Intelligent analysis method for superheat degree of aluminum electrolysis cell

Abstract

The invention provides an intelligent analysis method for the superheat degree of an aluminum electrolysis cell, and relates to the technical field of aluminum electrolysis, and the intelligent analysis method comprises the following steps: S1, setting a time period, and collecting multiple operating parameters of the aluminum electrolysis cell according to the time period, including aluminum oxide related parameters, element content, electrolyte related parameters and electrolysis cell operating parameters; s2, analyzing the multiple operation parameters by using a multi-factor trend judgment method to obtain a first analysis result; s3, analyzing the multiple operation parameters by using a big data judgment method to obtain a second analysis result; and S4, in combination with the first analysis result and the second analysis result, a final superheat degree analysis result is obtained through a comprehensive judgment method. The method can optimize the operation efficiency of the electrolytic cell, improve the product quality and prolong the service life of equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of aluminum electrolysis, and in particular to an intelligent analysis method for superheat of an aluminum electrolysis cell. Background Art

[0002] In the aluminum electrolysis industry, superheat is defined as the difference between the actual temperature of the electrolyte and its initial crystallization temperature. Maintaining the proper superheat is essential to forming an ideal furnace structure, which directly affects the stability of the electrolytic cell, the horizontal current distribution, and the current efficiency. Fluctuations in superheat can cause a series of chain reactions, including changes in the thickness of the furnace wall, the length of the cathode leg, and the level of the electrolyte and aluminum liquid. These changes in turn affect the magnetic field distribution of the electrolytic cell, which in turn changes the flow characteristics and surface morphology of the aluminum liquid.

[0003] Superheat is affected by many factors, including the cell voltage, electrolyte composition, and heat dissipation conditions. Studies have shown that reducing electrolyte temperature can be achieved in two ways: reducing superheat and reducing the crystallization temperature. The former is mainly achieved by optimizing the cell voltage and enhancing heat dissipation; the latter can be achieved by adjusting the electrolyte composition, such as reducing the molecular ratio and adding additives such as lithium salts and magnesium salts.

[0004] Maintaining a low superheat has been shown to improve current efficiency and extend the life of the electrolyzer. However, precise control of superheat still faces many challenges. Traditional superheat measurement methods mainly rely on periodic manual sampling and laboratory analysis, which has lags and discontinuities. In recent years, some online monitoring technologies such as thermocouple arrays and electromagnetic induction methods have been introduced, but there are still problems with accuracy and stability. In addition, existing superheat control strategies are usually based on simplified models or empirical rules, which are difficult to cope with complex operating conditions and multi-parameter coupling problems.

[0005] Although existing technologies have made certain progress in superheat monitoring and control, there is still a lack of a method that can comprehensively consider the influence of multiple factors and achieve precise analysis and intelligent control, which limits the further improvement of aluminum electrolysis production efficiency and quality. Summary of the invention

[0006] In view of this, the present invention proposes an intelligent analysis method for the superheat of aluminum electrolytic cells, which realizes high-precision, real-time and dynamic analysis of the superheat of aluminum electrolytic cells by comprehensively using multi-factor trend judgment method and big data analysis technology. This method can comprehensively consider a variety of influencing factors, including aluminizing parameters, element content, electrolyte characteristics and electrolytic cell operating status, and effectively integrate the advantages of traditional statistical methods and modern data mining technology through adaptive weights and multi-level decision models. By introducing fuzzy logic to deal with uncertainty, this method can more accurately reflect the complex situation in the actual production process, thereby providing more reliable and accurate decision support for production control, and ultimately achieving the goal of optimizing the operating efficiency of the electrolytic cell, improving product quality and extending the life of the equipment.

[0007] The technical solution of the present invention is implemented as follows: The present invention provides an intelligent analysis method for superheat of an aluminum electrolytic cell, comprising: S1 sets a time period, and collects multiple operating parameters of the aluminum electrolytic cell according to the time period, including alumina-related parameters, element content, electrolyte-related parameters, and electrolytic cell operating parameters; S2 uses a multi-factor trend judgment method to analyze multiple operating parameters and obtain a first analysis result; S3 uses a big data judgment method to analyze multiple operating parameters and obtain a second analysis result; S4 combines the first analysis result and the second analysis result and obtains the final superheat analysis result through a comprehensive judgment method.

[0008] On the basis of the above technical scheme, preferably, the alumina related parameters include alumina feed rate and alumina concentration; the element content includes Fe content, Si content, Ca content, Li content, K content and Mg content; the electrolyte related parameters include electrolyte temperature, electrolyte composition and electrolyte addition amount; the electrolytic cell operation parameters include the number of flicker effects, quality level, electrolytic cell temperature and electrolytic cell measurement efficiency.

[0009] Based on the above technical solution, preferably, step S2 includes: S21 Determine multi-factor indicators, including alumina feed quantity indicator, Fe, Si, Ca content indicator, scintillation effect number indicator, quality level indicator, electrolyte addition and removal indicator; S22 obtains the number of alumina feeding times within a time period, and records the corresponding alumina feeding amount, calculates the slope between every two adjacent alumina feeding amounts, calculates the rate of change of the slope, performs statistics on the rate of change of the slope within the time period, calculates statistical indicators, judges the change of superheat according to the statistical indicators, and obtains the judgment result of the first factor; wherein the statistical indicators include the average rate of change, the standard deviation of the rate of change, the proportion of positive rate of change, and the proportion of negative rate of change; S23 obtains data on the contents of Fe, Si and Ca within a time period, performs time series analysis on the three elements respectively, obtains slopes by fitting the data, calculates coefficient of variation CV based on the slope, calculates correlation coefficients between Fe and Si, and determines changes in superheat based on the slope, coefficient of variation CV and correlation coefficient to obtain determination results of the second factor; S24: obtaining the number of flicker effects within the time period, calculating a short-term moving average and a medium-term moving average, calculating a frequency distribution of the number of flicker effects within the time period, drawing a CUSUM chart according to the number of flicker effects within the time period, judging the degree of overheating according to the short-term moving average, the medium-term moving average, the frequency distribution, and the CUSUM chart, evaluating the confidence level based on the consistency of the judgments of the short-term moving average and the medium-term moving average, and if the results of the frequency analysis and the CUSUM chart are consistent, then increasing the confidence level, and obtaining the judgment result of the third factor based on the confidence level; S25: obtaining quality level measurement data within a time period, fitting the quality level measurement data within the time period, obtaining a slope, calculating the first-order difference and the second-order difference of two adjacent quality level measurement data according to the measurement time of the quality level measurement data, forming a first-order difference sequence and a second-order difference sequence, judging the change of superheat according to the slope, the first-order difference sequence and the second-order difference sequence, and obtaining the judgment result of the fourth factor; S26: obtaining the amount of electrolyte added and taken in the time period, the time interval for each electrolyte addition and the corresponding electrolytic cell temperature, calculating the average amount added and taken and the frequency of addition and taken in the time period, analyzing the correlation between the amount added and taken and the electrolytic cell temperature, judging the change of superheat according to the average amount added, the frequency of addition and taken and the temperature correlation, and obtaining the judgment result of the fifth factor; S27: Combining the first factor judgment result, the second factor judgment result, the third factor judgment result, the fourth factor judgment result, and the fifth factor judgment result to obtain a first analysis result.

[0010] Based on the above technical solution, preferably, step S27 includes: S271 assigns weights to the first factor judgment result, the second factor judgment result, the third factor judgment result, the fourth factor judgment result, and the fifth factor judgment result respectively, and the size of the weight is determined according to the importance of each factor on the superheat; S272 quantifies the judgment result of each factor, and divides the superheat change trend into three states: rising, falling and stable, and assigns values ​​of 1, -1 and 0 respectively; S273 calculates a weighted average value, that is, multiplying the quantized value of each factor judgment result by its corresponding weight and then summing them; S274 sets thresholds α and β, where 0<α<β<1, and determines the superheat change trend based on the weighted average value: If the weighted average value is greater than β, it is judged that the superheat is on an upward trend; If the weighted average value is less than -β, it is judged that the superheat is on a downward trend; If the absolute value of the weighted average is less than α, it is judged that the superheat is in a stable trend; If the absolute value of the weighted average is between α and β, it is determined that the superheat change trend is unclear; S275 Combine the consistency of the judgment results of each factor to evaluate the credibility of the judgment results: Calculate the number of factors supporting the main trend and increase the credibility if it exceeds the preset threshold; If there are factor judgment results that are contrary to the main trend, the credibility will be reduced; S276 forms a first analysis result according to the superheat variation trend and the credibility of the judgment result, including the superheat variation trend, the degree of variation and the credibility of the judgment.

