Method for predicting residual life of aluminum electrolysis cell based on SVR-RFE-XGBoost
By constructing a prediction model for the remaining life of aluminum electrolytic cells based on the SVR-RFE-XGBoost method, the problem of unreasonable maintenance cycles of aluminum electrolytic cells is solved, scientific maintenance of aluminum electrolytic cells is realized, production efficiency and prediction accuracy are improved, and maintenance costs are reduced.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-11
- Publication Date
- 2026-03-13
AI Technical Summary
The maintenance cycle settings for aluminum electrolytic cells in the existing technology are unreasonable, resulting in equipment that cannot be repaired in a timely manner or is repaired too frequently, which increases the probability of failure and wastes resources, and there is a lack of effective methods for predicting the remaining life.
By adopting the SVR-RFE-XGBoost method, a prediction model for the remaining life of aluminum electrolytic cells is established by constructing a subset of data features and adjusting parameters. Historical data is used for scientific and reasonable dynamic maintenance, key features are selected and the prediction model is trained to achieve accurate prediction of the remaining life of aluminum electrolytic cells.
It improved the production efficiency of aluminum electrolytic cells, reduced maintenance costs, decreased economic losses, extended the service life of electrolytic cells, and improved the accuracy of prediction models.
Smart Images
Figure CN121659718A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of artificial intelligence and non-ferrous metal smelting technology, and in particular to a method for predicting the remaining life of aluminum electrolytic cells based on SVR-RFE-XGBoost. Background Technology
[0002] my country boasts the world's largest aluminum production capacity, making it a pillar industry of the nation. However, the harsh environment of aluminum electrolysis production, influenced by the coupling and interaction of multiple physical fields, leads to frequent process failures, increased energy consumption, reduced production efficiency, and even directly impacts the lifespan of aluminum electrolytic cells. As the core equipment in industrial aluminum smelting, the lifespan of aluminum electrolytic cells directly affects the economic and environmental benefits of production enterprises. Therefore, predicting the lifespan of aluminum electrolytic cells has become a crucial aspect of research in the aluminum electrolysis industry.
[0003] The aluminum electrolysis production process is a complex and ever-changing physicochemical process. During production, various factors influence the development of the aluminum electrolytic cell. From material balance and energy balance to fluid dynamics, electromagnetic fields, and mechanical properties, all these phenomena work together in electrolysis. In this complex environment, process engineers ensure the normal operation of the production process by setting reasonable process parameters. However, even so, the lifespan of the aluminum electrolytic cell remains unstable, ranging from a few hundred days to thousands of days. Most domestic and international research on the lifespan of aluminum electrolytic cells focuses on the cell lining materials, damage mechanisms, and ways to extend lifespan; research on predicting the remaining lifespan is scarce.
[0004] Currently, maintenance of aluminum electrolytic cells is mainly based on expert opinions and fixed maintenance cycles. If the fixed maintenance cycle is not set reasonably, the equipment cannot be repaired in a timely manner, greatly increasing the probability of failure, or the number of maintenance is too frequent, resulting in waste of resources. Summary of the Invention
[0005] The purpose of this invention is to provide a method for predicting the remaining life of aluminum electrolytic cells based on SVR-RFE-XGBoost, so as to achieve prediction of the remaining life of aluminum electrolytic cells with high accuracy.
[0006] To achieve the above objectives, this invention provides a method for predicting the remaining life of aluminum electrolytic cells based on SVR-RFE-XGBoost, comprising the following steps: Historical data of aluminum electrolytic cells are acquired, and the historical data is preprocessed to construct a dataset, which consists of data features; An SVR-RFE model is constructed, the hyperparameters of the SVR-RFE model are adjusted, and the preprocessed dataset is input into the adjusted SVR-RFE model. After calculating the importance of features, the data features that affect the remaining life of the aluminum electrolytic cell are sorted by importance and selected to form a feature subset. Build the XGBoost algorithm framework; Parameters were adjusted for the XGBoost algorithm framework; The feature subset is input into the parameter-adjusted XGBoost algorithm framework for training to generate a prediction model for the remaining life of aluminum electrolytic cells. The characteristics of the data to be tested are input into the prediction model of the remaining life of the aluminum electrolytic cell to calculate the remaining life of the aluminum electrolytic cell.
