A method and device for predicting the remaining life of an aluminum electrolytic cell

Abstract

The present invention discloses a method and device for predicting the remaining life of an aluminum electrolytic cell. The method comprises the following steps: screening out degradation-related process parameters of the aluminum electrolytic cell and obtaining degradation data of a source domain electrolytic cell and a target domain electrolytic cell; dividing the degradation states of the source domain aluminum electrolytic cell over its entire life cycle to obtain degradation state labels corresponding to the degradation data of the source domain aluminum electrolytic cell; determining the current degradation state of the target domain aluminum electrolytic cell and obtaining the results of the existing degradation state division to obtain degradation state labels corresponding to the degradation data of the target domain aluminum electrolytic cell; using the degradation data and corresponding degradation state labels of the source domain aluminum electrolytic cell and the target domain aluminum electrolytic cell to train a feature migration model, migrate all degradation data of the source domain aluminum electrolytic cell, train a model for predicting the remaining life of the aluminum electrolytic cell using the migrated degradation data, and predict the remaining life of the target domain aluminum electrolytic cell. The present invention enables accurate prediction of the remaining life of the target domain aluminum electrolytic cell.

Description

Technical Field

[0001] The present invention relates to the technical field of aluminum electrolysis, and in particular to a method and device for predicting the remaining life of an aluminum electrolysis cell. Background Art

[0002] Aluminum reduction cells are essential equipment in aluminum electrolysis production, and their remaining lifespan is essential for ensuring safe and stable production. As large-scale equipment, they carry high capital and maintenance costs, and downtime for major repairs can also result in significant economic losses. Therefore, predicting the remaining lifespan of aluminum reduction cells and coordinating their overhaul and maintenance are significant challenges that have drawn widespread attention and research within the aluminum reduction industry.

[0003] According to relevant literature, remaining life prediction methods can be divided into two categories: physical model-based methods and data-driven methods. Physical model-based prediction methods focus on constructing degradation mechanism models, requiring professional and extensive theoretical support and long-term experimental verification, making it difficult to obtain accurate models. Data-driven prediction methods, on the other hand, do not require extensive prior knowledge and instead complete remaining life prediction by finding a mapping relationship between equipment operation monitoring data and remaining life. With the rapid development of information technology and data technology, data-driven methods are becoming the mainstream method for exploring the degradation patterns of industrial equipment.

[0004] Most current data-driven methods assume that the training and test sets follow the same data distribution. However, in actual production, when the training and test sets come from different work areas, differences in set voltages, operating modes, and daily management can lead to differences in the data distribution of process parameter data generated by aluminum electrolytic cell operations. Even within the same work area, data distributions can vary between different aluminum electrolytic cells due to the influence of many external factors, such as noise. This inconsistency in the data distribution of the training and test sets can lead to poor generalization and low prediction accuracy in the remaining life prediction model. Summary of the Invention

[0005] The present invention provides a method and device for predicting the remaining life of an aluminum electrolysis cell, which is used to predict the remaining life of an aluminum electrolysis cell. The technical solution is as follows:

[0006] In one aspect, a method for predicting the remaining life of an aluminum electrolysis cell is provided, the method comprising:

[0007] S1. Screen out the process parameters related to aluminum electrolytic cell degradation and obtain the degradation data of the source domain electrolytic cell and the target domain electrolytic cell;

[0008] S2. Dividing the degradation states of the source domain aluminum electrolysis cell over its entire life cycle to obtain degradation state labels corresponding to the degradation data of the source domain aluminum electrolysis cell;

[0009] S3. Determine the current degradation state of the aluminum electrolytic cell in the target domain and obtain the results of existing degradation state division to obtain a degradation state label corresponding to the degradation data of the aluminum electrolytic cell in the target domain;

[0010] S4. Using partial degradation data and corresponding degradation state labels of the source domain aluminum electrolysis cell, and all degradation data and corresponding degradation state labels of the target domain aluminum electrolysis cell, to train a feature transfer model;

[0011] S5. Using the feature migration model to migrate all degradation data of the source domain aluminum electrolytic cell, and using the migrated degradation data to train a remaining life prediction model for the aluminum electrolytic cell;

[0012] S6. Use the aluminum electrolysis cell remaining life prediction model to predict the remaining life of the aluminum electrolysis cell in the target domain.

[0013] Optionally, the step S1 of screening out degradation-related process parameters of the aluminum electrolytic cell and obtaining degradation data of the source domain electrolytic cell and the target domain electrolytic cell specifically includes:

[0014] Normalizing the source domain aluminum electrolytic cell full life cycle monitoring data;

[0015] Using the two indicators of monotonicity and correlation to screen out degradation-related process parameters of the aluminum electrolysis cell, and obtain degradation data of the source domain aluminum electrolysis cell, wherein the degradation data of the source domain aluminum electrolysis cell includes full life cycle data of the degradation-related process parameters;

[0016] Obtain the degradation data of the target domain electrolytic cell, wherein the degradation data of the target domain electrolytic cell includes the existing data of the degradation-related process parameters of the target domain aluminum electrolytic cell from the beginning of operation to the current state, and the degradation-related process parameters of the target domain aluminum electrolytic cell include the same degradation-related process parameters as those of the source domain electrolytic cell.

[0017] Optionally, the step S2 divides the degradation state of the source domain aluminum electrolytic cell over its entire life cycle to obtain degradation state labels corresponding to the degradation data of the source domain aluminum electrolytic cell, specifically including:

[0018] The degradation data of the source domain electrolytic cell is input into the fuzzy c-means clustering FCM model, and the number of clusters is set to 5. The dividing lines of each degradation state are determined according to the clustering results, and the degradation state of the source domain aluminum electrolysis from the start of operation to the end of life is divided, which corresponds to the healthy operation state, initial degradation state, slow degradation state, rapid degradation state and severe degradation state of the aluminum electrolytic cell respectively, and then the degradation state label corresponding to the degradation data of the source domain aluminum electrolytic cell is obtained.