[0011] Based on the above technical solution, preferably, step S3 includes: S31 In each time period, obtaining electrolyte samples, analyzing the contents and molecular ratios of Li, K, Ca, and Mg in the samples, and calculating the concentration of aluminum oxide; S32 calculates the crystallization temperature based on the content and molecular ratio of Li, K, Ca, Mg and the concentration of alumina, and calculates the initial superheat based on the electrolyte temperature and the crystallization temperature; S33 calculates the actual current efficiency of the electrolytic cell based on the data of the two aluminum inventory and the actual measured efficiency of the electrolytic cell, obtains the amount of alumina discharged during the time period based on the actual current efficiency, calculates the actual amount of alumina discharged based on the average value of the amount of alumina discharged during the time period and the alumina concentration, and calculates the theoretical amount of aluminum output based on the reaction equation of alumina; S34 collects electrolyte inlet and outlet data of the electrolytic cell, subaluminum level data, Fe and Si content of the electrolytic cell, and cell temperature change data according to a time period; S35 establishing an analysis model, the analysis model including a data preprocessing module, a feature extraction module, a correlation analysis module, a trend analysis module, an anomaly detection module, a prediction module, and a comprehensive scoring module; S36 takes the actual current efficiency, theoretical aluminum output, electrolyte in and out data of the electrolytic cell, subaqueous aluminum level data, Fe and Si content of the electrolytic cell and cell temperature change data as independent parameters, inputs them into the analysis model, uses the data preprocessing module to clean and standardize the data, uses the feature extraction module to extract key features from the independent parameters, and then inputs the key features into the correlation analysis module, trend analysis module, anomaly detection module and prediction module respectively, respectively analyzes the correlation between independent parameters, identifies the change trend of each independent parameter, identifies abnormal data points and predicts future trends, and inputs the outputs of the correlation analysis module, trend analysis module, anomaly detection module and prediction module into the comprehensive scoring module to obtain big data analysis results, including electrolytic cell status score, superheat change trend prediction and anomaly analysis; S37 combines the preliminary superheat and the big data analysis results to obtain a second analysis result.

[0012] Based on the above technical solution, preferably, step S32 includes: S321 Analyze the lithium fluoride content, magnesium fluoride content, calcium fluoride content, potassium fluoride content, molecular ratio and aluminum oxide concentration, and calculate the initial crystallization temperature:

[0013] In the formula, is the crystallization temperature, is the base temperature, is the aluminum oxide concentration, is the lithium fluoride content, is the magnesium fluoride content, is the calcium fluoride content, is the potassium fluoride content, R is the actual molecular ratio, is the base molecular ratio, a, b, c, d, e, f, g are adjustment coefficients; S322 Calculate preliminary superheat:

[0014] In the formula, is the initial superheat, is the electrolyte temperature.

[0015] Based on the above technical solution, preferably, step S33 includes: S331 Obtain the aluminum quantity data of two inventory and the actual measurement efficiency in the corresponding time period, and calculate the actual current efficiency η of the electrolytic cell t ; S332 Obtain the alumina feed quantity data within the time period and calculate the weekly average feed quantity Q ave ; S333 According to the purity of alumina and weekly average feed quantity Q ave , calculate the actual alumina feed amount Q t :

[0016] In the formula, It indicates the ratio coefficient of the actual alumina entering the electrolytic cell to the theoretical alumina; S334 Based on the reaction equation of alumina 2Al 2 O 3 +3C→4Al+3CO 2 , calculate the theoretical aluminum output m t :

[0017] in, is the stoichiometric coefficient, is the molar mass of aluminum, is the molar mass of aluminum oxide.

[0018] Based on the above technical solution, preferably, step S36 includes: S361 uses a data preprocessing module to perform data cleaning and standardization on the input independent parameters to obtain a standardized data set; S362 uses a feature extraction module to extract key features from the standardized data set, including time series features, statistical features, and frequency domain features; S363 Input key features into the correlation analysis module and calculate the Pearson correlation coefficient matrix R between features corr ; S364 Use the trend analysis module to perform time series decomposition and trend identification on key features to obtain trend indicator T ind ; S365 uses the anomaly detection module to detect anomalies of key features based on the local anomaly factor algorithm and generate anomaly index A ind ; S366 uses the prediction module, which includes the ARIMA network, to predict the future trend of key features and obtain the prediction result P res ; S367 R corr , T ind , A ind and P res Input the comprehensive scoring module and obtain the electrolytic cell status score S through weighted summation. score , Superheat change trend prediction T pred and Abnormal Analysis Report A report .

[0019] Based on the above technical solution, preferably, step S37 includes: S371 Construct the comprehensive evaluation function H(x) of superheat:

[0020] In the formula, is the initial superheat, w 1 、w 2 、w 3 、w 4 is the weight coefficient, f(A report ) is the quantitative function of the abnormal analysis report; S372 Set the superheat threshold value θ 1 ,θ 2 and θ 3 , where θ 1 <θ 2 <θ 3 ; S373 determines the superheat status according to the value of H(x): If H(x)<θ 1 , it is judged that the superheat decreases; If θ 1 ≤H(x)<θ 2 , then the superheat is determined to be stable; If θ 2 ≤H(x)<θ 3 , it is judged that the superheat increases; If H(x)≥θ 3 , then it is judged that the superheat is disordered; S374 Generate corresponding control strategy recommendations based on the superheat state judgment result, and output the superheat state judgment result and the control strategy recommendations as the second analysis result.

[0021] On the basis of the above technical solution, preferably, step S4 includes: S41 Design the fusion function F(x,y), where x represents the first analysis result, y represents the second analysis result, and introduces the dynamic weight coefficient α t and β t , used to balance the two analysis results, and define the comprehensive evaluation index Z=F(α t x,β t y); S42 Establish a weight adaptive algorithm based on historical data and dynamically adjust α t and β t ; S43 constructs a multi-level decision tree model, uses the Z value as the input of the decision tree, designs multiple decision nodes, including superheat state, change rate, and stability, and sets the superheat judgment result and confidence level at the leaf node; S44 converts the Z value into a fuzzy set, uses the Mamdani reasoning method to obtain the fuzzy output, and defuzzifies it through the centroid method to obtain the superheat judgment result; S45 integrates the outputs of decision tree and fuzzy reasoning, uses ensemble learning method, conducts voting, and obtains the final superheat judgment result; S46 selects a control strategy from the preset strategy library based on the final superheat judgment result and generates operation suggestions.

[0022] Compared with the prior art, the present invention has the following beneficial effects: (1) By comprehensively applying the multi-factor trend judgment method and big data analysis technology, a high-precision, real-time, and dynamic analysis of the superheat of aluminum electrolytic cells is achieved. This method can comprehensively consider multiple influencing factors, effectively integrate the advantages of traditional statistical methods and modern data mining technology, and provide more reliable and accurate decision support for production control, thereby optimizing the operating efficiency of the electrolytic cell, improving product quality, and extending the life of the equipment; (2) A multi-factor trend judgment method is used to analyze multiple indicators such as alumina feed rate, element content, number of scintillation effects, quality level, and electrolyte addition and take-up, and assign different weights to achieve a comprehensive assessment of superheat changes. This method can capture the slight changes in various factors and their mutual influence, improving the accuracy and reliability of the analysis; (3) Introducing the big data judgment method, by establishing an analysis model that includes modules such as data preprocessing, feature extraction, correlation analysis, trend analysis, anomaly detection and prediction, it has achieved in-depth mining and analysis of massive data. This method can extract valuable information from complex data and improve the comprehensiveness and foresight of the analysis; (4) Adopting a comprehensive judgment method, by integrating the results of the multi-factor trend judgment method and the big data judgment method, and introducing dynamic weight coefficients and multi-level decision tree models, the adaptive optimization of the analysis results is achieved. This method can balance the advantages of different analysis methods and improve the accuracy and credibility of the final analysis results; (5) Fuzzy logic is introduced to deal with uncertainty. By converting the comprehensive evaluation indicators into fuzzy sets and using the Mamdani reasoning method for defuzzification, the uncertainty and ambiguity in the actual production process are effectively handled. This method improves the practicality and robustness of the analysis results and better adapts to the complex production environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0024] Figure 1 is a flow chart of the method of the present invention; Figure 2 It is a technical implementation diagram of the present invention; Figure 3 It is a multi-factor trend judgment flow chart of the present invention; Figure 4 This is a big data judgment flow chart of the present invention. DETAILED DESCRIPTION

[0025] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] like Figure 1 and Figure 2 As shown, the present invention provides an intelligent analysis method for superheat of an aluminum reduction cell, comprising: S1 sets a time period, and collects multiple operating parameters of the aluminum electrolytic cell according to the time period, including alumina-related parameters, element content, electrolyte-related parameters, and electrolytic cell operating parameters; S2 uses a multi-factor trend judgment method to analyze multiple operating parameters and obtain a first analysis result; S3 uses a big data judgment method to analyze multiple operating parameters and obtain a second analysis result; S4 combines the first analysis result and the second analysis result and obtains the final superheat analysis result through a comprehensive judgment method.

