sodium fluoride and aluminum fluoride, obtaining the variation of aluminum fluoride, and obtaining a theoretical predicted value of the molecular ratio; according to the actual experimental period of the molecular ratio of the electrolyte in the aluminum factory, calculating the deviation of the theoretical predicted value of the molecular ratio and introducing negative feedback calculation to obtain a deviation vector; and correcting the molecular ratio, and adding the deviation vector to the theoretical predicted value of the molecular ratio to obtain final molecular ratio data. According to the method, accurate data can be given to the molecular ratio of the electrolytic cell, the requirement for refined operation of the electrolytic cell is met, the precision of subsequent large model training is improved, and the training process is more accurate.

Description

Technical Field

[0001] The invention relates to the technical field of electrolytic cell molecular ratio data interpolation prediction, and in particular to an electrolytic cell molecular ratio data correction method for large model training. Background Art

[0002] The molecular ratio is an important technical parameter in the production of electrolytic aluminum. It refers to the ratio of the amount of sodium fluoride to aluminum fluoride in the electrolyte. This parameter is obtained by X-ray diffractometer testing. However, there are many electrolytic cells in the plant and a large number of test samples. It is impossible to test each cell every day. The testing frequency of the electrolytic aluminum plant is generally between 2 and 7 days. In data processing, the aluminum plant will use the molecular ratio data of the previous day instead of the data that has not been tested on that day.

[0003] When training a large model, continuous molecular ratio data is required, so the data obtained from actual tests needs to be interpolated. If the molecular ratio data of the previous day is used to replace the data without tests on that day for data analysis, the interpolated data will inevitably be inconsistent with the actual data. This is because the process of molecular ratio changes in the electrolytic cell is a continuous process. For electrolytic cells with unstable cell conditions and large parameter fluctuations, the current data completion method will make their data inaccurate, which will lead to inaccurate large models or even training failures. Therefore, accurate estimation of molecular ratio calculation methods plays a very important role in further improving the accuracy of large model data. Summary of the invention

[0004] In view of the above shortcomings and deficiencies, the present invention provides a method for correcting the molecular ratio data of an electrolytic cell for large model training, and provides a method for interpolation prediction of the molecular ratio, which corrects and accurately predicts the molecular ratio data to meet the requirements for refined operation of the electrolytic cell.

[0005] In order to achieve the above-mentioned purpose, the main technical scheme adopted by the present invention includes the following steps: S1, obtaining non-continuous actual experimental data of the electrolyte molecular ratio of the aluminum plant, and performing abnormal value preprocessing on the data;

[0006] S2. Calculate the daily consumption of aluminum fluoride based on the daily aluminum production;

[0007] S3, calculating the amount of sodium fluoride and aluminum fluoride based on the total mass of the electrolyte and the proportion of each component in the electrolyte obtained by diffractometer analysis; and obtaining the change in aluminum fluoride based on the result of S2, and obtaining the theoretical prediction value of the molecular ratio;

[0008] S4. According to the actual experimental cycle of the molecular ratio of the aluminum plant electrolyte, the deviation of the theoretical predicted value of the molecular ratio is calculated and negative feedback calculation is introduced to obtain the deviation vector;

[0009] S5. Correct the molecular ratio, add the theoretical predicted value of the molecular ratio to the deviation vector to obtain the final molecular ratio data.

[0010] In S1, non-continuous actual experimental data of periodic electrolyte molecular ratios were obtained by X-ray diffractometer analysis, and the obtained data were threshold processed to remove data with molecular ratio values ​​greater than 3.1 and less than 2.1.

[0011] In S2, the aluminum fluoride consumption coefficient is set to α, which is determined by the cell temperature and cell age of the aluminum electrolytic cell. The daily aluminum fluoride consumption f is:

[0012] f = X α A;

[0013] Wherein, A is the daily aluminum production (tons), X is a fixed constant, Xα is the average aluminum fluoride consumption per ton of aluminum produced, and f is the predicted aluminum fluoride consumption (kg).

