Prediction method for discharge amount of CF4 generated by electrolytic aluminum production line

Abstract

The invention relates to a method for predicting the emission amount of CF4 generated by an electrolytic aluminum production line, and the method comprises the following steps: S1, building a CEMS system, detecting the flue gas emission position in the detected electrolytic aluminum production line, and obtaining the total emission amount of CF4 including dissipated CF4; s2, performing correlation analysis according to the anode effect parameters when the electrolytic aluminum generates the anode effect and the emission data of the CF4, and constructing a CF4 emission prediction model; and S3, utilizing the constructed CF4 emission prediction model to predict the CF4 emission generated by the measured electrolytic aluminum production line when the anode effect occurs. According to the invention, real-time online monitoring of flue gas emission in the electrolytic aluminum production line and CF4 concentration in the flue gas is realized, and the CF4 emission during the anode effect period is predicted in real time to obtain the influence of the duration, average voltage, power consumption and other parameters on the CF4 emission during the occurrence of the anode effect, so that prediction of the CF4 emission is realized.

Description

Technical Field

[0001] The present invention relates to a gas detection method, in particular to a method for predicting the CF4 emission generated by an electrolytic aluminum production line. Background Art

[0002] The modern aluminum industry produces aluminum using the Hall-Heroult molten salt electrolysis process. This process uses molten cryolite (Na₃AlF₆) as the solvent, alumina (Al₂O₃) as the solute, a carbon material as the anode, and liquid aluminum as the cathode. During the electrolysis process, molten aluminum forms at the cathode, while CO₂ and CO gases are released at the anode. However, when the alumina concentration in the electrolyte is low or the anode current density is high, the oxygen ion concentration near the carbon anode decreases and the fluoride ion (F⁻) concentration increases. When the anode potential reaches the discharge potential of fluoride ions, fluorocarbon gases, primarily carbon tetrafluoride (CF₄), are released, and the electrolytic cell experiences an anode effect. This anode effect not only increases energy consumption but also produces large amounts of PFCs, which have serious environmental impacts. CF₄ is one of the six greenhouse gases listed in the Kyoto Protocol and has a very high global warming potential. According to the IPCC, the aluminum electrolytic industry is the world's largest source of CF₄ emissions.

[0003] Existing gas detection methods mainly include chemical detection methods and spectroscopic detection methods. Chemical detection methods have the advantages of direct measurement methods and high measurement accuracy, but they require pre-sampling of the gas to be tested, which is cumbersome and not time-effective. Spectroscopic detection methods have the advantages of strong selectivity, high sensitivity, and fast response time, and are widely used in environmental testing. However, since electrolytic aluminum production companies generally lack effective detection methods for CF4 gas, and monitoring data is mostly obtained from environmental protection departments or third parties, resulting in a lag in information acquisition, it is difficult for companies to conduct real-time and effective self-inspection and supervision of CF4 emissions from electrolytic aluminum production lines. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for predicting the CF4 emissions generated by an electrolytic aluminum production line to solve the problem that production enterprises find it difficult to detect the CF4 emissions during the electrolytic aluminum production process by themselves.

[0005] The object of the present invention is achieved like this:

[0006] A method for predicting CF4 emissions from an electrolytic aluminum production line comprises the following steps:

[0007] S1. Use an infrared spectrometer to build a CEMS system to detect the flue gas emission locations in the electrolytic aluminum production line to obtain the total CF4 emissions including the escaped CF4 during the detection period.

[0008] S2. Based on the anode effect parameters including the occurrence time, duration, average voltage and power consumption when the anode effect occurs in electrolytic aluminum and the CF4 emission data obtained through step S1, a random forest method is used to perform correlation analysis to construct a CF4 emission prediction model.

[0009] S3. Using the constructed CF4 emission prediction model, the CF4 emission generated by the tested electrolytic aluminum production line when the anode effect occurs is predicted.