[0027] Specifically, in one embodiment of the present invention, the time period is 1 day, 3 days, 7 days or one month. Alumina-related parameters include alumina feed rate and alumina concentration; element content includes Fe content, Si content, Ca content, Li content, K content and Mg content; electrolyte-related parameters include electrolyte temperature, electrolyte composition and electrolyte addition and removal amount; electrolytic cell operation parameters include flicker effect times, quality level, electrolytic cell temperature and electrolytic cell measurement efficiency.

[0028] Specifically, Figure 3As shown, in one embodiment of the present invention, step S2 includes: S21 Determine multiple factor indicators, including alumina feed amount indicator, Fe, Si, Ca content indicator, scintillation effect number indicator, quality level indicator, and electrolyte addition amount indicator. This embodiment selects these five parameters as factor indicators based on the key influencing factors in the actual process.

[0029] S22 obtains the number of alumina feeding times within a time period, and records the corresponding alumina feeding amount, calculates the slope between every two adjacent alumina feeding amounts, calculates the rate of change of the slope, performs statistics on the rate of change of the slope within the time period, calculates statistical indicators, judges the change of superheat according to the statistical indicators, and obtains the judgment result of the first factor; wherein the statistical indicators include the average rate of change, the standard deviation of the rate of change, the proportion of positive rate of change, and the proportion of negative rate of change.

[0030] In this embodiment, the variation of the superheat of the electrolytic cell is inferred by analyzing the variation trend of the alumina feed rate. This is based on the close relationship between the alumina feed rate and the energy balance of the electrolytic cell. The process is as follows: At a preset time period Obtain the data of n times of alumina feeding, including the feeding time t i And the corresponding feeding amount m i (i=1,2,...,n), where n is the number of data obtained for alumina cutting, and i represents the data for the i-th alumina cutting.

[0031] For every two adjacent aluminum oxide feeds, calculate the slope S i :S i =(m i+1 -m i ) / (t i+1 -t i ), (i=1,2,...,n-1). Calculate the rate of change R between adjacent slopes i : R i =(S i+1 -S i ) / S i *100%, (i=1,2,...,n-2).

[0032] Statistical indicator calculation: Average rate of change R avg : R avg =(1 / (n-2))*ΣR i . Standard deviation of rate of change : . Positive rate of change ratio P pos :P pos =(number of positive change rates) / (n-2). Negative change rate ratio Pneg :P neg =(number of negative rates of change) / (n-2).

[0033] Setting the threshold used to determine the superheat in step S22 , judge the superheat change according to the following rules: if R avg > and P pos > : Overheat judgment = "rising" elif R avg <- and P neg > : Overheat judgment = "down" elif |R avg |< and σ R < : Superheat judgment = "stable" else: Overheat judgment = "uncertain" Introducing the confidence index C 1 , used to quantify the reliability of the judgment results: C 1 =(1-|R avg | / R max )*(1-σ R / σ Rmax )*max(P pos ,P neg ), where max is the maximum statistical function, max(P pos ,P neg ) refers to the pos , P neg Take the maximum value, R max is the maximum allowable value of the preset average change rate, σ Rmax It is the maximum allowable value of the preset standard deviation of the rate of change.

[0034] Output the first factor judgment result, including: superheat change trend and judgment confidence C 1 .

[0035] This embodiment uses multi-dimensional statistical indicators, including slope change rate, average change rate, standard deviation, and positive and negative change rate ratio, to fully capture the dynamic characteristics of the material discharge amount. The introduced confidence calculation further quantifies the credibility of the judgment result, providing a more solid basis for operational decision-making.

[0036] S23 obtains data on the contents of Fe, Si and Ca within a time period, performs time series analysis on the three elements respectively, obtains slopes by fitting the data, calculates coefficient of variation CV based on the slope, calculates correlation coefficients between Fe and Si, and determines changes in superheat based on the slope, coefficient of variation CV and correlation coefficient to obtain determination results of the second factor.

[0037] This example analyzes the changing trends of the contents of the three elements Fe, Si, and Ca to infer the changes in the superheat of the electrolytic cell. The specific process is as follows: At a preset time period Obtain n times of Fe, Si and Ca content measurement data within t, including the measurement time t i And the corresponding content value Fe i 、Si i , Ca i , (i=1,2,...,n).

[0038] Perform time series analysis on each element separately: Use the least squares method to perform linear regression fitting and get the slope k Fe , k Si , k Ca :

[0039] Where y represents the content of Fe, Si or Ca, i Indicates the measurement time t i The corresponding Fe, Si or Ca content values ​​are shown below, k represents the slope. After linear regression fitting based on the measured data of Fe, Si or Ca content, the slope k obtained is recorded as k Fe , k Si , k Ca .

[0040] Calculate the coefficient of variation CV for each element: CV = σ / μ; where σ is the standard deviation and μ is the mean.

[0041] The Pearson correlation coefficient r between Fe and Si contents was calculated.

[0042] Introducing a comprehensive indicator I 1 To evaluate superheat changes: I 1 =w 1 (k Fe +k Si )+w 2 (CV Fe +CV Si )+w 3 |r|+w 4 k Ca , where w1 、w 2 、w 3 、w 4 is the weight coefficient, which will be adjusted according to the actual situation.

[0043] Set the threshold used to determine the superheat in step S23 , judge the superheat change according to the following rules: if I 1 > : Overheat judgment = "rising" elif I 1 <- : Overheat judgment = "down" elif |I 1 |≤ : Superheat judgment = "stable" else: Overheat judgment = "uncertain" Introducing the confidence index C 2 , used to quantify the reliability of the judgment results: C 2 =(1-max(|k Fe |,|k Si |,|k Ca |) / k max )*(1-max(CV Fe ,CV Si ,CV Ca ) / CV max )*(1-|r|), where max represents the maximum statistical function, |k Fe |,|k Si |,|k Ca | Refers to the absolute value of the slope of the measured data of elements Fe, Si, and Ca after linear regression fitting, CV Fe ,CV Si ,CV Ca Refers to the coefficient of variation of elements Fe, Si, and Ca, max(|k Fe |,|k Si |,|k Ca |) and max(CV Fe ,CV Si ,CV Ca ) means taking the maximum value in brackets, k max is the preset maximum allowable slope value, CV max The preset maximum allowable value of the coefficient of variation.

[0044] Output the second factor judgment result, including: superheat change trend and judgment confidence C 2 .

[0045] In this embodiment, by comprehensively considering the changing trends, volatility and correlation of multiple elements, the comprehensiveness and reliability of the analysis are improved. 1 and confidence C 2 , making the judgment results more objective and credible.

[0046] S24 obtains the number of flicker effects within a time period, calculates a short-term moving average and a medium-term moving average, calculates a frequency distribution of the number of flicker effects within the time period, draws a CUSUM chart according to the number of flicker effects within the time period, judges the overheating degree according to the short-term moving average, the medium-term moving average, the frequency distribution, and the CUSUM chart, evaluates the confidence level based on the consistency of the judgments of the short-term moving average and the medium-term moving average, if the results of the frequency analysis and the CUSUM chart are consistent, then the confidence level is increased, and a judgment result of the third factor is obtained based on the confidence level.

[0047] Specifically, in a preset time period The number of flickering effects at c time points is obtained, recorded as {(t i ,F i ),i=1,2,...,c}, where t i is the time point, F i is the corresponding number of flickering effects.