[0019] Optionally, the step S3 of determining the current degradation state of the aluminum electrolytic cell in the target domain and obtaining the result of existing degradation state division to obtain the degradation state label corresponding to the degradation data of the aluminum electrolytic cell in the target domain specifically includes:

[0020] The degradation data of the aluminum electrolytic cell in the target domain is input into the FCM model, the number of clusters is set to M (M≤5), and the silhouette coefficient values corresponding to different M are calculated. The calculation formula is:

[0021]

[0022]

[0023] Where S represents the total silhouette coefficient, n t is the number of samples in the target domain, s(i) represents the silhouette coefficient of the i-th sample point, a(i) represents the cohesion of the sample point, which is the average distance between the sample point i and other sample points in the same category, and b(i) represents the separation, which is the minimum value of the average distance between the sample point i and the sample points of other categories. The larger the silhouette coefficient, the better the clustering effect. The M value corresponding to the maximum silhouette coefficient is counted as C, and C is the optimal number of clusters. It is determined that the target domain aluminum electrolytic cell is currently in the C-th degradation state, and according to the clustering result when the number of clusters is C, the first C degradation states of the target domain aluminum electrolytic cell are divided, and then the degradation state label corresponding to the degradation data of the target domain aluminum electrolytic cell is obtained.

[0024] Optionally, the step S4 utilizes partial degradation data and corresponding degradation state labels of the source domain aluminum electrolytic cell, and all degradation data and corresponding degradation state labels of the target domain aluminum electrolytic cell to train a feature migration model, specifically including:

[0025] All degradation data and corresponding degradation state labels of the target domain aluminum electrolytic cell, as well as the degradation data and corresponding degradation state labels of the source domain aluminum electrolytic cell corresponding to the existing degradation state of the target domain aluminum electrolytic cell, are input into the improved JDA migration algorithm. While retaining the original attributes of each data, the marginal distribution difference of the two data and the conditional distribution difference of the same degradation state are reduced, and finally a migration matrix is obtained. The migration matrix is the feature migration model.

[0026] The improved JDA migration algorithm includes three steps: preserving the original attributes of the source and target domain data, minimizing the marginal distribution of the source and target domains, and minimizing the conditional distribution of the source and target domains. Specifically, the algorithm includes:

[0027] The original attributes of the data are retained by maximizing the variance of the transformed data, X=[x1,x2,...,x n] is the matrix after the source domain and target domain data are merged, n=n s +n t , where n s is the number of source domain samples, n t is the number of samples in the target domain, H is the centralization matrix, then the covariance matrix is XHX T , assuming A is the migration matrix, the problem of retaining the original attributes of the data can be transformed into the following expression:

[0028] max tr(A T XHX T A)

[0029] The maximum mean difference (MMD) distance is used to measure the difference in the marginal probability distribution of the two data distributions. The problem of minimizing the marginal distribution between the source domain and the target domain is transformed into minimizing the MMD distance after mapping. The expression is as follows:

[0030]

[0031] By introducing the kernel method, the above formula is transformed into the following expression:

[0032]

[0033] Where M0 is an MMD matrix, and its expression is

[0034]

[0035] The same processing method as marginal distribution adaptation, conditional distribution adaptation can be transformed into minimizing the MMD distance between classes, expressed as

[0036]

[0037] Among them, n c , m c are the number of samples from the cth class in the source domain and the target domain respectively, C is the total number of categories, is the source domain sample belonging to category c, For the target domain sample belonging to the cth class, the kernel method is also used to obtain the following formula:

[0038]

[0039] M c The expression is:

[0040]

[0041] By combining c=0...C, we combine the marginal distribution difference with the conditional distribution difference, while also preserving the original attributes of the source domain data and the target domain data. Combining the above three optimization goals, we can obtain the overall optimization goal:

[0042]

[0043] max A T XHX T A

[0044] in is a regular term used to reduce overfitting. Let the denominator be the unit matrix. The above formula can be further transformed into:

[0045]

[0046] Finally, the Lagrangian method is used to obtain the migration matrix A, which is the feature migration model.

[0047] Optionally, the step S5 uses the feature migration model to migrate all degradation data of the source domain aluminum electrolytic cell, and uses the migrated degradation data to train a remaining life prediction model for the aluminum electrolytic cell, specifically including:

[0048] The degradation data of the entire life cycle of the source domain aluminum electrolytic cell is input into the trained feature migration model to obtain the migrated degradation data, and the migrated degradation data is used as the input of the long short-term memory neural network LSTM, and the corresponding remaining life label is used as the output. The remaining life prediction model of the aluminum electrolytic cell is constructed and the model is trained to obtain the remaining life prediction model of the aluminum electrolytic cell.

[0049] Optionally, the step S6 uses the aluminum electrolysis cell remaining life prediction model to predict the remaining life of the aluminum electrolysis cell in the target domain, specifically including:

[0050] Process parameter data related to the degradation of the target aluminum electrolytic cell are screened out from the online monitoring data generated by the operation of the target aluminum electrolytic cell, and are input into the trained feature migration model for migration. The migrated data are then input into the trained remaining life prediction model of the aluminum electrolytic cell to predict the corresponding remaining life.