[0010] Furthermore, the CEMS system includes:

[0011] The flue gas sampling probe is installed in the flue gas exhaust duct of the electrolytic aluminum production line to collect the flue gas emitted by the electrolytic aluminum production line;

[0012] The flue gas filtering device is connected to the flue gas sampling probe through a heat tracing pipeline and is used to filter water vapor and dust from the collected flue gas;

[0013] an infrared spectrometer connected to the flue gas filter device through a pipeline, for collecting spectral data of the gas concentration of the filtered flue gas and transmitting the collected spectral data to a computer; and

[0014] The computer is connected to the infrared spectrometer through a data line and is used to perform baseline correction on the input spectral data using the endpoint weighting method, and to obtain the gas spectrum and current concentration of CF4 in the flue gas emitted by the electrolytic aluminum production line under test by calculating the corrected data.

[0015] Furthermore, the total CF4 emissions, including the escaped CF4, of the electrolytic aluminum production line under test during the test in step S1 are calculated as follows:

[0016] S1-1 uses an infrared spectrometer to collect spectral signals of CF4 standard gas with different concentration gradients. According to the corresponding relationship between the absorption peak signal in the collected spectral signal and the CF4 gas concentration, the corresponding linear fitting formula is established:

[0017] y i =a·x i +b

[0018] Among them, y i is the absorption peak signal, x i is the concentration of CF4 standard gas, a and b are fitting coefficients, i=1,…,n.

[0019] S1-2 performs baseline correction on the CF4 gas spectrum signal in the flue gas emitted by the electrolytic aluminum production line collected by the infrared spectrometer to obtain the corrected CF4 gas spectrum signal y corrected(x) is:

[0020] y corrected (x) = x(x) - B(x)

[0021] Where x is the wave number of the spectrum signal, y(x) is the collected CF4 gas spectrum signal, and B(x) is the fitting baseline.

[0022] S1-3 calculates the current CF4 gas concentration data based on the corrected CF4 gas spectrum signal and the linear fitting formula, which is the CF4 concentration data measured by the CEMS system.

[0023] S1-4 Convert the CF4 concentration data measured by the CEMS system into CF4 mass data according to the following formula:

[0024]

[0025] Among them, m ti is the mass of CF4, C ti is the concentration of CF4, t is the flue gas scanning time period of the infrared spectrometer, and F is the standard flue gas flow rate at the flue gas sampling point in the flue.

[0026] S1-5 Total CF4 emissions during the test period M td for:

[0027]

[0028] Where ti is the test start time and tl is the test end time.

[0029] S1-6 Total CF4 emissions including fugitive emissions during the test period M t for:

[0030] M t =M td / E col

[0031] Among them, E col is the electrolysis gas collection efficiency.

[0032] Furthermore, step S2 includes the following sub-steps:

[0033] S2-1 collects sample information and constructs an original data set: the sample information includes the anode effect parameters including duration, peak voltage and power consumption each time the anode effect occurs, and the actual value of the CF4 emission corresponding to the anode effect; the anode effect parameters each time the anode effect occurs and the corresponding CF4 emission constitute a sample (x i ,y i ), where x iis the characteristic vector of the anode effect parameter, y i is the CF4 emission corresponding to a certain anode effect;

[0034] Eigenvector x i for:

[0035]

[0036] Among them, t i Indicates the duration of the i-th anode effect, V i represents the peak voltage of the i-th anode effect, P i Represents the power consumption of the i-th anode effect.

[0037] The set of all samples forms the sample set D: D={(x1,y1),(x 2,y2 ),…,(x N ,y N )}; where N is the number of samples. The sample set D constitutes the original data set.

[0038] S2-2 Self-service sampling to construct training subsets: n samples are extracted from the original data set with replacement to form T training subsets.

[0039] S2-3 uses random forest method to select random features: anode effect parameter x i and CF4 emissions y i As a feature of the random forest method, when each decision tree node splits, m features are randomly selected from the sample set D as candidate features.

[0040] S2-4 constructs and trains a decision tree: uses these m candidate features to process the samples in the training subset and use them to construct a decision tree; based on the mean square error minimization criterion, the constructed decision tree is trained.

[0041] S2-5 obtains the prediction result through the decision tree: For each sample in the training subset (x i ,y i ), the prediction value of its decision tree is: The mean square error is: Based on the mean square error minimization criterion, nodes are recursively divided until the termination condition is reached, forming the prediction results of T decision trees; the average of the prediction results of T decision trees is taken as the final prediction result output:

[0042]

[0043] Where x is the anode effect parameter vector; f t (x) is the prediction function of the t-th decision tree for the input anode effect parameter vector x.