[0014] S3 includes the following steps:

[0015] S301, by measuring the daily change in the electrolyte liquid level, combined with the length and width data of the electrolytic cell body, the total mass m of the electrolyte is obtained E ; The proportion of each component in the electrolyte is obtained by diffractometer analysis, and the mass of all components in the electrolyte is obtained:

[0016]

[0017] in, represents the total mass of aluminum fluoride in the electrolyte, m NaF represents the total mass of sodium fluoride in the electrolyte, represents the total mass of alumina in the electrolyte, m MgF represents the total mass of magnesium fluoride in the electrolyte, m LiF represents the total mass of lithium oxide in the electrolyte, m CaF represents the total mass of calcium fluoride in the electrolyte, m E represents the total mass of electrolyte;

[0018] S302, calculation of aluminum fluoride and sodium fluoride NaF The amount of substance and the molecular ratio of the day

[0019]

[0020] Among them, M NaF represents the molar mass of sodium fluoride, which is 41.9882 g / mol; represents the molar mass of aluminum fluoride, which is 83.997 g / mol;

[0021] S303. Calculate the change in mass of aluminum fluoride And the amount of aluminum fluoride converted into

[0022]

[0023] Among them, the daily amount of aluminum fluoride added is obtained by querying the control report records of the day.

[0024] S304, calculate the theoretical predicted value of molecular ratio

[0025]

[0026] S4 specifically includes the following steps:

[0027] S401, set the molecular ratio test on both sides of the aluminum plant as a test cycle, so that the results of S301 The actual molecular ratio measured on the first day of the cycle, CR Tn+1 The molecular ratio is actually measured on the last day of the cycle. The molecular ratio theoretical prediction value of S3 is used to complete the molecular ratio data that was not actually measured in an experimental cycle, and the theoretical prediction value CR on the last day is calculated. Tn ; where (n>1,n∈N);

[0028] S402, introduce negative feedback calculation to eliminate the residual e(T) between the actual measured molecular ratio and the theoretical predicted value:

[0029]

[0030] S403, calculate deviation vector Δ CR :

[0031]

[0032] Among them, K p is a constant proportionality coefficient, T i is an integral coefficient whose value is a constant.

[0033] The final molecular ratio data CR corrected by the deviation vector in S5 n The formula is as follows:

[0034]

[0035] The present invention has the following beneficial effects and advantages:

[0036] The present invention collects the electrolyte liquid level height, obtains the total mass and change of the electrolyte by analyzing the data, calculates the theoretical prediction value of the molecular ratio by the change, and finally calculates the deviation vector to correct the theoretical value according to the deviation; the method of the present invention can provide accurate data for the molecular ratio of the electrolytic cell, meets the requirements for fine operation of the electrolytic cell, helps to improve the accuracy of subsequent large-scale model training, and makes the training process more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 The present invention is a flow chart of the electrolytic cell molecular ratio data correction method;

[0038] Figure 2 This is a comparison table of aluminum fluoride consumption coefficient α. DETAILED DESCRIPTION

[0039] The present invention will be further described below in conjunction with the accompanying drawings. Figure 1 As shown, the present invention is a method for correcting electrolytic cell molecular ratio data for large model training, comprising the following steps:

[0040] S1. Obtain the non-continuous actual experimental data of the electrolyte molecular ratio of the aluminum plant. The molecular ratio values ​​recorded in the electrolytic cell control machine database may have input errors or large test value errors, so the data needs to be pre-processed for abnormal values;

[0041] Specifically, non-continuous actual experimental data of periodic electrolyte molecular ratios are obtained by X-ray diffractometer analysis, and threshold processing is performed on the obtained data to remove data with molecular ratio values ​​greater than 3.1 and less than 2.1.

[0042] S2. Calculate the daily consumption of aluminum fluoride based on the daily aluminum production. The purpose of calculating the aluminum fluoride consumption is to maintain a stable thermal balance of the electrolytic cell and a stable electrolysis temperature. In addition to maintaining voltage stability, the most important thing is the stability of the electrolyte composition in the electrolytic cell. Under the premise of ensuring constant current power supply, stable cell voltage and alumina concentration control, coupled with stable electrolyte composition control, can achieve stable electrolysis temperature control.

[0043] According to data statistics, the average consumption of aluminum fluoride in 500KA electrolytic cells is about Xkg / tAl (the value will vary for different cell types). For normal electrolytic cells, the consumption of fluoride salts (aluminum fluoride) is not large, but the consumption of fluoride salts in new cells is higher, while the consumption of fluoride salts in old cells is lower. This is mainly due to the fact that the cathode absorbs a large amount of sodium fluoride and cryolite when the new cell is started up.

[0044] Specifically, the aluminum fluoride consumption coefficient is set to α, and the α is determined by the cell temperature and cell age of the aluminum electrolytic cell (e.g. Figure 2 As shown), the daily aluminum fluoride consumption f is:

[0045] f = X α A;

[0046] Wherein, A is the daily aluminum production (tons), X is a fixed constant, Xα is the average aluminum fluoride consumption per ton of aluminum produced, and f is the predicted aluminum fluoride consumption (kg).