[0051] In another aspect, a device for predicting the remaining life of an aluminum electrolysis cell is provided, the device comprising:

[0052] The acquisition module is used to screen out the process parameters related to the degradation of the aluminum electrolytic cell and obtain the degradation data of the source domain electrolytic cell and the target domain electrolytic cell;

[0053] a partitioning module, configured to partition the degradation states of the source domain aluminum electrolytic cell over its entire life cycle, and obtain degradation state labels corresponding to the degradation data of the source domain aluminum electrolytic cell;

[0054] a determination module, configured to determine the current degradation state of the aluminum electrolytic cell in the target domain and obtain the results of the existing degradation state division, thereby obtaining a degradation state label corresponding to the degradation data of the aluminum electrolytic cell in the target domain;

[0055] A first training module is configured to train a feature transfer model using partial degradation data and corresponding degradation state labels of the source domain aluminum electrolysis cell and all degradation data and corresponding degradation state labels of the target domain aluminum electrolysis cell;

[0056] A second training module is configured to migrate all degradation data of the source domain aluminum electrolytic cell using the feature migration model, and train a remaining life prediction model for the aluminum electrolytic cell using the migrated degradation data;

[0057] A prediction module is used to predict the remaining life of the aluminum electrolysis cell in the target domain using the remaining life prediction model of the aluminum electrolysis cell.

[0058] On the other hand, an electronic device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the above-mentioned method for predicting the remaining life of an aluminum electrolysis cell.

[0059] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement the above-mentioned method for predicting the remaining life of an aluminum electrolysis cell.

[0060] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0061] The present invention can enhance the generalization capability of the remaining life prediction model of aluminum electrolysis cells, thereby improving the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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 creative work.

[0063] Figure 1 This is a flow chart of a method for predicting the remaining life of an aluminum electrolysis cell provided by an embodiment of the present invention;

[0064] Figure 2This is a flow chart of another method for predicting the remaining life of an aluminum electrolysis cell provided by an embodiment of the present invention;

[0065] Figure 3 This is a block diagram of a device for predicting the remaining life of an aluminum electrolytic cell provided by an embodiment of the present invention;

[0066] Figure 4 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0067] like Figure 1 As shown, an embodiment of the present invention provides a method for predicting the remaining life of an aluminum electrolysis cell, the method comprising:

[0068] S1. Screen out the process parameters related to aluminum electrolytic cell degradation and obtain the degradation data of the source domain electrolytic cell and the target domain electrolytic cell;

[0069] S2. Dividing the degradation states of the source domain aluminum electrolysis cell over its entire life cycle to obtain degradation state labels corresponding to the degradation data of the source domain aluminum electrolysis cell;

[0070] S3. Determine the current degradation state of the aluminum electrolytic cell in the target domain and obtain the results of existing degradation state division to obtain a degradation state label corresponding to the degradation data of the aluminum electrolytic cell in the target domain;

[0071] S4. Using partial degradation data and corresponding degradation state labels of the source domain aluminum electrolysis cell, and all degradation data and corresponding degradation state labels of the target domain aluminum electrolysis cell, to train a feature transfer model;

[0072] S5. Using the feature migration model to migrate all degradation data of the source domain aluminum electrolytic cell, and using the migrated degradation data to train a remaining life prediction model for the aluminum electrolytic cell;

[0073] S6. Use the aluminum electrolysis cell remaining life prediction model to predict the remaining life of the aluminum electrolysis cell in the target domain.

[0074] The following combination Figure 2 , a method for predicting the remaining life of an aluminum electrolysis cell according to an embodiment of the present invention is described in detail, the method comprising:

[0075] S1. Screen out the process parameters related to aluminum electrolytic cell degradation and obtain the degradation data of the source domain electrolytic cell and the target domain electrolytic cell;

[0076] Optionally, the step S1 of screening out degradation-related process parameters of the aluminum electrolytic cell and obtaining degradation data of the source domain electrolytic cell and the target domain electrolytic cell specifically includes:

[0077] Normalizing the source domain aluminum electrolytic cell full life cycle monitoring data;

[0078] Using the two indicators of monotonicity and correlation to screen out degradation-related process parameters of the aluminum electrolysis cell, and obtain degradation data of the source domain aluminum electrolysis cell, wherein the degradation data of the source domain aluminum electrolysis cell includes full life cycle data of the degradation-related process parameters;

[0079] Obtain the degradation data of the target domain electrolytic cell, wherein the degradation data of the target domain electrolytic cell includes the existing data of the degradation-related process parameters of the target domain aluminum electrolytic cell from the beginning of operation to the current state, and the degradation-related process parameters of the target domain aluminum electrolytic cell include the same degradation-related process parameters as those of the source domain electrolytic cell.

[0080] In this embodiment of the present invention, aluminum electrolysis production process parameters are obtained through a daily on-site database, including 15 dimensions of data: operating voltage, average voltage, aluminum level, electrolyte level, molecular ratio, fluoride salt, cell temperature, anode stroke, aluminum output, current intensity, noise, alumina content, cathode voltage drop, silicon content, and iron content. The source domain aluminum electrolysis cell process parameter data includes full life cycle monitoring data from the start of operation to the shutdown of the cell, and the target domain aluminum electrolysis cell process parameter data includes existing data from the start of operation to the current state (operation to a certain degradation stage). The data sampling interval is 1 day.

[0081] After normalizing the source domain aluminum electrolytic cell full life cycle monitoring data, the monotonicity and correlation of each process parameter are calculated. Based on the on-site daily database, the correlation between each process parameter and the degradation of the aluminum electrolytic cell is different. Some parameters can show correlation with the degradation of the aluminum electrolytic cell, while some parameters show no correlation at all. Parameters with weak correlation with the degradation of the aluminum electrolytic cell will affect the training accuracy and generalization ability of the model when used as input to the prediction model. In order to remove parameters with low correlation with the degradation of the aluminum electrolytic cell, it is necessary to screen the process parameters in the daily database. Monotonicity and correlation are commonly used performance indicators for evaluating the degradation of aluminum electrolytic cells. Monotonicity represents an upward or downward trend, and correlation measures the degree to which a feature is linearly related to time. The specific calculation formula is as follows:

[0082]

[0083]

[0084] Where Mon and Corr represent monotonicity and correlation respectively, T is the total number of operating days, that is, the number of samples, no.of(dH>0) is the number of times the previous value of the process parameter is greater than the next value, no.of(dH<0) is the number of times the previous value of the process parameter is less than the next value, t iis the number of operating days corresponding to the i-th number of process parameters, t is the mean of the operating days sequence, a i is the i-th value of the process parameter, and a is the mean value of the process parameter.