[0007] Historical data consists of daily reports from the electrolytic cells, primarily consisting of online automatic data recorded by the on-site cell control machine and offline manual data from laboratory tests. All data are authentic historical data obtained by sampling within the same remaining lifespan of the electrolytic cells, and the data to be tested and the filtered data share the same characteristics.
[0008] The SVR-RFE model uses SVR as its base model, transforming the feature selection problem into a feature importance ranking problem. SVR uses a nonlinear mapping function to map the sample set from a low-dimensional space to a high-dimensional space, and then performs linear fitting in the high-dimensional space. The functional relationship can be expressed by the following formula: In the formula, ω is the weight, Φ ( x ) is a nonlinear mapping function. b This is a bias term. According to the principle of minimizing structural risk, f ( x This can be equivalent to solving an optimization problem, and the solution model is shown in equation X: In the formula, C is the regularization constant. This represents the loss function with an acceptable deviation of ε.
[0009] Introducing slack variables and and Lagrange multipliers , The formula can be written as Solving the above equation, we obtain the expression for SVR as follows: In the formula, This is the kernel function.
[0010] This invention provides a method for predicting the remaining life of aluminum electrolytic cells based on an SVR-RFE-XGBoost fusion model. This improves upon previous periodic maintenance by providing dynamic maintenance based on data-driven predictions of the remaining life of aluminum electrolytic cells. Before building the prediction model, the SVR-RFE model is used to prioritize the data for training, selecting key features for predicting the remaining life of aluminum electrolytic cells as a subset of features for training. This subset of features is then used to train the prediction model, thereby enabling the scientific and rational scheduling of cell shutdowns, improving the production efficiency of aluminum electrolytic cells, reducing maintenance costs, and minimizing economic losses.
[0011] Furthermore, the preprocessing includes the following steps: Historical data is organized and cleaned to remove invalid and abnormal data features, retain valid data features, and fill in missing values with the mean. Standard deviation standardization is performed on the imputed data features to eliminate differences in the units and magnitudes of different data features, so that all data features are unified to the same scale. The processed data features can be represented as data vectors, and the data vectors shown are scaled to the [0,1] interval.
[0012] Furthermore, the data features within the dataset include set voltage, operating voltage, average voltage, aluminum level, electrolyte level, molecular ratio, fluoride salt, tank temperature, anode stroke, aluminum output, effect coefficient, noise, current density, alumina concentration, current efficiency, cathode voltage drop, silicon content in the aluminum liquid, iron content in the aluminum liquid, and power consumption.
[0013] Furthermore, the step of sorting and filtering data features that affect the remaining lifespan of aluminum electrolytic cells by importance to form a feature subset includes the following steps: Each data feature is sorted from highest to lowest importance according to its weight, and a predetermined number of data features are selected to form a feature subset.
[0014] The preset quantity is 12.
[0015] RFE uses SVR as its base model, transforming the feature selection problem into a feature importance ranking problem. It ranks each feature according to its weights from highest to lowest importance, selecting the optimal number of features to generate the optimal feature subset. The ranking criteria for the i features are: The preprocessed dataset was input into the parameter-tuned SVR-RFE model to calculate feature importance. Features with the lowest importance were progressively removed to generate the optimal feature subset for predicting the remaining lifespan of the aluminum electrolytic cell. After model selection, 12 feature vectors were ultimately chosen as key features for predicting the remaining lifespan of the aluminum electrolytic cell. Ranked by importance, they are: effect coefficient, alumina concentration, fluoride salts, current density, operating voltage, molecular ratio, silicon content in the molten aluminum, aluminum output, electrolyte level, aluminum level, anode travel, and iron content in the molten aluminum.
[0016] Furthermore, the parameter tuning of the XGBoost algorithm framework includes the following steps: The range of predefined parameters; K-fold cross-validation is performed using grid search. The parameter combination corresponding to the maximum efficiency returned after K-fold cross-validation is the optimal parameter of the XGBoost algorithm framework.
[0017] The K-fold cross-validation method involves dividing the training samples into K equal parts. Each time, K-1 parts are taken out as training data, and the remaining part is taken as test data. This process is repeated K times, and the average goodness of fit (r2) of the K cross-validations is obtained as the result. After K-fold cross-validation, an efficiency value is returned. The parameter combination corresponding to the maximum efficiency value is the optimal parameter of the model.