[0048] Select the window size m 1 , calculate the short-term moving average SMA i : SMA i =(1 / m 1 )*Σ(F i-j ),j=0,1,...,m 1 -1, where j represents a counting variable used to traverse the selected window size m 1 Each data point in the range is from 0 to m 1 -1; F i-j Represents the data value at time ij, that is, the current time i and its previous m 1 -Data for 1 moment.

[0049] Select the window size m 2 (m 2 >m 1 ), calculate the medium-term moving average MMA i : MMA i =(1 / m 2 )*Σ(F i-j ),j=0,1,...,m2 -1.

[0050] Divide the number of flickering effects into k intervals; calculate the frequency p of each interval a :p a =(number of data points in interval a) / n, a=1,2,...,k; calculate the entropy H of the frequency distribution: H=-Σ(p a *log2(p a )),a=1,2,...,k.

[0051] Calculate the cumulative sum S i :S i =Σ(F b -F̄),b=1,2,...,i, where F̄ is the average value of all data, F b is the number of flickering effects at the bth time point; calculate the slope change ΔS of the CUSUM graph i :ΔS i =(S i -S i-1 ) / (t i -t i-1 ).

[0052] Introducing a comprehensive indicator I 2 To evaluate superheat changes: I 2 =w 1 *(SMA c -MMA c )+w 2 *H+w 3 *ΔS c , where SMA c Indicates time point t c The short-term moving average, MMA c The time point is t c The medium-term moving average, ΔS c The time point is t c The slope change of the CUSUM graph, w 1 、w 2 、w 3 is the weight coefficient, which will be adjusted according to the actual situation.

[0053] Setting the threshold used to determine the superheat degree in step S24 , judge the superheat change according to the following rules: if I 2 > : Overheat judgment = "rising" elif I 2 <- : Overheat judgment = "down" elif |I 2 |≤ : Superheat judgment = "stable" else: Overheat judgment = "uncertain" Introducing the confidence index C 3 , used to quantify the reliability of the judgment results: C 3 =(1-|ΔSMA-ΔMMA| / ΔMA max )*(1-H / H max )*(1-|ΔS| / ΔS max ), ΔSMA and ΔMMA are the rates of change of the short-term and medium-term moving averages, ΔS is the slope change of the CUSUM chart, and ΔMA max is the maximum allowed value of the preset moving average, H max is the maximum permissible value of the entropy of the preset frequency distribution, ΔS max The maximum allowed value for the slope change of the preset CUSUM chart.

[0054] Output the third factor judgment result, including: superheat change trend and judgment confidence C 3 .

[0055] In this embodiment, the moving average, frequency distribution and CUSUM chart analysis are comprehensively considered to provide a more comprehensive and reliable flicker effect analysis. 2 and confidence C 3 , making the judgment results more objective and credible.

[0056] S25 obtains the quality level measurement data within the time period, fits the quality level measurement data within the time period, obtains the slope, calculates the first-order difference and the second-order difference of two adjacent quality level measurement data according to the measurement time of the quality level measurement data, forms a first-order difference sequence and a second-order difference sequence, judges the change of superheat according to the slope, the first-order difference sequence and the second-order difference sequence, and obtains the judgment result of the fourth factor.

[0057] In this embodiment, during a preset time period Within the time period, the quality level measurement data of c time points are obtained, which is recorded as {(t i ,L i ),i=1,2,...,c}, where t i is the time point, L i is the corresponding quality level value.

[0058] Use polynomial regression to fit the quality level data L(t): L(t)=a 0 +a1 t+a 2 t 2 +...+a m t m , where m is the polynomial order, a 0 、a 1、 ...a m represents the polynomial coefficient, a m is the coefficient of the mth order, t m Represents the mth power of the time variable t, and the polynomial coefficient a is determined by the least squares method 0 、a 1、 ...a m .

[0059] Calculate the first-order derivative L'(t) and second-order derivative L''(t) of the fitting curve: L'(t)=a 1 +2a 2 t+...+ma m t m-1 ; L''(t)=2a 2 +...+m(m-1)a m t m-2 , m is the polynomial order.

[0060] Calculate the first-order difference sequence ΔL i :ΔL i =L i -L i-1 ,i=2,3,...,c; calculate the second-order difference sequence Δ²L i :Δ²L i =ΔL i -ΔL i-1 ,i=3,4,...,c.

[0061] The quality level data is decomposed by wavelet to obtain detail coefficients and approximate coefficients of different scales: {cA j ,cD j ,cD j-1 ,...,cD 1}=WaveletDecompose(L), where j is the number of decomposition layers, cA j is the j-th layer approximation coefficient, cD j is the j-th layer detail coefficient, WaveletDecompose(L) is the wavelet decomposition function, and L represents the quality level data.

[0062] Calculate the Lyapunov exponent λ of the quality level time series: λ=lim(t→∞)(1 / t)*ln(|dL(t) / dL(0)|), where t represents the time variable.

[0063] Calculate the correlation dimension D of the quality level time series: D = lim(ε→0)[log(C(ε)) / log(ε)], where C(ε) is the correlation integral.

[0064] Introducing a comprehensive indicator I 3 To evaluate superheat changes: I 3 =w 1 *L'(t c )+w 2 *L''(t c )+w 3 *std(ΔL)+w 4 *std(Δ²L)+w 5 *λ+w 6 *D+w 7 *E(cD 1 ); where w 1 ,w 2 ,...,w 7 is the weight coefficient, E(cD 1 ) is the energy of the first layer wavelet detail coefficient, t c is a time variable, indicating the cth time point; ΔL is the first-order difference sequence of quality level L; std represents the standard deviation; Δ²L is the second-order difference sequence of quality level L.

[0065] Set the threshold used to determine the superheat in step S25 , judge the superheat change according to the following rules: if I 3 > : Overheat judgment = "rising" elif I 3 <- : Overheat judgment = "down" elif |I 3 |≤ : Superheat judgment = "stable" else: Overheat judgment = "uncertain" Introduce a confidence index C 4 , used to quantify the reliability of the judgment results: C 4 =(R²+1)*(1-σ ΔL / σ ΔLmax )*(1-σ Δ²L / σ Δ²Lmax )*(1-|λ| / λ max )*(1-D / Dmax ), R² is the coefficient of determination of the polynomial fit, σ ΔL and σ Δ²L are the standard deviations of the first-order and second-order difference series, σ ΔLmax is the maximum allowable value of the standard deviation of the preset first-order difference sequence, σ Δ²Lmax is the maximum allowable value of the standard deviation of the preset second-order difference sequence, λ max is the maximum allowable value of the preset Lyapunov exponent, D max The maximum allowed value of the correlation dimension of the time series of the preset quality level.

[0066] Output the fourth factor judgment result, including: superheat change trend and judgment confidence C 4 .

[0067] This embodiment comprehensively considers time series analysis, differential series analysis, wavelet decomposition and nonlinear dynamic characteristics, and provides a more comprehensive and reliable quality level analysis.

[0068] S26 obtains the amount of electrolyte added in the time period, the time interval for each electrolyte addition and the corresponding electrolytic cell temperature, calculates the average amount added and the frequency of addition in the time period, analyzes the correlation between the amount added and the electrolytic cell temperature, and determines the change in superheat based on the average amount added, the frequency of addition and the temperature correlation, thereby obtaining the fifth factor determination result.

[0069] In this embodiment, during a preset time period In the period, the data of adding and removing electrolytes are obtained for n times, which is recorded as {(t i ,M i ,T i ),i=1,2,...,n}, where t i is the addition time, M i is the amount added, T i is the corresponding electrolytic cell temperature.

[0070] Calculate the addition and subtraction interval time series: Δt i =t i -t i-1 ,i=2,3,...,n. Use exponential smoothing method to calculate the moving average S(M i ): S(M i )=α*M i +(1-α)*S(M i-1 ), where α is the smoothing coefficient, 0<α<1.

[0071] Calculate the frequency of addition and extraction f: f=n / T. Use Fourier transform to analyze the periodicity of the amount of addition and extraction F(ω): F(ω)=Σ(M i *e -jωti),i=1,2,...,n, where ω is the angular frequency, j here represents an imaginary unit, and e is an exponential function.