[0085] The average value of the two is taken as the comprehensive evaluation index, the threshold is set to 0.5, and the process parameters with the normalized comprehensive evaluation index greater than 0.5 are screened out as the degradation-related process parameters of the aluminum electrolytic cell, that is, the process parameters that can characterize the degradation of the aluminum electrolytic cell. The number is recorded as d, and the historical data of the corresponding degradation-related process parameters of the source domain aluminum electrolytic cell and the target domain aluminum electrolytic cell are obtained.

[0086] S2. Dividing the degradation states of the source domain aluminum electrolysis cell over its entire life cycle to obtain degradation state labels corresponding to the degradation data of the source domain aluminum electrolysis cell;

[0087] Optionally, the step S2 divides the degradation state of the source domain aluminum electrolytic cell over its entire life cycle to obtain degradation state labels corresponding to the degradation data of the source domain aluminum electrolytic cell, specifically including:

[0088] The degradation data of the source domain electrolytic cell is input into the fuzzy c-means clustering FCM model, and the number of clusters is set to 5. The dividing lines of each degradation state are determined according to the clustering results, and the degradation state of the source domain aluminum electrolysis from the start of operation to the end of life is divided, which corresponds to the healthy operation state, initial degradation state, slow degradation state, rapid degradation state and severe degradation state of the aluminum electrolytic cell respectively, and then the degradation state label corresponding to the degradation data of the source domain aluminum electrolytic cell is obtained.

[0089] As a typical clustering algorithm, the Fuzzy C-means Clustering (FCM) algorithm integrates the essence of fuzzy theory, can provide more flexible clustering results, and realize unsupervised learning. From the perspective of health assessment, the extracted degradation data has its own rules and characteristics in different degradation stages of the aluminum electrolytic cell, but at the same time there is a characteristic of fuzzy boundaries between different degradation states. Therefore, the embodiment of the present invention uses the FCM model to divide the full life of the source domain aluminum electrolytic cell into five degradation stages: state1, state2, state3, state4, and state5, which correspond to the healthy operation state, initial degradation state, slow degradation state, rapid degradation state, and severe degradation state of the aluminum electrolytic cell, respectively, and then obtain the degradation state label corresponding to the degradation data of the source domain aluminum electrolytic cell. Specifically, all degradation data of the d process parameters corresponding to the source domain are input into the FCM model, the number of clusters is set to 5, and the boundary of each degradation state is determined according to the clustering results, completing the division of the degradation state of the source domain aluminum electrolysis from the start of operation to the end of life.

[0090] The FCM model introduces the concept of membership to determine which category a sample point belongs to. The problem of calculating the membership of each sample to each category can be transformed into the problem of minimizing the following objective function.

[0091]

[0092]

[0093] Where N is the total number of samples, L is the number of cluster categories, u ij is the membership degree of the i-th sample to the j-th class, ||x i -c j || 2 is the distance between the i-th sample and the j-th cluster center, and m is a weighted index in the range of [1,∞).

[0094] This optimization problem can be solved using the Lagrange multiplier method, and we can get the following two equations:

[0095]

[0096]

[0097] The specific solution steps are as follows:

[0098] (1) First, initialize the membership matrix that meets the conditions;

[0099] (2) Calculate cluster centers;

[0100] (3) Calculate the new membership matrix;

[0101] (4) Repeat steps (2) and (3) until the number of iterations is reached or the convergence condition is met;

[0102] (5) Obtain the final membership matrix and obtain the final clustering result based on the principle that the sample is classified into the class to which it has the largest membership.

[0103] S3. Determine the current degradation state of the aluminum electrolytic cell in the target domain and obtain the results of existing degradation state division to obtain a degradation state label corresponding to the degradation data of the aluminum electrolytic cell in the target domain;

[0104] The FCM model is also used to perform cluster analysis on the degradation data of the target domain electrolytic cells, and the silhouette coefficient is used to determine the optimal number of clusters. Based on the optimal number of clusters and the clustering results, the current degradation stage of the target domain aluminum electrolytic cells is judged and the results of the degradation stage division of the target domain aluminum electrolytic cells are obtained.

[0105] The silhouette coefficient is a metric used to evaluate clustering effectiveness. It consists of two factors: cohesion and separation. Cohesion represents the closeness of the same category, while separation represents the dispersion of different categories. The lower the cohesion within the same category, the greater the separation between different categories. In other words, a larger silhouette coefficient indicates a better clustering effect.

[0106] Optionally, the step S3 of determining the current degradation state of the aluminum electrolytic cell in the target domain and obtaining the result of existing degradation state division to obtain the degradation state label corresponding to the degradation data of the aluminum electrolytic cell in the target domain specifically includes:

[0107] The degradation data of the aluminum electrolytic cell in the target domain is input into the FCM model, the number of clusters is set to M (M≤5), and the silhouette coefficient values corresponding to different M are calculated. The calculation formula is:

[0108]

[0109]

[0110] Where S represents the total silhouette coefficient, n t is the number of samples in the target domain, s(i) represents the silhouette coefficient of the i-th sample point, a(i) represents the cohesion of the sample point, which is the average distance between the sample point i and other sample points in the same category, and b(i) represents the separation, which is the minimum value of the average distance between the sample point i and the sample points of other categories. The larger the silhouette coefficient, the better the clustering effect. The M value corresponding to the maximum silhouette coefficient is counted as C, and C is the optimal number of clusters. It is determined that the target domain aluminum electrolytic cell is currently in the C-th degradation state, and according to the clustering result when the number of clusters is C, the first C degradation states of the target domain aluminum electrolytic cell are divided, and then the degradation state label corresponding to the degradation data of the target domain aluminum electrolytic cell is obtained.