[0047] S3, calculating the amount of sodium fluoride and aluminum fluoride based on the total mass of the electrolyte and the proportion of each component in the electrolyte obtained by diffractometer analysis; and obtaining the change in aluminum fluoride based on the result of S2, and obtaining the theoretical prediction value of the molecular ratio;

[0048] Specifically:

[0049] S301. By measuring the daily change in the electrolyte level, the total volume of the electrolyte can be calculated by cutting the length and width data of the electrolytic cell body (the cross-sectional area of ​​the electrolytic cell can be calculated), and then the total mass m of the electrolyte can be calculated by multiplying the total volume and the electrolyte density. E ;

[0050] It should be noted that: in the electrolytic cell, the electrolyte is above the aluminum liquid level, and the anode carbon block is immersed in the electrolyte. At the same time, due to the different thickness of the furnace wall, each electrolytic cell needs to consider the above factors when calculating the total mass of the electrolyte. In the process of measuring the electrolyte height, the anode effect moment should be avoided. When the anode effect occurs, the electrolyte liquid level will fluctuate greatly. If the measurement is made at this moment, the calculation will be inaccurate. The measurement should be started after the anode effect treatment is completed.

[0051] In the electrolyte testing process, the proportion of each component in the electrolyte can be obtained through diffractometer testing, thereby obtaining the mass of all components in the electrolyte:

[0052]

[0053] in, represents the total mass of aluminum fluoride in the electrolyte, m NaF represents the total mass of sodium fluoride in the electrolyte, represents the total mass of alumina in the electrolyte, m MgF represents the total mass of magnesium fluoride in the electrolyte, m LiF represents the total mass of lithium oxide in the electrolyte, m CaF represents the total mass of calcium fluoride in the electrolyte, m E represents the total mass of electrolyte;

[0054] S302, due to and m NaF Knowing that aluminum fluoride can be calculated and sodium fluoride NaF The amount of substance and the molecular ratio of the day

[0055]

[0056] Among them, M NaF represents the molar mass of sodium fluoride, which is 41.9882 g / mol; represents the molar mass of aluminum fluoride, which is 83.997 g / mol;

[0057] S303, the amount of AlF3 in the electrolyte will change. The main reasons for the change are: consumption during the electrolysis process and the amount of fluoride salt (aluminum fluoride) added, daily loss of aluminum fluoride mass (i.e., the change in aluminum fluoride mass) ) can be obtained by:

[0058]

[0059] Among them, the daily amount of aluminum fluoride added is obtained by querying the control report records of the day.

[0060] The amount of substance converted into aluminum fluoride

[0061]

[0062] S304, calculate the theoretical predicted value of molecular ratio

[0063]

[0064] S4. According to the actual experimental cycle of the molecular ratio of the aluminum plant electrolyte, the deviation of the theoretical predicted value of the molecular ratio is calculated and negative feedback calculation is introduced to obtain the deviation vector;

[0065] Specifically:

[0066] S401, set the molecular ratio test on both sides of the aluminum plant as a test cycle (in this embodiment, one test cycle is 7 days), and let S301 obtain The actual molecular ratio measured on the first day of the cycle, CR Tn+1 The molecular ratio is actually measured on the last day of the cycle (i.e., the seventh day). The molecular ratio theoretical prediction value of S3 is used to complete the molecular ratio data that was not actually measured within an experimental cycle (the second to sixth days), and the theoretical prediction value CR on the last day (the seventh day) is calculated. Tn ;

[0067] S402, introduce negative feedback calculation to eliminate the residual e(T) between the actual measured molecular ratio and the theoretical predicted value:

[0068]

[0069] S403, calculate deviation vector ΔCR :

[0070]

[0071] Among them, K p is a constant proportionality coefficient, T i is an integral coefficient whose value is a constant.

[0072] S5. Correct the molecular ratio by using the deviation vector, and add the deviation vector to the theoretical predicted value of the molecular ratio to obtain the final molecular ratio data CR. n :

[0073]

[0074] The present invention completes data through theoretical calculation and deviation correction, performs interpolation processing on the molecular ratio, and sends the interpolation result to the database, laying a good foundation for subsequent large model training; predicts and interpolates the molecular ratio of the electrolytic cell to improve the accuracy, and improves the model precision effect for subsequent large model training.