[0044] S2-6 Correlation analysis: The occurrence time in the anode effect parameter vector is matched with the CF4 emission time, and the three parameters of duration, average voltage and power consumption are trained with the total CF4 emission data using the random forest algorithm. The final prediction result output from step S2-5 is used as the predicted value of CF4 emissions corresponding to the three parameters of duration, average voltage and power consumption when the anode effect occurs at that moment, so as to obtain the CF4 emission prediction model.

[0045] Furthermore, the specific method of step S3 is: inputting the three parameters of duration, average voltage and power consumption when a certain anode effect occurs into the CF4 emission prediction model, that is, obtaining the CF4 emission generated by the anode effect.

[0046] This method uses an endpoint weighting method to perform baseline correction on spectral data, effectively eliminating baseline drift and noise interference, and significantly improving the signal-to-noise ratio of the spectral signal. By integrating the prediction results of multiple decision trees, the model is robust and has significant resistance to overfitting, providing reliable data support for optimizing production processes.

[0047] The present invention achieves real-time online monitoring of flue gas emissions and CF4 concentration in the electrolytic aluminum production line by integrating FTIR spectral analysis equipment with a CEMS system and monitoring CF4 concentration data through the CEMS system, thereby solving the problems of cumbersome operation and poor timeliness of traditional chemical methods. By using a machine learning algorithm to perform correlation analysis on the collected data and the process parameters of the anode effect of electrolytic aluminum, the established standard model is used to predict the CF4 emissions generated when the anode effect occurs, solving the problem of low prediction accuracy of the traditional single linear model. By predicting the CF4 emissions during the anode effect in real time, the influence of parameters such as the duration, average voltage and power consumption of the anode effect on the CF4 emissions can be obtained, thereby realizing the prediction of CF4 emissions, and optimizing the production process according to the prediction results, and making corresponding process adjustments in a timely manner to reduce CF4 emissions from the electrolytic aluminum production line. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a schematic diagram of the system structure of the CEMS system.

[0049] Figure 2 It is the original spectrum of CF4 signal.

[0050] Figure 3 The original spectrum of CF4 signal and the baseline fitted by endpoint weighting method are shown in Figure 2.

[0051] Figure 4 This is the CF4 spectrum after endpoint weighting correction.

[0052] Figure 5 This is the random forest training result graph.

[0053] Figure 6 This is the random forest prediction result graph.

[0054] In the figure: 1. Flue, 2. Sampling probe, 3. Heating pipeline, 4. Monitoring station, 5. Flue gas filtration device, 6. Infrared spectrometer, 7. Computer. DETAILED DESCRIPTION

[0055] The present invention will be further described below in conjunction with the accompanying drawings.

[0056] The method for predicting CF4 emissions from an electrolytic aluminum production line of the present invention comprises the following steps:

[0057] S1. Use FTIR spectroscopy equipment to build a CEMS system to detect the flue gas emission locations in the electrolytic aluminum production line to obtain the total CF4 emissions including the escaped CF4 during the detection period.

[0058] like Figure 1 As shown, the CEMS system includes: a flue gas sampling probe 2, a flue gas filtering device 5, an infrared spectrometer 6 and a computer 7. The flue gas sampling probe 2 is installed in the flue gas exhaust duct 1 of the electrolytic aluminum production line under test, and is used to collect the flue gas discharged by the electrolytic aluminum production line. The flue gas filtering device 5 is connected to the flue gas sampling probe 2 through a heating pipeline 3, and is used to filter water vapor and dust in the collected flue gas. The infrared spectrometer 6 is connected to the flue gas filtering device 6 through a pipeline, and is used to collect spectral data of the gas concentration of the filtered flue gas, and transmit the collected spectral data to the computer 7. The computer 7 is connected to the infrared spectrometer 7 through a data line, and is used to perform baseline correction on the input spectral data using the endpoint weighting method, and calculate the gas spectrum line and current concentration of CF4 in the flue gas discharged by the electrolytic aluminum production line under test by calculating the corrected data.