[0111] The remaining life prediction method of an aluminum electrolytic cell according to an embodiment of the present invention utilizes an FCM model to perform cluster analysis on existing data on degradation process parameters of a target domain electrolytic cell, and determines the optimal number of clusters based on a silhouette coefficient. This method eliminates the need for prior knowledge of a source domain aluminum electrolytic cell to identify and classify the degradation state of the target domain aluminum electrolytic cell, thereby simplifying the calculation and avoiding classification errors caused by inconsistent data distribution.

[0112] S4. Using partial degradation data and corresponding degradation state labels of the source domain aluminum electrolysis cell, and all degradation data and corresponding degradation state labels of the target domain aluminum electrolysis cell, to train a feature transfer model;

[0113] The degradation data of the first C degradation states of the source domain aluminum electrolytic cell and the target domain aluminum electrolytic cell, as well as the corresponding degradation state labels, are input into the improved JDA migration algorithm. While retaining the original attributes of each data, the marginal distribution differences between the two data and the conditional distribution differences of the same degradation state are reduced, and finally a migration matrix is obtained. This migration matrix is the feature migration model trained using partial source domain data and existing target domain data.

[0114] The embodiment of the present invention adopts an improved JDA migration algorithm to construct a feature migration model, and uses the data of the existing degradation stage in the target domain and the corresponding partial data in the source domain for learning. This makes the marginal distribution difference and the conditional distribution difference of the same degradation state between the migrated source domain data and the target domain data as small as possible, while retaining the characteristics of their respective original data, thereby effectively improving the generalization ability of the remaining life prediction model.

[0115] Optionally, the step S4 utilizes partial degradation data and corresponding degradation state labels of the source domain aluminum electrolytic cell and all degradation data and corresponding degradation state labels of the target domain aluminum electrolytic cell to train a feature migration model, specifically including:

[0116] All degradation data and corresponding degradation state labels of the target domain aluminum electrolytic cell, as well as the degradation data and corresponding degradation state labels of the source domain aluminum electrolytic cell corresponding to the existing degradation state of the target domain aluminum electrolytic cell, are input into the improved JDA migration algorithm. While retaining the original attributes of each data, the marginal distribution difference of the two data and the conditional distribution difference of the same degradation state are reduced, and finally a migration matrix is obtained. The migration matrix is the feature migration model.

[0117] The improved JDA migration algorithm includes three steps: preserving the original attributes of the source and target domain data, minimizing the marginal distribution of the source and target domains, and minimizing the conditional distribution of the source and target domains. Specifically, the algorithm includes:

[0118] (1) Retain the original attributes

[0119] The original attributes of the data are retained by maximizing the variance of the transformed data, X=[x1,x2,...,x n ] is the matrix after the source domain and the target domain are merged, n=n s +n t , where n s is the number of source domain samples, n t is the number of samples in the target domain, H is the centralization matrix, then the covariance matrix is XHX T , assuming A is the migration matrix, this problem can be transformed into the following expression:

[0120] max tr(AT XHX T A)

[0121] (2) Edge distribution adaptation

[0122] Maximum Mean Discrepancy (MMD) is the most commonly used distance metric in transfer learning. It maps the original data into Hilbert space and calculates the average distance between samples in the new space as the distance between the two data distributions. Therefore, the MMD distance is used to measure the difference in the marginal probability distributions of two data distributions. This problem is transformed into minimizing the mapped MMD distance, as expressed as follows:

[0123]

[0124] By introducing the kernel method, the above formula is transformed into the following expression:

[0125]

[0126] Where M0 is an MMD matrix, which is expressed as

[0127]

[0128] (3) Conditional distribution adaptation

[0129] The same processing method as marginal distribution adaptation, conditional distribution adaptation can be transformed into minimizing the MMD distance between classes, expressed as

[0130]

[0131] Among them, n c , m c are the number of samples from the cth class in the source domain and the target domain respectively, C is the total number of categories, is the source domain sample belonging to category c, For the target domain sample belonging to the cth class, the kernel method is also used to obtain the following formula:

[0132]

[0133] M c The expression is:

[0134]

[0135] By combining c=0...C, we combine the marginal distribution difference with the conditional distribution difference, while also preserving the original attributes of the source domain data and the target domain data. Combining the above three optimization goals, we can obtain the overall optimization goal:

[0136]

[0137] max A T XHX T A

[0138] in is a regular term used to reduce overfitting. Let the denominator be the unit matrix. The above formula can be further transformed into:

[0139]

[0140] Finally, the Lagrangian method is used to obtain the migration matrix A, which is the feature migration model.

[0141] The remaining life prediction method of an aluminum electrolytic cell in an embodiment of the present invention utilizes an improved JDA migration algorithm to learn a feature migration model. It not only takes minimizing the edge distribution difference between the source domain and target domain data as the optimization goal, but also considers the distribution difference under the same degradation state conditions. It can effectively reduce the data distribution difference between the source domain aluminum electrolytic cell and the target domain aluminum electrolytic cell, realize online prediction of the remaining life of the aluminum electrolytic cell in the same work area or between different work areas, improve the remaining life prediction accuracy, and reduce the occurrence of negative migration.