[0072] Calculate the Pearson correlation coefficient ρ(M,T) between the amount added and the temperature: ρ(M,T)=cov(M,T) / (σ M *σ T ), M is the amount of addition and removal, T is the temperature of the electrolytic cell, cov(M,T) represents the covariance between M and T, σ M is the standard deviation of the amount added or taken, σ T is the standard deviation of temperature; Granger causality test is used to analyze the effect of addition and removal on temperature: T(t)=Σ(a i *T(ti))+Σ(b i *M(ti))+ε(t), where T(t) is the temperature at time t, a i and b i is the regression coefficient, and ε(t) is the error term.

[0073] Introducing a comprehensive indicator I 4 To evaluate superheat changes: I 4 =w 1 *S(M n )+w 2 *f+w 3 *|F(ω max )|+w 4 *ρ(M,T), where w 1 ,w 2 ,w 3 ,w 4 is the weight coefficient, ω max is the main frequency component.

[0074] Setting the threshold used to determine the superheat degree in step S26 , judge the superheat change according to the following rules: if I 4 > : Overheat judgment = "rising" elif I 4 <- : Overheat judgment = "down" elif |I 4 |≤ : Superheat judgment = "stable" else: Overheat judgment = "uncertain" Introducing the confidence index C 5, used to quantify the reliability of the judgment results: C 5 =(R²+1)*(1-σ Δt / σ Δtmax )*(|ρ(M,T)|+1) / 2, R² is the coefficient of determination of Granger causality test, σ Δt is the standard deviation of the addition and subtraction interval, σ Δtmax The maximum allowed value of the standard deviation of the preset addition and subtraction interval.

[0075] S27: Combining the first factor judgment result, the second factor judgment result, the third factor judgment result, the fourth factor judgment result, and the fifth factor judgment result to obtain a first analysis result.

[0076] In this embodiment, step S27 includes: S271 assigns weights to the first factor judgment result, the second factor judgment result, the third factor judgment result, the fourth factor judgment result, and the fifth factor judgment result respectively, and the size of the weight is determined according to the importance of each factor on the superheat; S272 quantifies the judgment result of each factor, and divides the superheat change trend into three states: rising, falling and stable, and assigns values ​​of 1, -1 and 0 respectively; S273 calculates a weighted average value, that is, multiplying the quantized value of each factor judgment result by its corresponding weight and then summing them; S274 sets thresholds α and β, where 0<α<β<1, and determines the superheat change trend based on the weighted average value: If the weighted average value is greater than β, it is judged that the superheat is on an upward trend; If the weighted average value is less than -β, it is judged that the superheat is on a downward trend; If the absolute value of the weighted average is less than α, it is judged that the superheat is in a stable trend; If the absolute value of the weighted average is between α and β, it is determined that the superheat change trend is unclear; S275 Combine the consistency of the judgment results of each factor to evaluate the credibility of the judgment results: Calculate the number of factors supporting the main trend and increase the credibility if it exceeds the preset threshold; If there are factor judgment results that are contrary to the main trend, the credibility will be reduced; S276 forms a first analysis result according to the superheat variation trend and the credibility of the judgment result, including the superheat variation trend, the degree of variation and the credibility of the judgment.

[0077] A specific example is used to illustrate: According to the influence of each factor on superheat, the following weights are assigned to the five factors: The first factor (alumina feeding amount): w1 =0.25; Second factor (Fe, Si, Ca content): w 2 =0.20; The third factor (number of flickering effects): w 3 =0.15; the fourth factor (quality level): w 4 =0.25; Fifth factor (amount of electrolyte added): w 5 =0.15.

[0078] The judgment results of each factor are quantified: the first factor: increase (1); the second factor: stable (0); the third factor: decrease (-1); the fourth factor: increase (1); the fifth factor: increase (1).

[0079] Calculate the weighted average: 0.25*1+0.20*0+0.15*(-1)+0.25*1+0.15*1=0.50.

[0080] Set the threshold α=0.3, β=0.6. Since the weighted average value 0.50 is between α and β, the superheat change trend is judged to be "increasing, but not obvious".

[0081] The credibility of the evaluation results is as follows: the number of factors supporting the upward trend: 3 (the first, fourth, and fifth factors); the number of factors supporting the downward trend: 1 (the third factor); the number of factors supporting the stable trend: 1 (the second factor). After comprehensive consideration, a medium to high credibility is given, such as 75%.

[0082] The first analysis result is formed: superheat change trend: rising; change degree: slight; credibility of judgment: 75%.

[0083] Specifically, Figure 4 As shown, in one embodiment of the present invention, step S3 includes: S31 In each time period, an electrolyte sample is obtained, the content and molecular ratio of Li, K, Ca, and Mg in the sample are analyzed, and the aluminum oxide concentration is calculated.

[0084] Specifically, in this embodiment, the aluminum electrolytic cell is analyzed for 7 days, and an electrolyte sample is taken once a day. Sampling is performed regularly every day, and about 100g of electrolyte sample is taken from a specified position of the electrolytic cell, placed in a container of the corresponding date, and the temperature of the electrolytic cell at the time of sampling is recorded. The samples are analyzed using an X-ray fluorescence spectrometer (XRF). The analysis results are recorded every day, including: Li content (wt%), K content (wt%), Ca content (wt%), Mg content (wt%), and alumina concentration (wt%). The molecular ratio R = NaF / (AlF 3 +CaF 2 +MgF 2), NaF is sodium fluoride, AlF 3 Aluminum fluoride, CaF 2 is calcium fluoride, MgF 2 For magnesium fluoride, NaF and AlF are used in calculating the molecular ratio R. 3 , CaF 2 MgF 2 to calculate the concentration.

[0085] S32 calculates the crystallization temperature according to the contents and molecular ratios of Li, K, Ca, and Mg, and the alumina concentration, and calculates the initial superheat based on the electrolyte temperature and the crystallization temperature.

[0086] Specifically, step S32 includes: S321 Analyze the lithium fluoride content, magnesium fluoride content, calcium fluoride content, potassium fluoride content, molecular ratio and aluminum oxide concentration, and calculate the initial crystallization temperature:

[0087] In the formula, is the crystallization temperature, is the base temperature, is the aluminum oxide concentration, is the lithium fluoride content, is the magnesium fluoride content, is the calcium fluoride content, is the potassium fluoride content, R is the actual molecular ratio, is the base molecular ratio, a, b, c, d, e, f, g are adjustment coefficients; S322 Calculate preliminary superheat:

[0088] In the formula, is the initial superheat, is the electrolyte temperature.

[0089] In this embodiment, the basic temperature is 1094℃, a is 4.6, b is 7.8, c is 6, d is 2.4, e is 3, f is 60, g is 30, and the standard molecular ratio is is 3.

[0090] S33 calculates the actual current efficiency of the electrolytic cell based on the data of the two aluminum inventory measurements and the actual measured efficiency of the electrolytic cell. Taking the actual current efficiency as a benchmark, obtains the alumina discharge amount within the time period. The actual alumina discharge amount is calculated based on the average value of the alumina discharge amount within the time period and the alumina concentration. The theoretical aluminum output is calculated based on the reaction equation of alumina.

[0091] Specifically, step S33 includes: S331 Obtain the aluminum quantity data of two inventory and the actual measurement efficiency in the corresponding time period, and calculate the actual current efficiency η of the electrolytic cell t ; S332 Obtain the alumina feed quantity data within the time period and calculate the weekly average feed quantity Q ave ; S333 According to the purity of alumina and weekly average feed quantity Q ave , calculate the actual alumina feed amount Q t :

[0092] In the formula, Indicates the ratio of the actual aluminum oxide entering the electrolytic cell to the theoretical aluminum oxide; in this embodiment, Take 0.95. Alumina purity It is 98%.

[0093] S334 Based on the reaction equation of alumina 2Al 2 O 3 +3C→4Al+3CO 2 , calculate the theoretical aluminum output m t :

[0094] in, is the stoichiometric coefficient, is the molar mass of aluminum, is the molar mass of aluminum oxide. In this embodiment, is 1, is 27, is 102.

[0095] S34 collects electrolyte in and out data of the electrolytic cell, subanode aluminum level data, Fe and Si content of the electrolytic cell, and cell temperature change data according to a time period.

[0096] Specifically, electrolyte in and out data collection: Record the time, amount and composition of each electrolyte addition. Record the time, amount and composition of each electrolyte discharge. Calculate the net increase in electrolyte and the average addition frequency.