[0075] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments after knowing the basic creative concepts; therefore, the claims should be interpreted as including the preferred embodiments and various modifications and variations of the invention without departing from the spirit and scope of the invention, so that these modifications and variations of the invention fall within the scope of the claims of the present invention and their equivalent technologies.

Claims

1. A method for correcting electrolytic cell molecular ratio data for large model training, comprising the following steps: S1. Obtain non-continuous actual experimental data of the electrolyte molecular ratio of the aluminum plant and perform outlier preprocessing on the data; S2. Calculate the daily consumption of aluminum fluoride based on the daily aluminum production; S3, calculating the amount of sodium fluoride and aluminum fluoride based on the total mass of the electrolyte and the proportion of each component in the electrolyte obtained by diffractometer analysis; and obtaining the change in aluminum fluoride based on the result of S2, and obtaining the theoretical prediction value of the molecular ratio; S4. According to the actual experimental cycle of the molecular ratio of the aluminum plant electrolyte, the deviation of the theoretical predicted value of the molecular ratio is calculated and negative feedback calculation is introduced to obtain the deviation vector; S5. Correct the molecular ratio, add the theoretical predicted value of the molecular ratio to the deviation vector to obtain the final molecular ratio data.

2. The method for correcting electrolytic cell molecular ratio data for large model training according to claim 1, characterized in that: In S1, non-continuous actual experimental data of periodic electrolyte molecular ratios were obtained by X-ray diffractometer analysis, and the obtained data were threshold processed to remove data with molecular ratio values ​​greater than 3.1 and less than 2.

1.

3. The method for correcting electrolytic cell molecular ratio data for large model training according to claim 1, characterized in that: In S2, the aluminum fluoride consumption coefficient is set to α, which is determined by the cell temperature and cell age of the aluminum electrolytic cell. The daily aluminum fluoride consumption f is: f = X α A; Wherein, A is the daily aluminum production (tons), X is a fixed constant, Xα is the average aluminum fluoride consumption per ton of aluminum produced, and f is the predicted aluminum fluoride consumption (kg).

4. The method for correcting electrolytic cell molecular ratio data for large model training according to claim 1, characterized in that: S3 includes the following steps: S301, by measuring the daily change in the electrolyte liquid level, combined with the length and width data of the electrolytic cell body, the total mass m of the electrolyte is obtained E ; The proportion of each component in the electrolyte is obtained by diffractometer analysis, and the mass of all components in the electrolyte is obtained: in, represents the total mass of aluminum fluoride in the electrolyte, m NaF represents the total mass of sodium fluoride in the electrolyte, represents the total mass of alumina in the electrolyte, m MgF represents the total mass of magnesium fluoride in the electrolyte, m LiF represents the total mass of lithium oxide in the electrolyte, m CaF represents the total mass of calcium fluoride in the electrolyte, m E represents the total mass of electrolyte; S302, calculation of aluminum fluoride and sodium fluoride NaF The amount of substance and the molecular ratio of the day Among them, M NaF represents the molar mass of sodium fluoride, which is 41.9882 g / mol; represents the molar mass of aluminum fluoride, which is 83.997 g / mol; S303. Calculate the change in mass of aluminum fluoride And the amount of aluminum fluoride converted into Among them, the daily amount of aluminum fluoride added is obtained by querying the control report records of the day. S304, calculate the theoretical predicted value of molecular ratio 5. The method for correcting electrolytic cell molecular ratio data for large model training according to claim 1, characterized in that , S4 includes the following steps: S401, set the molecular ratio test on both sides of the aluminum plant as a test cycle, so that the results of S301 The actual molecular ratio measured on the first day of the cycle, CR Tn+1 The molecular ratio is actually measured on the last day of the cycle. The molecular ratio theoretical prediction value of S3 is used to complete the molecular ratio data that was not actually measured in an experimental cycle, and the theoretical prediction value CR on the last day is calculated. Tn ; where (n>1,n∈N); S402, introduce negative feedback calculation to eliminate the residual e(T) between the actual measured molecular ratio and the theoretical predicted value: S403, calculate deviation vector Δ CR : Among them, K p is a constant proportionality coefficient, T i is an integral coefficient whose value is a constant.

6. The method for correcting electrolytic cell molecular ratio data for large model training according to claim 1, characterized in that The final molecular ratio data CR corrected by the deviation vector in S5 n The formula is as follows:

Images