[0142] Compared with the traditional JDA migration algorithm, the improved JDA migration algorithm in the embodiment of the present invention does not need to use source domain data to train a simple classifier to obtain target domain pseudo labels, which greatly simplifies the computational complexity. Moreover, the source domain and target domain degradation state category labels obtained using the FCM algorithm have higher accuracy than the target domain pseudo labels obtained using a simple classifier, which can better improve the migration effect.

[0143] S5. Using the feature migration model to migrate all degradation data of the source domain aluminum electrolytic cell, and using the migrated degradation data to train a remaining life prediction model for the aluminum electrolytic cell;

[0144] Optionally, the step S5 uses the feature migration model to migrate all degradation data of the source domain aluminum electrolytic cell, and uses the migrated degradation data to train a remaining life prediction model for the aluminum electrolytic cell, specifically including:

[0145] The degradation data of the entire life cycle of the source domain aluminum electrolytic cell is input into the trained feature migration model to obtain the migrated degradation data, and the migrated degradation data is used as the input of the long short-term memory neural network LSTM, and the corresponding remaining life label is used as the output. The remaining life prediction model of the aluminum electrolytic cell is constructed and the model is trained to obtain the remaining life prediction model of the aluminum electrolytic cell.

[0146] Aluminum electrolytic cells degrade over time, and historical data also has a significant impact on their current degradation state. Their remaining lifespan is related not only to current data but also to historical data. Long Short-Term Memory (LSTM) is a time-recurrent network with powerful memory and time series data processing capabilities. Therefore, in this embodiment of the present invention, LSTM is used to construct a regression prediction model, while simultaneously using data from the previous n-1 days (e.g., the previous 9 days) and the current day to predict the current remaining lifespan. The degradation data from the entire lifespan of the source domain is input into a trained feature migration model to obtain the migrated degradation data. The degradation data corresponding to each consecutive n days after migration is used as a sample as the input to the LSTM, and the remaining lifespan corresponding to the last of these n days is used as the output. A model for predicting the remaining lifespan of aluminum electrolytic cells is constructed and trained to obtain the model.

[0147] S6. Use the aluminum electrolysis cell remaining life prediction model to predict the remaining life of the aluminum electrolysis cell in the target domain.

[0148] Optionally, the step S6 uses the aluminum electrolysis cell remaining life prediction model to predict the remaining life of the aluminum electrolysis cell in the target domain, specifically including:

[0149] Process parameter data related to the degradation of the target aluminum electrolytic cell are screened out from the online monitoring data generated by the operation of the target aluminum electrolytic cell, and are input into the trained feature migration model for migration. The migrated data are then input into the trained remaining life prediction model of the aluminum electrolytic cell to predict the corresponding remaining life.

[0150] Specifically, process parameter data related to the degradation of the target aluminum electrolytic cell are screened out from the online monitoring data generated by the operation of the target aluminum electrolytic cell, and the process parameter data and the degradation data of the previous n-1 days are input into the trained feature migration model for migration. The migrated data are then input into the trained aluminum electrolytic cell remaining life prediction model to predict the corresponding remaining life.

[0151] The embodiment of the present invention can use the process parameter data related to the degradation of the target aluminum reduction cell in the current period and the n-1 days before to predict the remaining life of the current target aluminum reduction cell. It can also use the process parameter data related to the degradation of the target aluminum reduction cell in the day after the current period and the n-1 days before to predict the remaining life of the target aluminum reduction cell in the day after the current period. The embodiment of the present invention performs online prediction, meaning that when new parameters are generated for a new day of operation, the remaining life corresponding to the new day of operation can be predicted.

[0152] An embodiment of the present invention further provides a device for predicting the remaining life of an aluminum electrolysis cell, the device comprising:

[0153] An acquisition module 310 is used to screen out degradation-related process parameters of the aluminum electrolytic cell and obtain degradation data of the source domain electrolytic cell and the target domain electrolytic cell;

[0154] a classification module 320 for classifying the degradation states of the source domain aluminum electrolytic cell over its entire life cycle, and obtaining degradation state labels corresponding to the degradation data of the source domain aluminum electrolytic cell;

[0155] A determination module 330 is configured to determine the current degradation state of the aluminum electrolytic cell in the target domain and obtain the results of existing degradation state divisions to obtain a degradation state label corresponding to the degradation data of the aluminum electrolytic cell in the target domain;

[0156] A first training module 340 is configured to train a feature transfer model using partial degradation data and corresponding degradation state labels of the source domain aluminum electrolysis cell and all degradation data and corresponding degradation state labels of the target domain aluminum electrolysis cell;

[0157] A second training module 350 is configured to migrate all degradation data of the source domain aluminum electrolytic cell using the feature migration model, and train a remaining life prediction model for the aluminum electrolytic cell using the migrated degradation data;

[0158] The prediction module 360 is used to predict the remaining life of the aluminum electrolysis cell in the target domain using the aluminum electrolysis cell remaining life prediction model.

[0159] An embodiment of the present invention provides an apparatus for predicting the remaining life of an aluminum electrolytic cell, the functional structure of which corresponds to an embodiment of the present invention provides a method for predicting the remaining life of an aluminum electrolytic cell, which will not be described in detail here.

[0160] Figure 4 It is a structural diagram of an electronic device 400 provided in an embodiment of the present invention. The electronic device 400 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 401 and one or more memories 402, wherein the memory 402 stores at least one instruction, and the at least one instruction is loaded and executed by the processor 401 to implement the steps of the above-mentioned aluminum electrolytic cell remaining life prediction method.

[0161] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including instructions. The instructions are executable by a processor in a terminal to implement the above-described method for predicting the remaining life of an aluminum electrolysis cell. For example, the computer-readable storage medium may be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, or the like.

[0162] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or by a program to instruct the relevant hardware, and the program may be stored in a computer-readable storage medium, which may be a read-only memory, a disk, or an optical disk, etc.