[0059] The sampling point can be set in an area of the flue after the boiler dust collector where the airflow is relatively stable, and should avoid the flue elbows and areas where the cross-section changes sharply. The installation position of the flue gas sampling probe 2 should be set at a distance of not less than 4 times the diameter downstream of the elbow, valve, and reducer, and not less than 2 times the diameter upstream of the above components. For a rectangular flue, the equivalent diameter is: D = 2AB / (A+B), where A and B are the side lengths of the flue. If this condition cannot be achieved, the unopened distance can be divided by the sampling tube at the following ratio: 2 / 3 from the inlet and 1 / 3 from the outlet.

[0060] The total CF4 emissions, including fugitive emissions, from the electrolytic aluminum production line tested during the test period are calculated as follows:

[0061] S1-1 uses an infrared spectrometer to collect spectral signals of CF4 standard gas with different concentration gradients. According to the corresponding relationship between the absorption peak signal in the collected spectral signal and the CF4 gas concentration, the corresponding linear fitting formula is established:

[0062] y i =a·x i +b

[0063] Among them, y i is the absorption peak signal, x i is the concentration of CF4 standard gas, a and b are fitting coefficients, i=1,…,n.

[0064] The CF4 gas spectrum signal in the flue gas emitted by the electrolytic aluminum production line under test, collected by the S1-2 infrared spectrometer, is as follows: Figure 2 As shown, the spectral signal is baseline corrected to obtain Figure 4 The corrected CF4 gas spectrum signal y is shown corrected (x) is:

[0065] y corrected (x) = y(x) - B(x)

[0066] Where x is the wave number of the spectrum signal, y(x) is the collected CF4 gas spectrum signal, and B(x) is Figure 3 The baseline of the fit is shown in red.

[0067] S1-3 calculates the current CF4 gas concentration data based on the corrected CF4 gas spectrum signal and the linear fitting formula, which is the CF4 concentration data measured by the CEMS system.

[0068] S1-4 Convert the CF4 concentration data measured by the CEMS system into CF4 mass data according to the following formula:

[0069]

[0070] Among them, m ti is the mass of CF4, C ti is the concentration of CF4, t is the flue gas scanning time period of the infrared spectrometer, and F is the standard flue gas flow rate at the flue gas sampling point in the flue.

[0071] S1-5 Total CF4 emissions during the test period M td for:

[0072]

[0073] Where ti is the test start time and tl is the test end time.

[0074] S1-6 Total CF4 emissions including fugitive emissions during the test period M t for:

[0075] M t =M td / E col

[0076] Among them, E col is the electrolysis gas collection efficiency.

[0077] S2. Based on the anode effect parameters including the occurrence time, duration, average voltage and power consumption of the anode effect during electrolytic aluminum production and the CF4 emission data obtained in step S1, a correlation analysis is performed using the random forest method to construct a CF4 emission prediction model. This specifically includes the following sub-steps:

[0078] S2-1 collects sample information and constructs an original data set: the sample information includes the anode effect parameters including duration, peak voltage and power consumption each time the anode effect occurs, and the actual value of the CF4 emission corresponding to the anode effect; the anode effect parameters each time the anode effect occurs and the corresponding CF4 emission constitute a sample (x i ,y i ), where x i is the characteristic vector of the anode effect parameter, y i is the CF4 emission corresponding to a certain anode effect;

[0079] Eigenvector x i Expressed as:

[0080]

[0081] Among them, t i Indicates the duration of the i-th anode effect, V i represents the peak voltage of the i-th anode effect, P i Represents the power consumption of the i-th anode effect.

[0082] The set of all samples forms the sample set D: D = {(x1, y1), (x2, y2), ..., (x N ,y N )}; where N is the number of samples. The sample set D constitutes the original data set.

[0083] S2-2 Self-service sampling to construct training subsets: n samples are extracted from the original data set with replacement to form T training subsets.

[0084] S2-3 uses random forest method to select random features: anode effect parameter x i and CF4 emissions yi As a feature of the random forest method, when each decision tree node splits, m features are randomly selected from the sample set D as candidate features.