[0097] Data collection of sub-polar aluminum level: Use an acoustic depth sounder to regularly measure the sub-polar aluminum level. Record the measurement time and the corresponding level height. Calculate the level change rate and fluctuation range.

[0098] Fe and Si content data collection: Collect electrolyte samples regularly and use a spectrometer to measure the Fe and Si contents. Record the sampling time and the corresponding Fe and Si content values. Calculate the average value, change trend and fluctuation range of Fe and Si contents.

[0099] Tank temperature change data collection: Use a thermocouple or infrared thermometer to regularly measure the temperature of the electrolytic tank. Record the measurement time and the corresponding temperature value. Calculate the average temperature, change trend and fluctuation range.

[0100] S35 establishes an analysis model, which includes a data preprocessing module, a feature extraction module, a correlation analysis module, a trend analysis module, an anomaly detection module, a prediction module, and a comprehensive scoring module.

[0101] S36 takes the actual current efficiency, theoretical aluminum output, electrolyte inlet and outlet data of the electrolytic cell, sub-electrode aluminum level data, Fe and Si content of the electrolytic cell and cell temperature change data as independent parameters, and inputs them into the analysis model. The data preprocessing module is used to clean and standardize the data. The feature extraction module is used to extract key features from the independent parameters. The key features are then input into the correlation analysis module, trend analysis module, anomaly detection module and prediction module respectively to analyze the correlation between independent parameters, identify the change trend of each independent parameter, identify abnormal data points and predict future trends. The outputs of the correlation analysis module, trend analysis module, anomaly detection module and prediction module are all input into the comprehensive scoring module to obtain big data analysis results, including the electrolytic cell status score, superheat change trend prediction and anomaly analysis.

[0102] In this embodiment, step S36 includes: S361 uses the data preprocessing module to clean and standardize the input independent parameters to obtain a standardized data set.

[0103] Specifically, the Z-score method is used, and data points with |Z|>3 are considered outliers for outlier detection. Forward filling is used for missing values. Min-Max normalization is used to scale the data to the [0,1] interval.

[0104] S362 uses a feature extraction module to extract key features from standardized data sets, including time series features, statistical features, and frequency domain features.

[0105] In this embodiment, the feature extraction module uses a CNN network for feature extraction. The CNN network has a simple structure. An example structure is as follows: Convolutional layer 1: 32 filters, kernel size 3, stride 1, ReLU activation; Pooling layer 1: max pooling, pooling size 2; Convolutional layer 2: 64 filters, kernel size 3, stride 1, ReLU activation; Pooling layer 2: max pooling, pooling size 2; Fully connected layers: 128 nodes, ReLU activation.

[0106] Specifically, time series features: use the sliding window method to extract mean, variance, skew, etc. Statistical features: calculate skewness, kurtosis, quartiles, etc. Frequency domain features: use fast Fourier transform (FFT) to extract frequency components.

[0107] S363 Input key features into the correlation analysis module and calculate the Pearson correlation coefficient matrix R between features corr .

[0108] In this embodiment, the module can be analyzed using a 6-layer autoencoder, where the encoder has 3 layers, the number of nodes is 64, 32, and 16, respectively, and ReLU is activated, and the decoder has 3 layers, the number of nodes is 16, 32, and 64, respectively, and ReLU is activated. According to the calculation formula of the autoencoder and the Pearson correlation coefficient, the Pearson correlation coefficient matrix R between the features is obtained corr .

[0109] S364 Use the trend analysis module to perform time series decomposition and trend identification on key features to obtain trend indicator T ind .

[0110] Specifically, in this module, STL is used to decompose the time series, and then the Mann-Kendall trend test is used. A two-layer LSTM structure is built in to implement time series analysis to obtain the trend indicator T ind .

[0111] S365 uses the anomaly detection module to detect anomalies of key features based on the local anomaly factor algorithm and generate anomaly index A ind .

[0112] Specifically, the module adopts the VAE structure, including: 2 layers of encoders, the number of nodes is 64 and 32, 16-dimensional latent space, 2 layers of decoders, the number of nodes is 32 and 64. The algorithm adopts the local anomaly factor algorithm, namely the LOF algorithm, and its process is: calculate the k nearest neighbors, calculate the local reachable density, calculate the LOF score based on the results of the k nearest neighbors and the local reachable density distribution, and finally use the LOF score as the anomaly indicator A. ind .

[0113] S366 uses the prediction module, which includes the ARIMA network, to predict the future trend of key features and obtain the prediction result P res .

[0114] Specifically, this module uses the ARIMA-LSTM hybrid model. ARIMA part: autoregression order p, difference order d, moving average order q; LSTM part: 2 layers of LSTM, 64 units per layer; connection layer: 32 nodes, ReLU activation; output layer: nodes with the number of prediction steps. Among them, ARIMA uses the Box-Jenkins method to determine parameters, and LSTM uses the Adam optimizer. The key features are input into the ARIMA-LSTM hybrid model, that is, the output prediction probability sequence is obtained to obtain the prediction result P res .

[0115] S367 R corr , T ind , A ind and P res Input the comprehensive scoring module and obtain the electrolytic cell status score S through weighted summation. score , Superheat change trend prediction T pred and Abnormal Analysis Report A report .

[0116] S37 combines the preliminary superheat and the big data analysis results to obtain a second analysis result.

[0117] In this embodiment, step S37 includes: S371 Construct the comprehensive evaluation function H(x) of superheat:

[0118] In the formula, is the initial superheat, w 1 、w 2 、w 3 、w 4 is the weight coefficient, f(A report ) is the quantitative function of the abnormal analysis report; S372 Set the superheat threshold value θ 1 ,θ 2 and θ 3 , where θ 1 <θ 2 <θ 3 ; S373 determines the superheat status according to the value of H(x): If H(x)<θ 1 , it is judged that the superheat decreases; If θ 1 ≤H(x)<θ 2 , then the superheat is determined to be stable; If θ 2 ≤H(x)<θ 3 , it is judged that the superheat increases; If H(x)≥θ 3 , then it is judged that the superheat is disordered; S374 Generate corresponding control strategy recommendations based on the superheat state judgment result, and output the superheat state judgment result and the control strategy recommendations as the second analysis result.

[0119] Specifically, in one embodiment of the present invention, step S4 includes: S41 Design the fusion function F(x,y), where x represents the first analysis result, y represents the second analysis result, and introduces the dynamic weight coefficient α t and β t , used to balance the two analysis results, and define the comprehensive evaluation index Z=F(α t x,β t y).

[0120] Specifically, Z = F(α t x,β t y)=α t ·f 1 (x)+β t ·f 2 (y)+γ·g(x,y), where f 1 (x) and f 2 (y) are nonlinear transformation functions of x and y, respectively, used to capture their respective information. g(x,y) is an interaction term, which represents the mutual influence between x and y. γ is the weight coefficient of the interaction term.

[0121] S42 Establish a weight adaptive algorithm based on historical data and dynamically adjust α t and β t .

[0122] In this embodiment, the gradient descent method is used to dynamically adjust α t and β t The weight update formula is: α t+1 =α t -η·∂E / ∂α t +λ·(α t -α 0 ); β t+1 =β t -η·∂E / ∂β t +λ·(β t -β 0 ); Where: E is the prediction error function; η is the learning rate; λ is the regularization parameter; α 0 and β 0 is the initial weight.

[0123] S43 constructs a multi-level decision tree model, takes the Z value as the input of the decision tree, designs multiple decision nodes, including superheat state, change rate, and stability, and sets the superheat judgment result and confidence level at the leaf node.

[0124] Specifically, the C4.5 algorithm is used to recursively construct a decision tree, and the decision tree structure is as follows: root node, Z value; internal nodes, superheat state, change rate, and stability; leaf nodes, superheat judgment results and confidence.

[0125] S44 converts the Z value into a fuzzy set, uses the Mamdani reasoning method to obtain the fuzzy output, and defuzzifies it through the centroid method to obtain the superheat judgment result.

[0126] Specifically, the specific process of step S44 is as follows: 1. Define the fuzzy set membership function of Z value : , e represents the exponential function; adaptive adjustment parameter z 0 and σ: Use the sliding window method to collect the most recent N historical Z values ​​and calculate the mean of these N values ​​as z 0 , calculate the standard deviation of these N values ​​as σ, and convert the current Z value into a fuzzy set: Calculate The value of , and the membership degree of Z value is obtained.