[0163] 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 scope of protection of the present invention.

Claims

1. A method for predicting the remaining life of an aluminum electrolysis cell, characterized in that: The method comprises: S1. Screen out the process parameters related to the degradation of the aluminum electrolytic cell and obtain the degradation data of the aluminum electrolytic cell in the source domain and the aluminum electrolytic cell in the target domain; S2. Dividing the degradation states of the source domain aluminum electrolysis cell over its entire life cycle to obtain degradation state labels corresponding to the degradation data of the source domain aluminum electrolysis cell; S3. Determine the current degradation state of the aluminum electrolytic cell in the target domain and obtain the results of existing degradation state division to obtain a degradation state label corresponding to the degradation data of the aluminum electrolytic cell in the target domain; S4. Using partial degradation data and corresponding degradation state labels of the source domain aluminum electrolysis cell, and all degradation data and corresponding degradation state labels of the target domain aluminum electrolysis cell, to train a feature transfer model; S5. Using the feature migration model to migrate all degradation data of the source domain aluminum electrolytic cell, and using the migrated degradation data to train a remaining life prediction model for the aluminum electrolytic cell; S6. Using the aluminum electrolysis cell remaining life prediction model, predict the remaining life of the aluminum electrolysis cell in the target domain; The step S4 utilizes partial degradation data and corresponding degradation state labels of the source domain aluminum electrolytic cell and all degradation data and corresponding degradation state labels of the target domain aluminum electrolytic cell to train a feature transfer model, specifically including: All degradation data and corresponding degradation state labels of the target domain aluminum electrolytic cell, as well as the degradation data and corresponding degradation state labels of the source domain aluminum electrolytic cell corresponding to the existing degradation state of the target domain aluminum electrolytic cell, are input into the improved JDA migration algorithm. While retaining the original attributes of each data, the marginal distribution difference of the two data and the conditional distribution difference of the same degradation state are reduced, and finally a migration matrix is obtained. The migration matrix is the feature migration model. The improved JDA migration algorithm includes three steps: preserving the original attributes of the source and target domain data, minimizing the marginal distribution of the source and target domains, and minimizing the conditional distribution of the source and target domains. Specifically, it includes: The original attributes of the data are retained by maximizing the variance of the transformed data. is the matrix after the source domain and target domain data are merged, ,in is the number of source domain samples, is the number of samples in the target domain, is a centered matrix, then the covariance matrix is , assuming is the migration matrix, then the problem of retaining the original attributes of the data is transformed into the following expression: ； The maximum mean difference (MMD) distance is used to measure the difference in the marginal probability distribution of the two data distributions. The problem of minimizing the marginal distribution between the source domain and the target domain is transformed into minimizing the MMD distance after mapping. The expression is as follows: ； By introducing the kernel method, the above formula is transformed into the following expression: ； in Is an MMD matrix, its expression is ； The same processing method as marginal distribution adaptation, conditional distribution adaptation can be transformed into minimizing the MMD distance between classes, expressed as ； in, , are from the source domain and the target domain respectively. The number of samples of the class, is the total number of categories, For the Source domain samples of the class, For the The target domain sample of the class is also obtained by the kernel method: ； The expression is: ； pass Combining the marginal distribution difference with the conditional distribution difference, while also preserving the original attributes of the source domain data and the target domain data, the overall optimization goal is obtained by combining the above three optimization goals: ； in is a regular term used to reduce overfitting. Let the denominator be the unit matrix, and the above formula is transformed into: ； Finally, the Lagrangian method is used to obtain the transfer matrix ,matrix This is the feature migration model.

2. The method according to claim 1, characterized in that The step S1 selects the degradation-related process parameters of the aluminum electrolysis cell and obtains the degradation data of the source domain aluminum electrolysis cell and the target domain aluminum electrolysis cell, specifically including: Normalizing the source domain aluminum electrolytic cell full life cycle monitoring data; Using the two indicators of monotonicity and correlation to screen out degradation-related process parameters of the aluminum electrolysis cell, and obtain degradation data of the source domain aluminum electrolysis cell, wherein the degradation data of the source domain aluminum electrolysis cell includes full life cycle data of the degradation-related process parameters; Obtain the degradation data of the target domain aluminum electrolysis cell, wherein the degradation data of the target domain aluminum electrolysis cell includes the existing data of the degradation-related process parameters of the target domain aluminum electrolysis cell from the beginning of operation to the current state, and the degradation-related process parameters of the target domain aluminum electrolysis cell include the same degradation-related process parameters as those of the source domain aluminum electrolysis cell.

3. The method according to claim 1, characterized in that The step S2 divides the degradation state of the source domain aluminum electrolytic cell over its entire life cycle to obtain degradation state labels corresponding to the degradation data of the source domain aluminum electrolytic cell, specifically including: The degradation data of the source domain aluminum electrolytic cell is input into the fuzzy c-means clustering FCM model, and the number of clusters is set to 5. The dividing lines of each degradation state are determined according to the clustering results, and the degradation state of the source domain aluminum electrolytic cell from the start of operation to the end of its life is divided, corresponding to the healthy operation state, initial degradation state, slow degradation state, rapid degradation state and severe degradation state of the aluminum electrolytic cell, respectively, and then the degradation state label corresponding to the degradation data of the source domain aluminum electrolytic cell is obtained.