[0085] S2-4 constructs and trains a decision tree: uses these m candidate features to process the samples in the training subset and use them to construct a decision tree; based on the mean square error minimization criterion, the constructed decision tree is trained.

[0086] S2-5 obtains the prediction result through the decision tree: For each sample in the training subset (x i ,y i ), the prediction value of its decision tree is: The mean square error is: Based on the mean square error minimization criterion, nodes are recursively divided until the termination condition is reached, forming the prediction results of T decision trees; the average of the prediction results of T decision trees is taken as the final prediction result output:

[0087]

[0088] Where x is the anode effect parameter vector; f t (x) is the prediction function of the t-th decision tree for the input anode effect parameter vector x.

[0089] S2-6 Correlation analysis: The occurrence time in the anode effect parameter vector is matched with the CF4 emission time, and the three parameters of duration, average voltage and power consumption are trained with the total CF4 emission data using the random forest algorithm. The final prediction result output from step S2-5 is used as the predicted value of CF4 emissions corresponding to the three parameters of duration, average voltage and power consumption when the anode effect occurs at that moment, so as to obtain the CF4 emission prediction model.

[0090] S3. Using the constructed CF4 emission prediction model, the CF4 emissions generated by the tested electrolytic aluminum production line during the anode effect are predicted. Specifically, the duration, average voltage, and power consumption of a particular anode effect are input into the CF4 emission prediction model to obtain the CF4 emissions generated by that anode effect.

[0091] A certain aluminum smelter used the present invention's method for predicting CF4 emissions from its aluminum smelter production line. Combined with CF4 concentration data from a CEMS system, the company monitored CF4 concentration for one month. Anode effect data for the company during that monitoring month was also obtained. Partial data is shown in the "Table of Anode Effect Parameters and CF4 Emissions" below. Total CF4 emissions, including fugitive emissions, were calculated using the present invention's method.

[0092]

[0093] Substitute the average voltage, duration, and power consumption information into the feature vector: The corresponding CF4 emission y i Combined into a sample data. Train the sample data and get Figure 5 The training results are shown.

[0094] The CF4 emission prediction model constructed is used to predict the CF4 emission generated by the tested electrolytic aluminum production line when the anode effect occurs. The specific method is: the duration, average voltage and power consumption of a certain anode effect are input into the CF4 emission prediction model to obtain Figure 6 The prediction results are shown in Figure 2. The correlation coefficient between the predicted value and the actual value is: R 2 =0.9722; mean square error is: MSE=0.058.

Claims

1. A method for predicting CF4 emissions from an electrolytic aluminum production line, characterized by: The following steps are involved: S1. Use an infrared spectrometer to build a CEMS system to detect the flue gas emission location in the electrolytic aluminum production line to obtain the total CF4 emission during the detection period, including the escaped CF4; S2. Based on the anode effect parameters including the occurrence time, duration, average voltage and power consumption when the anode effect occurs in the electrolytic aluminum, and the CF4 emission data obtained in step S1, a random forest method is used to perform correlation analysis to construct a CF4 emission prediction model; S3. Using the constructed CF4 emission prediction model, the CF4 emission generated by the tested electrolytic aluminum production line when the anode effect occurs is predicted.

2. The method for predicting CF4 emissions from an electrolytic aluminum production line according to claim 1, wherein: The CEMS system includes: The flue gas sampling probe is installed in the flue gas exhaust duct of the electrolytic aluminum production line to collect the flue gas emitted by the electrolytic aluminum production line; The flue gas filter device is connected to the flue gas sampling probe through a heat tracing pipeline and is used to filter water vapor and dust in the collected flue gas; the infrared spectrometer is connected to the flue gas filter device through a pipeline and is used to collect spectral data of the gas concentration of the filtered flue gas and transmit the collected spectral data to a computer; The computer is connected to the infrared spectrometer through a data line and is used to perform baseline correction on the input spectral data using the endpoint weighting method, and to obtain the gas spectrum and current concentration of CF4 in the flue gas emitted by the electrolytic aluminum production line under test by calculating the corrected data.