[0127] 2. Define language variables: Z:{LOW,MEDIUM,HIGH}; superheat:{DECREASING,STABLE,INCREASING}.

[0128] Construct fuzzy rules: R1:IF Z is LOW THEN superheat is DECREASING; R2:IF Z is MEDIUM THEN superheat is STABLE; R3:IF Z is HIGH THEN superheat is INCREASING; Define the fuzzy set membership function for each linguistic variable: Use Gaussian function to define LOW, MEDIUM, HIGH; Define DECREASING, STABLE, and INCREASING using trigonometric functions.

[0129] 3. Calculate the trigger strength of each rule : for i=1,2,3. Calculate the output fuzzy set of each rule: B i'(y)=min(ω i ,μ Bi(y) ), where B i is the output fuzzy set of rule i, μ Bi(y) is the output fuzzy set B of rule i i Membership degree at point y. Aggregate the output of all rules: B(y)=max(B 1 '(y),B 2 '(y),B 3 '(y)).

[0130] 4. Use the centroid method to calculate the clear output y*: y*=∫y·B(y)dy / ∫B(y)dy, where B(y) represents the membership of the aggregated output fuzzy set at y. Discretize the output domain into n points and calculate the membership of each point B(y i ), use the numerical integration method to calculate y*: y*=Σ(y i ·B(y i )) / ΣB(y i ), y i is the i-th discretization point in the output domain.

[0131] 5. Map y* to language description: If y* <threshold 1 , output "superheat reduction"; if threshold 1 ≤y* <threshold 2 , output "superheat stable"; if y*≥threshold 2 , output "superheat increases". Calculate the output membership confidence: confidence = max(B(y)).

[0132] 6. Result output: superheat status: {DECREASING, STABLE, INCREASING}; numerical superheat index: y*; judge the credibility of the result.

[0133] S45 integrates the outputs of decision tree and fuzzy reasoning, uses ensemble learning method, conducts voting, and obtains the final superheat judgment result.

[0134] Specifically, the Adaboost algorithm is used for ensemble learning, and the decision tree and fuzzy reasoning are regarded as two base learners. The output of the decision tree is recorded as: DT result (Superheat state), DT conf (confidence); fuzzy reasoning is recorded as: FL result (Superheat state), FL memb (Degree of membership).

[0135] Maps superheat states to numeric values: Decreasing: -1; Stable: 0; Increasing: 1; Chaotic: 2.

[0136] Calculate decision tree weight WT DT : WT DT =DT conf *(1+|DT result |) / 3.

[0137] Calculate the fuzzy inference weight WT FL : WT FL =FL memb *(1+|FL result |) / 3.

[0138] Calculate the weighted score Score: Score=(WT DT *DT result +WT FL *FL result ) / (WT DT +WT FL ).

[0139] Use an exponential moving average (EMA) to smooth the score: EMA Score (t)=α*Score(t)+(1-α)*EMA Score (t-1), where EMA Score (t) is the exponential moving average smoothed score at time point t, Score(t) is the unsmoothed weighted score at time point t, EMA Score (t-1) is the score after exponential moving average smoothing at time point t-1, and α is the smoothing factor, which is between 0.2 and 0.3.

[0140] The final score is calculated according to the voting score formula: Final Score =tanh(EMA Score *(1+δ*ΔTrend))*(1+ε*Consistency).

[0141] Where tanh() is the hyperbolic tangent function, which is used to compress the score to the range of [-1,1]. δ is the trend adjustment coefficient. ΔTrend is the trend change rate of the last N time points, that is, the time point of the N data collections closest to time point t. ε is the consistency adjustment coefficient. Consistency is the consistency index of the decision tree and fuzzy reasoning results.

[0142] ΔTrend calculation method: ΔTrend=(EMA Score (t)-EMA Score (tN)) / N, EMA Score (tN) is the exponential moving average score at the Nth time point before the current time point t.

[0143] Consistency calculation method: Consistency=1-|DT result -FL result | / 3.

[0144] According to Final Score The value of determines the superheat state: If Final Score <-0.6, it is judged as "superheat decreases significantly"; if -0.6≤Final Score <-0.2, it is judged as "superheat slightly reduced"; if -0.2≤Final Score ≤0.2, it is judged as "superheat stable"; if 0.2 <Final Score ≤0.6, it is judged as "slight increase in superheat"; if Final Score >0.6, it is judged as "superheat increases significantly".

[0145] Confidence calculation: Confidence=(|Final Score |+Consistency) / 2.

[0146] S46 selects a control strategy from the preset strategy library based on the final superheat judgment result and generates operation suggestions.

[0147] Specifically, the preset strategy library can be a hierarchical strategy library, with the first-level classification: significantly reduced, slightly reduced, stable, slightly increased, and significantly increased overheating; the second-level classification: short-term strategy, medium-term strategy, and long-term strategy. Each strategy includes: strategy ID, applicable conditions, operation steps, expected effects, and risk assessment.

[0148] According to Final Score Select the first-level classification, select the second-level classification based on Confidence and trend duration, and generate specific operation steps based on the selected strategy.

[0149] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent analysis method for superheat of an aluminum electrolytic cell, characterized in that: include: S1 sets a time period, and collects multiple operating parameters of the aluminum electrolytic cell according to the time period, including alumina-related parameters, element content, electrolyte-related parameters, and electrolytic cell operating parameters; alumina-related parameters include alumina feed rate and alumina concentration; element content includes Fe content, Si content, Ca content, Li content, K content, and Mg content; electrolyte-related parameters include electrolyte temperature, electrolyte composition, and electrolyte addition and removal amount; electrolytic cell operating parameters include the number of flicker effects, quality level, electrolytic cell temperature, and electrolytic cell measurement efficiency; S2 uses a multi-factor trend judgment method to analyze multiple operating parameters and obtain a first analysis result; S3 uses a big data judgment method to analyze multiple operating parameters and obtain a second analysis result; S4 combines the first analysis result and the second analysis result and obtains the final superheat analysis result through a comprehensive judgment method.

2. The intelligent analysis method for superheat of an aluminum reduction cell according to claim 1, characterized in that: Step S2 includes: S21 Determine multi-factor indicators, including alumina feed quantity indicator, Fe, Si, Ca content indicator, scintillation effect number indicator, quality level indicator, electrolyte addition and removal indicator; S22 obtains the number of alumina feeding times within a time period, and records the corresponding alumina feeding amount, calculates the slope between every two adjacent alumina feeding amounts, calculates the rate of change of the slope, performs statistics on the rate of change of the slope within the time period, calculates statistical indicators, judges the change of superheat according to the statistical indicators, and obtains the judgment result of the first factor; wherein the statistical indicators include the average rate of change, the standard deviation of the rate of change, the proportion of positive rate of change, and the proportion of negative rate of change; S23 obtains data on the contents of Fe, Si and Ca within a time period, performs time series analysis on the three elements respectively, obtains slopes by fitting the data, calculates coefficient of variation CV based on the slope, calculates correlation coefficients between Fe and Si, and determines changes in superheat based on the slope, coefficient of variation CV and correlation coefficient to obtain determination results of the second factor; S24: obtaining the number of flicker effects within the time period, calculating a short-term moving average and a medium-term moving average, calculating a frequency distribution of the number of flicker effects within the time period, drawing a CUSUM chart according to the number of flicker effects within the time period, judging the degree of overheating according to the short-term moving average, the medium-term moving average, the frequency distribution, and the CUSUM chart, evaluating the confidence level based on the consistency of the judgments of the short-term moving average and the medium-term moving average, and if the results of the frequency analysis and the CUSUM chart are consistent, then increasing the confidence level, and obtaining the judgment result of the third factor based on the confidence level; S25: obtaining quality level measurement data within a time period, fitting the quality level measurement data within the time period, obtaining a slope, calculating the first-order difference and the second-order difference of two adjacent quality level measurement data according to the measurement time of the quality level measurement data, forming a first-order difference sequence and a second-order difference sequence, judging the change of superheat according to the slope, the first-order difference sequence and the second-order difference sequence, and obtaining the judgment result of the fourth factor; S26: obtaining the amount of electrolyte added and taken in the time period, the time interval for each electrolyte addition and the corresponding electrolytic cell temperature, calculating the average amount added and taken and the frequency of addition and taken in the time period, analyzing the correlation between the amount added and taken and the electrolytic cell temperature, judging the change of superheat according to the average amount added, the frequency of addition and taken and the temperature correlation, and obtaining the judgment result of the fifth factor; S27: Combining the first factor judgment result, the second factor judgment result, the third factor judgment result, the fourth factor judgment result, and the fifth factor judgment result to obtain a first analysis result.