4. The method according to claim 1, wherein The step S3 of determining the current degradation state of the aluminum electrolytic cell in the target domain and obtaining the results of existing degradation state division to obtain the degradation state label corresponding to the degradation data of the aluminum electrolytic cell in the target domain specifically includes: The degradation data of the target domain aluminum electrolytic cell is input into the FCM model, and the number of clusters is set to , calculate different The corresponding silhouette coefficient value is calculated as follows: ； ； in represents the overall silhouette coefficient, is the number of target domain samples, represents the silhouette coefficient of the jth sample point, Represents the cohesion of the sample point, which is the average distance between the sample point j and other sample points in the same category. Represents the degree of separation, which is the minimum value of the average distance between sample point j and other category sample points; the larger the silhouette coefficient, the better the clustering effect. The value is calculated as , is the optimal number of clusters, then it is determined that the target domain aluminum electrolysis cell is currently in the degenerate states, and according to the number of clusters The clustering results of the target domain aluminum electrolysis cell are completed. The degradation state is divided into the degradation states, and then the degradation state labels corresponding to the degradation data of the aluminum electrolytic cell in the target domain are obtained.

5. The method according to claim 1, wherein The step S5 uses the feature migration model to migrate all degradation data of the source domain aluminum electrolytic cell, and uses the migrated degradation data to train a remaining life prediction model for the aluminum electrolytic cell, specifically including: The degradation data of the entire life cycle of the source domain aluminum electrolytic cell is input into the trained feature migration model to obtain the migrated degradation data, and the migrated degradation data is used as the input of the long short-term memory neural network LSTM, and the corresponding remaining life label is used as the output. The remaining life prediction model of the aluminum electrolytic cell is constructed and the model is trained to obtain the remaining life prediction model of the aluminum electrolytic cell.

6. The method according to claim 1, characterized in that The step S6 uses the aluminum electrolysis cell remaining life prediction model to predict the remaining life of the aluminum electrolysis cell in the target domain, specifically including: The process parameter data related to the degradation of the aluminum electrolytic cell is screened out from the online monitoring data generated by the operation of the aluminum electrolytic cell in the target domain, and is input into the trained feature migration model for migration. The migrated data is then input into the trained remaining life prediction model of the aluminum electrolytic cell to predict the corresponding remaining life.

7. A device for predicting the remaining life of an aluminum electrolytic cell, characterized in that: The device comprises: An acquisition module is used to screen out the process parameters related to the degradation of the aluminum electrolytic cell and obtain the degradation data of the aluminum electrolytic cell in the source domain and the aluminum electrolytic cell in the target domain; a partitioning module, configured to partition the degradation states of the source domain aluminum electrolytic cell over its entire life cycle, and obtain degradation state labels corresponding to the degradation data of the source domain aluminum electrolytic cell; a determination module, configured to determine the current degradation state of the aluminum electrolytic cell in the target domain and obtain the results of the existing degradation state division, thereby obtaining a degradation state label corresponding to the degradation data of the aluminum electrolytic cell in the target domain; A first training module is configured to train a feature transfer model using partial degradation data and corresponding degradation state labels of the source domain aluminum electrolysis cell and all degradation data and corresponding degradation state labels of the target domain aluminum electrolysis cell; A second training module is configured to migrate all degradation data of the source domain aluminum electrolytic cell using the feature migration model, and train a remaining life prediction model for the aluminum electrolytic cell using the migrated degradation data; A prediction module, configured to predict the remaining life of the aluminum electrolysis cell in the target domain using the remaining life prediction model of the aluminum electrolysis cell; The first training module is specifically used to: All degradation data and corresponding degradation state labels of the target domain aluminum electrolytic cell, as well as the degradation data and corresponding degradation state labels of the source domain aluminum electrolytic cell corresponding to the existing degradation state of the target domain aluminum electrolytic cell, are input into the improved JDA migration algorithm. While retaining the original attributes of each data, the marginal distribution difference of the two data and the conditional distribution difference of the same degradation state are reduced, and finally a migration matrix is obtained. The migration matrix is the feature migration model. The improved JDA migration algorithm includes three steps: preserving the original attributes of the source and target domain data, minimizing the marginal distribution of the source and target domains, and minimizing the conditional distribution of the source and target domains. Specifically, the algorithm includes: The original attributes of the data are retained by maximizing the variance of the transformed data. is the matrix after the source domain and target domain data are merged, ,in is the number of source domain samples, is the number of samples in the target domain, is a centered matrix, then the covariance matrix is , assuming is the migration matrix, then the problem of retaining the original attributes of the data is transformed into the following expression: ； The maximum mean difference (MMD) distance is used to measure the difference in the marginal probability distribution of the two data distributions. The problem of minimizing the marginal distribution between the source domain and the target domain is transformed into minimizing the MMD distance after mapping. The expression is as follows: ； By introducing the kernel method, the above formula is transformed into the following expression: ； in Is an MMD matrix, its expression is ； The same processing method as marginal distribution adaptation, conditional distribution adaptation can be transformed into minimizing the MMD distance between classes, expressed as ； in, , are from the source domain and the target domain respectively. The number of samples of the class, is the total number of categories, For the Source domain samples of the class, For the The target domain sample of the class is also obtained by the kernel method: ； The expression is: ； pass Combining the marginal distribution difference with the conditional distribution difference, while also preserving the original attributes of the source domain data and the target domain data, the overall optimization goal is obtained by combining the above three optimization goals: ； in is a regular term used to reduce overfitting. Let the denominator be the unit matrix, and the above formula is transformed into: ； Finally, the Lagrangian method is used to obtain the transfer matrix ,matrix This is the feature migration model.

8. An electronic device comprising a processor and a memory, wherein the memory stores at least one instruction, characterized in that: The at least one instruction is loaded and executed by the processor to implement the method for predicting the remaining life of an aluminum electrolysis cell as described in any one of claims 1 to 6.

9. A computer-readable storage medium, wherein at least one instruction is stored in the storage medium, characterized in that: The at least one instruction is loaded and executed by the processor to implement the method for predicting the remaining life of an aluminum electrolysis cell as described in any one of claims 1 to 6.

Images