3. The method for predicting CF4 emissions from an electrolytic aluminum production line according to claim 2, wherein: The total CF4 emissions, including fugitive emissions, from the electrolytic aluminum production line tested during the test period are calculated as follows: S1-1 uses an infrared spectrometer to collect spectral signals of CF4 standard gas with different concentration gradients. According to the corresponding relationship between the absorption peak signal in the collected spectral signal and the CF4 gas concentration, the corresponding linear fitting formula is established: y i =a·x i +b Among them, y i is the absorption peak signal, x i is the concentration of CF4 standard gas, a and b are fitting coefficients, i=1,…,n; S1-2 performs baseline correction on the CF4 gas spectrum signal in the flue gas emitted by the electrolytic aluminum production line collected by the infrared spectrometer to obtain the corrected CF4 gas spectrum signal y corrected (x) is: y corrected (x)=y(x)-B(x) Where x is the wave number of the spectrum signal, y(x) is the collected CF4 gas spectrum signal, and B(x) is the fitted baseline; S1-3 calculates the current CF4 gas concentration data based on the corrected CF4 gas spectrum signal and the linear fitting formula, which is the CF4 concentration data measured by the CEMS system; S1-4 Convert the CF4 concentration data measured by the CEMS system into CF4 mass data according to the following formula: Among them, m ti is the mass of CF4, C ti is the concentration of CF4, t is the flue gas scanning time period of the infrared spectrometer, and F is the standard flue gas flow rate at the flue gas sampling point in the flue; S1-5 Total CF4 emissions during the test period M td for: Among them, ti is the test start time; tl is the test end time; S1-6 Total CF4 emissions including fugitive emissions during the test period M t for: M t =M td / E col Among them, E col is the electrolysis gas collection efficiency.

4. The method for predicting CF4 emissions from an electrolytic aluminum production line according to claim 1, wherein: Step S2 includes the following sub-steps: S2-1 collects sample information and constructs an original data set: the sample information includes the anode effect parameters including duration, peak voltage and power consumption each time the anode effect occurs, and the actual value of the CF4 emission corresponding to the anode effect; the anode effect parameters each time the anode effect occurs and the corresponding CF4 emission constitute a sample (x i ,y i ), where x i is the characteristic vector of the anode effect parameter, y i is the CF4 emission corresponding to a certain anode effect; Eigenvector x i for: Among them, t i Indicates the duration of the i-th anode effect, V i represents the peak voltage of the i-th anode effect, P i represents the power consumption of the i-th anode effect; The set of all samples forms the sample set D: D={(x1,y1),(x2,y2),…,(x N ,y N )}; where N is the number of samples; the sample set D constitutes the original data set; S2-2 self-service sampling, constructing training subsets: n samples are extracted from the original data set with replacement to form T training subsets; S2-3 uses random forest method to select random features: anode effect parameter x i and CF4 emissions y i As a feature of the random forest method, when each decision tree node splits, m features are randomly selected from the sample set D as candidate features; S2-4 builds and trains a decision tree: uses the m candidate features to process the samples in the training subset and build a decision tree; trains the constructed decision tree based on the mean square error minimization criterion; S2-5 obtains the prediction result through the decision tree: For each sample in the training subset (x i ,y i ), the prediction value of its decision tree is: The mean square error is: Based on the mean square error minimization criterion, nodes are recursively divided until the termination condition is reached, forming the prediction results of T decision trees; the average of the prediction results of T decision trees is taken as the final prediction result output: Where x is the anode effect parameter vector; f t (x) is the prediction function of the t-th decision tree for the input anode effect parameter vector x; S2-6 Correlation analysis: The occurrence time in the anode effect parameter vector is matched with the CF4 emission time, and the three parameters of duration, average voltage and power consumption are trained with the total CF4 emission data using the random forest algorithm. The final prediction result output from step S2-5 is used as the predicted value of CF4 emissions corresponding to the three parameters of duration, average voltage and power consumption when the anode effect occurs at that moment, so as to obtain the CF4 emission prediction model.

5. The method for predicting CF4 emissions from an electrolytic aluminum production line according to claim 1, wherein: The specific method of step S3 is: inputting the three parameters of duration, average voltage and power consumption when a certain anode effect occurs into the CF4 emission prediction model, that is, obtaining the CF4 emission generated when the anode effect occurs.

Images