3. The intelligent analysis method for superheat of an aluminum reduction cell according to claim 2, characterized in that: Step S27 includes: S271 assigns weights to the first factor judgment result, the second factor judgment result, the third factor judgment result, the fourth factor judgment result, and the fifth factor judgment result respectively, and the size of the weight is determined according to the importance of each factor on the superheat; S272 quantifies the judgment result of each factor, and divides the superheat change trend into three states: rising, falling and stable, and assigns values ​​of 1, -1 and 0 respectively; S273 calculates a weighted average value, that is, multiplying the quantized value of each factor judgment result by its corresponding weight and then summing them; S274 sets thresholds α and β, where 0<α<β<1, and determines the superheat change trend based on the weighted average value: If the weighted average value is greater than β, it is judged that the superheat is on an upward trend; If the weighted average value is less than -β, it is judged that the superheat is on a downward trend; If the absolute value of the weighted average is less than α, it is judged that the superheat is in a stable trend; If the absolute value of the weighted average is between α and β, it is determined that the superheat change trend is unclear; S275 Combine the consistency of the judgment results of each factor to evaluate the credibility of the judgment results: Calculate the number of factors supporting the main trend and increase the credibility if it exceeds the preset threshold; If there are factor judgment results that are contrary to the main trend, the credibility will be reduced; S276 forms a first analysis result according to the superheat variation trend and the credibility of the judgment result, including the superheat variation trend, the degree of variation and the credibility of the judgment.

4. The intelligent analysis method for superheat of an aluminum electrolytic cell according to claim 2, characterized in that: Step S3 includes: S31 In each time period, obtaining electrolyte samples, analyzing the contents and molecular ratios of Li, K, Ca, and Mg in the samples, and calculating the concentration of aluminum oxide; S32 calculates the crystallization temperature based on the content and molecular ratio of Li, K, Ca, Mg and the concentration of alumina, and calculates the initial superheat based on the electrolyte temperature and the crystallization temperature; S33 calculates the actual current efficiency of the electrolytic cell based on the data of the two aluminum inventory and the actual measured efficiency of the electrolytic cell, obtains the amount of alumina discharged during the time period based on the actual current efficiency, calculates the actual amount of alumina discharged based on the average value of the amount of alumina discharged during the time period and the alumina concentration, and calculates the theoretical amount of aluminum output based on the reaction equation of alumina; S34 collects electrolyte inlet and outlet data of the electrolytic cell, subaluminum level data, Fe and Si content of the electrolytic cell, and cell temperature change data according to a time period; S35 establishing an analysis model, the analysis model including a data preprocessing module, a feature extraction module, a correlation analysis module, a trend analysis module, an anomaly detection module, a prediction module, and a comprehensive scoring module; S36 takes the actual current efficiency, theoretical aluminum output, electrolyte in and out data of the electrolytic cell, subaqueous aluminum level data, Fe and Si content of the electrolytic cell and cell temperature change data as independent parameters, inputs them into the analysis model, uses the data preprocessing module to clean and standardize the data, uses the feature extraction module to extract key features from the independent parameters, and then inputs the key features into the correlation analysis module, trend analysis module, anomaly detection module and prediction module respectively, respectively analyzes the correlation between independent parameters, identifies the change trend of each independent parameter, identifies abnormal data points and predicts future trends, and inputs the outputs of the correlation analysis module, trend analysis module, anomaly detection module and prediction module into the comprehensive scoring module to obtain big data analysis results, including electrolytic cell status score, superheat change trend prediction and anomaly analysis; S37 combines the preliminary superheat and the big data analysis results to obtain a second analysis result.

5. The intelligent analysis method for superheat of an aluminum electrolytic cell according to claim 4, characterized in that: Step S32 includes: S321 Analyze the lithium fluoride content, magnesium fluoride content, calcium fluoride content, potassium fluoride content, molecular ratio and aluminum oxide concentration, and calculate the initial crystallization temperature: ； In the formula, is the crystallization temperature, is the base temperature, is the aluminum oxide concentration, is the lithium fluoride content, is the magnesium fluoride content, is the calcium fluoride content, is the potassium fluoride content, R is the actual molecular ratio, is the base molecular ratio, a, b, c, d, e, f, g are adjustment coefficients; S322 Calculate preliminary superheat: ； In the formula, is the initial superheat, is the electrolyte temperature.

6. The intelligent analysis method for superheat of an aluminum electrolytic cell according to claim 5, characterized in that: Step S33 includes: S331 Obtain the aluminum quantity data of two inventory and the actual measurement efficiency in the corresponding time period, and calculate the actual current efficiency η of the electrolytic cell t ; S332 Obtain the alumina feed quantity data within the time period and calculate the weekly average feed quantity Q ave ; S333 According to the purity of alumina and the average weekly feed quantity Q ave , calculate the actual alumina feed amount Q t : ； In the formula, It indicates the ratio coefficient of the actual alumina entering the electrolytic cell to the theoretical alumina; S334 Calculate the theoretical aluminum output m based on the aluminum oxide reaction equation 2Al2O3+3C→4Al+3CO2 t : ； in, is the stoichiometric coefficient, is the molar mass of aluminum, is the molar mass of aluminum oxide.

7. The intelligent analysis method for superheat of an aluminum reduction cell according to claim 6, characterized in that: Step S36 includes: S361 uses a data preprocessing module to perform data cleaning and standardization on the input independent parameters to obtain a standardized data set; S362 uses a feature extraction module to extract key features from the standardized data set, including time series features, statistical features, and frequency domain features; S363 Input key features into the correlation analysis module and calculate the Pearson correlation coefficient matrix R between features corr ; S364 Use the trend analysis module to perform time series decomposition and trend identification on key features to obtain trend indicator T ind ; S365 uses the anomaly detection module to detect anomalies of key features based on the local anomaly factor algorithm and generate anomaly index A ind ; S366 uses the prediction module, which includes the ARIMA network, to predict the future trend of key features and obtain the prediction result P res ; S367 R corr , T ind , A ind and P res Input the comprehensive scoring module and obtain the electrolytic cell status score S through weighted summation. score , Superheat change trend prediction T pred and Abnormal Analysis Report A report .

8. The intelligent analysis method for superheat of an aluminum reduction cell according to claim 7, characterized in that: Step S37 includes: S371 Construct the comprehensive evaluation function H(x) of superheat: ； In the formula, is the initial superheat, w1, w2, w3, w4 are weight coefficients, f(A report ) is a quantitative function for abnormal analysis report; S372 sets superheat thresholds θ1, θ2 and θ3, where θ1<θ2<θ3; S373 determines the superheat status according to the value of H(x): If H(x)<θ1, it is judged that the superheat is reduced; If θ1≤H(x)<θ2, the superheat is determined to be stable; If θ2≤H(x)<θ3, it is judged that the superheat increases; If H(x)≥θ3, it is judged that the superheat is disordered; S374 Generate corresponding control strategy recommendations based on the superheat state judgment result, and output the superheat state judgment result and the control strategy recommendations as the second analysis result.

9. The intelligent analysis method for superheat of an aluminum reduction cell according to claim 4, characterized in that: Step S4 includes: S41 Design the fusion function F(x,y), where x represents the first analysis result, y represents the second analysis result, and introduces the dynamic weight coefficient α t and β t , used to balance the two analysis results, and define the comprehensive evaluation index Z=F(α t x,β t y); S42 Establish a weight adaptive algorithm based on historical data and dynamically adjust α t and β t ; S43 constructs a multi-level decision tree model, uses the Z value as the input of the decision tree, designs multiple decision nodes, including superheat state, change rate, and stability, and sets the superheat judgment result and confidence level at the leaf node; S44 converts the Z value into a fuzzy set, uses the Mamdani reasoning method to obtain the fuzzy output, and defuzzifies it through the centroid method to obtain the superheat judgment result; S45 integrates the outputs of decision tree and fuzzy reasoning, uses ensemble learning method, conducts voting, and obtains the final superheat judgment result; S46 selects a control strategy from the preset strategy library based on the final superheat judgment result and generates operation suggestions.

Images