[0018] Furthermore, the aforementioned construction of the XGBoost algorithm framework refers to building an XGBoost algorithm framework based on a tree model, whose objective function is: in, Represents the true value of the sample. Let be the predicted value for the i-th sample, and T be the number of leaf nodes in the decision tree. Let λ represent the output score of each leaf node in the decision tree, where λ is the coefficient of the leaf node. The leaf weight penalty regularization term.
[0019] Furthermore, the data vector shown is scaled to the [0,1] interval, that is, the dataset is normalized using the standard deviation standardization principle, as shown in the following formula: in, The dataset after normalization. x For the original dataset, μ Let σ be the mean of all data, and σ be the standard deviation of all data.
[0020] The SVR-RFE-XGBoost-based method for predicting the remaining life of aluminum electrolytic cells proposed in this invention improves upon previous periodic maintenance by dynamic maintenance based on data-driven predictions of the remaining life of aluminum electrolytic cells. The predicted remaining life of aluminum electrolytic cells can provide a basis for monitoring and maintenance, thereby extending the service life of the electrolytic cells and reducing economic losses. In the process of building the prediction model, the importance of the data used for training is also ranked to improve the accuracy of the prediction model in predicting the life of aluminum electrolytic cells. Attached Figure Description
[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0022] Figure 1 This is the framework of the SVR-RFE-XGBoost fusion model provided by this invention.
[0023] Figure 2 This is a graph showing the prediction results of the SVR-RFE-XGBoost model provided by this invention.
[0024] Figure 3 This is a comparison chart of simulation results for remaining lifetime prediction between the present invention and mainstream machine learning algorithms. Detailed Implementation
[0025] The technical solutions of this embodiment will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.
[0026] The invention discloses a method for determining the remaining lifespan of aluminum electrolytic cells based on the SVR-RFE-XGBoost algorithm, the method comprising: Obtain historical data; The original historical data collected by this invention consists of daily reports for each electrolytic cell, online automatic data recording by the on-site cell control machine, and offline manual data testing. All data are real historical data sampled from the electrolytic cells within the same remaining lifespan, including data from the aluminum electrolytic cell from the time it was put into production to the time of its first major overhaul. The collected data features include: set voltage, operating voltage, average voltage, aluminum level, electrolyte level, molecular ratio, fluoride salt, tank temperature, anode stroke, aluminum output, effect coefficient, noise, current density, alumina concentration, current efficiency, cathode voltage drop, silicon content in aluminum liquid, iron content in aluminum liquid, and power consumption. Data preprocessing: The collected raw data contains many missing values, which are filled using the mean. Since the dimensions and magnitudes of the data features are different, this invention uses standard deviation standardization to ensure that different features have the same scale, resulting in a final data vector falling within the range [0,1].
[0027] Feature extraction; Construct an SVR-RFE model; Constructing an SVR-RFE model involves the following steps: SVR uses a nonlinear mapping function to map the sample set from a low-dimensional space to a high-dimensional space, and then performs linear fitting in the high-dimensional space. The functional relationship can be expressed by the following formula.
[0028] In the formula, ω is the weight, Φ ( x ) is a nonlinear mapping function. b This is a bias term. According to the principle of minimizing structural risk, f ( x This can be equivalent to solving an optimization problem, and the solution model is shown in the following equation: In the formula, C is the regularization constant. The loss function, representing the allowable deviation ε, is expressed as follows: In the formula, .
[0029] Introducing slack variables and To correct for irregular factors, the solution model and expression are rewritten as follows: At this point, the regression problem is transformed into a problem concerning the parameters to be determined. ω, b For convex quadratic programming problems, we introduce Lagrange multipliers. , The dual form is obtained.
[0030] The expression for SVR is obtained by solving the problem: In the formula, This is the kernel function.
[0031] RFE uses SVR as its base model, transforming the feature selection problem into a feature importance ranking problem. It ranks each feature according to its weight from highest to lowest importance, selecting the optimal number of features. i The ranking criteria for each feature are: Adjust model parameters: This invention utilizes a grid search method combined with prior knowledge to adjust parameters. The parameters to be adjusted and their ranges are as follows: the range of C is [2]. -8 ,2 8 The range of ] and γ is [2 -8 ,2 8 ].
[0032] The preprocessed dataset was input into the parameter-tuned SVR-RFE model to calculate feature importance. Features with the lowest importance were progressively removed to generate the optimal feature subset for predicting the remaining lifespan of the aluminum electrolytic cell. After model selection, 12 feature vectors were ultimately chosen as key features for predicting the remaining lifespan of the aluminum electrolytic cell. Ranked by importance, they are: effect coefficient, alumina concentration, fluoride salts, current density, operating voltage, molecular ratio, silicon content in the molten aluminum, aluminum output, electrolyte level, aluminum level, anode travel, and iron content in the molten aluminum.
[0033] Develop a prediction model for the remaining lifespan of XGBoost aluminum electrolytic cells; XGBoost is an ensemble algorithm that combines the features of linear and tree models. It approximates the target function quickly by performing a second-order Taylor expansion on the loss function and adds a regularization term to avoid overfitting, making XGBoost more advantageous in prediction accuracy. The construction of the XGBoost algorithm framework based on the tree model includes the following steps: In the formula, This represents the i-th input sample. Let be the predicted value for the i-th sample. Indicates the first t A decision tree, F This is the set of all regression trees.
[0034] The objective function consists of a loss function and regularization, and is expressed as follows: in, Indicates the true value of the sample; The loss function; For regularization terms, specifically: In the formula, T represents the number of leaf nodes in the decision tree. Let λ represent the output score of each leaf node in the decision tree, where λ is the coefficient of the leaf node. The leaf weight penalty regularization term.
[0035] XGBoost uses the error generated by the model composed of the first t-1 trees as a reference to build the t-th tree, thus continuously reducing the loss function. Therefore, the objective function can be rewritten as: The objective function is approximated using a second-order Taylor expansion, as shown below: In the formula ， ， They represent the first and second derivatives of the objective function, respectively.
[0036] Given a fixed tree structure, the optimal weights are determined by setting the first derivative to zero. The results are as follows: Substituting the obtained weights into the objective function yields the optimized objective function: In the formula, ， ， for Falling into a leaf node j The sample set.
[0037] This invention utilizes a grid search method combined with prior knowledge to adjust parameters. The parameters that need adjustment include: learning_rate, n_estimators, max_depth, γ, min_child_weight, subsample, and colsample_bytree.
[0038] Adjust the range of parameters: learning_rate is [0.01, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5]; n_estimators is [100, 800]; max_depth is [1, 6]; γ is [0.1, 1]; min_child_weight is [3, 20]; subsample is [0.1, 1]; colsample_bytree is [0.5, 0.9].
[0039] The selected key features are input into the parameter-tuned XGBoost algorithm framework for training, generating a prediction model for the remaining life of aluminum electrolytic cells. This invention uses a Python version of the XGBoost algorithm for modeling, with 80% of the data in the dataset used as the training set and 20% as the test set. Five-fold cross-validation is used to optimize the model parameters. Predicting the remaining lifespan of aluminum electrolysis cells; The test data is input into the aluminum electrolysis cell remaining life prediction model to calculate its remaining life. Figure 2 The prediction results are for the SVR-RFE-XGBoost model. Figure 3 A comparison chart of the prediction results of the current mainstream machine learning model and the SVR-RFE-XGBoost model is presented. As can be seen from the chart, the method provided by this invention has a higher degree of fit with the actual remaining life of aluminum electrolytic cells compared with the traditional mechanism model.
[0040] The technical effects of the present invention will be further illustrated by the following simulation experiments: Simulation conditions and content: The simulations for predicting the remaining life of aluminum electrolytic cells in this invention and the prior art are both implemented using the sklearn library in Python 3.6. 80% of the acquired dataset is used as the training sample set, and the remaining data is used as the test sample set.
[0041] Simulation Result Analysis: Using the prediction method based on the SVR-RFE-XGBoost model, this invention performs simulation prediction on the test sample set, and compares the performance results using the mean absolute error (MAE), mean squared error (MSE), and goodness of fit (r2), as shown in Table 1.
[0042] Table 1. Multi-model prediction results As can be seen from Table 1, the prediction error obtained by the method of the present invention is significantly lower than that of XGBoost, Ridge Regression, and SVR models, which verifies the accuracy improvement effect of the present invention on the prediction of the remaining life of aluminum electrolytic cells.
[0043] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.
Claims
1. A method for predicting the remaining life of aluminum electrolytic cells based on SVR-RFE-XGBoost, characterized in that, Includes the following steps: Historical data of aluminum electrolytic cells are acquired, and the historical data is preprocessed to construct a dataset, which consists of data features; An SVR-RFE model is constructed, the hyperparameters of the SVR-RFE model are adjusted, and the preprocessed dataset is input into the adjusted SVR-RFE model. After calculating the importance of features, the data features that affect the remaining life of the aluminum electrolytic cell are sorted by importance and selected to form a feature subset. Build the XGBoost algorithm framework; Parameters were adjusted for the XGBoost algorithm framework; The feature subset is input into the parameter-adjusted XGBoost algorithm framework for training to generate a prediction model for the remaining life of aluminum electrolytic cells. The characteristics of the data to be tested are input into the prediction model of the remaining life of the aluminum electrolytic cell to calculate the remaining life of the aluminum electrolytic cell.
2. The method for predicting the remaining life of an aluminum electrolytic cell based on SVR-RFE-XGBoost as described in claim 1, characterized in that, The preprocessing includes the following steps: Historical data is organized and cleaned to remove invalid and abnormal data features, retain valid data features, and fill in missing values with the mean. Standard deviation standardization is performed on the imputed data features to eliminate differences in the units and magnitudes of different data features, so that all data features are unified to the same scale. The processed data features can be represented as data vectors, and the data vectors shown are scaled to the [0,1] interval.
3. The method for predicting the remaining life of an aluminum electrolytic cell based on SVR-RFE-XGBoost as described in claim 1, characterized in that, The data features within the dataset include set voltage, operating voltage, average voltage, aluminum level, electrolyte level, molecular ratio, fluoride salt, tank temperature, anode stroke, aluminum output, effect coefficient, noise, current density, alumina concentration, current efficiency, cathode voltage drop, silicon content in the molten aluminum, iron content in the molten aluminum, and power consumption.
4. The method for predicting the remaining life of an aluminum electrolytic cell based on SVR-RFE-XGBoost as described in claim 2, characterized in that, The process of sorting and filtering data features that affect the remaining lifespan of aluminum electrolytic cells by importance to form a feature subset includes the following steps: Each data feature is sorted from highest to lowest importance according to its weight, and a predetermined number of data features are selected to form a feature subset.
5. The method for predicting the remaining life of an aluminum electrolytic cell based on SVR-RFE-XGBoost as described in claim 3, characterized in that, The preset quantity is 12.
6. The method for predicting the remaining life of an aluminum electrolytic cell based on SVR-RFE-XGBoost as described in claim 1, characterized in that, The parameter tuning of the XGBoost algorithm framework includes the following steps: The range of predefined parameters; K-fold cross-validation is performed using grid search. The parameter combination corresponding to the maximum efficiency returned after K-fold cross-validation is the optimal parameter of the XGBoost algorithm framework.
7. The method for predicting the remaining life of aluminum electrolytic cells based on SVR-RFE-XGBoost as described in claim 1, characterized in that, The order of importance, its first... i The ranking criteria for each feature are: 。 8. The method for predicting the remaining life of an aluminum electrolytic cell based on SVR-RFE-XGBoost as described in claim 1, characterized in that, The hyperparameters of the SVR-RFE model are adjusted, and the parameters and ranges to be adjusted are as follows: the range of C is [2]. -8 ,2 8 The range of ] and γ is [2 -8 ,2 8 ].
9. The method for predicting the remaining life of an aluminum electrolytic cell based on SVR-RFE-XGBoost as described in claim 1, characterized in that, The XGBoost algorithm framework based on the tree model is as follows: In the formula, This represents the i-th input sample. Let be the predicted value for the i-th sample. Indicates the first t A decision tree, F This is the set of all regression trees.
10. The method for predicting the remaining life of an aluminum electrolytic cell based on SVR-RFE-XGBoost as described in claim 9, characterized in that, The kernel function of the XGBoost algorithm framework based on the tree model is: 。