Molecular ratio optimization control model and prediction method under complex electrolyte system

Abstract

The invention relates to the technical field of automation control, and discloses a molecular ratio optimization control model and prediction method under a complex electrolyte system, and the model comprises a data collection module which is used for collecting cell control daily report and inspection data in an aluminum electrolysis production process as basic data; the feature selection module is used for screening core features; a prediction model is arranged in the model training and prediction module; a prediction model adopts a regression prediction model fused with a self-attention mechanism, and the prediction model takes the screened core features as input and outputs an aluminum fluoride addition amount prediction value through a dynamic weighting feature dependency relationship; the control execution module is used for converting the additive amount predicted value into an aluminum fluoride adding instruction and issuing the aluminum fluoride adding instruction to a production line control system to realize automatic execution of an adding task; and the man-machine interaction module is used for providing a visual interface to display information in real time. According to the invention, the addition amount of the aluminum fluoride can be accurately and efficiently predicted, and the intelligent and automatic control of the addition process of the aluminum fluoride is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of automation control technology, in particular to a complex electrolyte system under the molecular ratio optimization control model and prediction method. BACKGROUND

[0002] In the industrial aluminum electrolysis production process, maintaining the appropriate molecular ratio in the electrolyte system is the core element to achieve efficient, stable and energy-saving operation. The accurate control of the molecular ratio directly affects the heat balance of the electrolytic cell, the dissolution characteristics of alumina, the current efficiency and the energy consumption level. Specifically, the amount of aluminum fluoride added is the key operating variable to regulate the molecular ratio. Traditionally, the decision of the amount of aluminum fluoride added is highly dependent on the accumulation of experience of the operator and the regular offline electrolyte composition analysis.

[0003] This mode has significant limitations: first, manual experience is limited by individual differences and is difficult to quantify inheritance, which is prone to subjectivity; second, regular detection has serious hysteresis and cannot capture the dynamic changes of electrolyte composition in real time, especially when the production load fluctuates, the raw material characteristics change or the environmental conditions (such as ambient temperature) are disturbed, the hysteresis effect is more prominent; third, manual operation is slow and prone to errors, making it difficult to achieve fine and continuous control. As a result, the electrolyte molecular ratio is often in a non-ideal state, with large fluctuations and high frequency. The excessive fluctuation of the molecular ratio directly damages the stability of the production process, reduces the current efficiency (which means that the production of aluminum per unit power consumption decreases), increases energy consumption (especially invalid ohmic heat loss), and may increase the frequency of abnormal working conditions (such as anode effect) in the electrolytic cell, ultimately affecting the consistency of the primary aluminum product quality and production costs.

[0004] Therefore, with the continuous improvement of the modern aluminum industry's requirements for fine, intelligent, efficient and green production processes, it is urgent to abandon the traditional extensive control mode that relies on manual experience and low-frequency detection, and instead develop a complex electrolyte system molecular ratio optimization control model and prediction method that can be based on real-time data, with high-precision prediction capability and dynamic optimization and adjustment functions. SUMMARY

[0005] The present application aims to provide a complex electrolyte system under the molecular ratio optimization control model and prediction method, which can accurately and efficiently predict the amount of aluminum fluoride added, and realize intelligent and automated control of the aluminum fluoride addition process, helping to improve the overall technical and economic indicators and operational stability of the electrolytic production.

[0006] To achieve the above purpose, the present application provides the following basic scheme.

[0007] Scheme One A complex electrolyte system under the molecular ratio optimization control model, comprising: A data collection module is configured to collect cell control daily reports and inspection data in the aluminum electrolysis production process as basic data. A feature selection module is configured to calculate feature importance scores of the basic data based on a random forest algorithm and to filter a preset number of core features. A model training and prediction module is internally provided with a prediction model; the prediction model adopts a regression prediction model with a fusion self-attention mechanism, takes the filtered core features as input, and outputs an aluminum fluoride addition amount prediction value through dynamic weighting feature dependency; A control execution module is configured to convert the addition amount prediction value into an aluminum fluoride addition instruction and issue it to a production line control system to realize automatic execution of the addition task; the aluminum fluoride addition instruction includes an additive amount, a feeding rate, and a feeding duration. A human-computer interaction module is configured to provide a visual interface to display the running state of the electrolytic cell, the addition amount prediction value, and the molecular ratio threshold alarm information in real time.

[0008] Further, the cell control daily report includes cell control parameters, and the cell control parameters at least include current intensity, voltage, uniformity difference, temperature, aluminum level, quality level, alumina discharge amount, and furnace bottom pressure drop; the inspection data at least includes calcium fluoride content, sodium oxide content, potassium fluoride content, iron content, silicon content, and molecular ratio.

[0009] Further, the system further comprises a preprocessing module; the preprocessing module is configured to perform standardization processing, outlier removal, and missing value completion operations on the collected basic data.

[0010] Further, when performing outlier removal, the preprocessing module adopts the interquartile range method to remove outliers, including the following operations: S1, calculate the 25th percentile Q1 and the 75th percentile Q3 of each type of data to determine the interquartile range IQR = Q3-Q1; S2, set the outlier determination interval as [Q1-1.5×IQR, Q3+1.5×IQR], and remove data points outside the interval; S3, repeat S1-S2 until no outliers exist in the basic data, and use a linear interpolation algorithm to complete the missing data.

[0011] Further, the number of core features filtered by the feature selection module is the top 10 features ranked by feature importance scores, and includes calcium fluoride content, aluminum level, quality level, current intensity, voltage, and temperature; and the hyperparameters of the random forest algorithm are optimized by grid search or Bayesian optimization method.

[0012] Further, the regression prediction model is a gradient boosting decision tree model; the self-attention mechanism of the prediction model is implemented in the following manner: receiving a feature vector generated by a gradient boosting decision tree; calculating a correlation weight between pairs of features in the feature dimension or the time dimension for the feature vector; and generating a dynamically weighted feature expression by weighted fusion according to the correlation weight; mapping the weighted feature expression to a predicted value of the aluminum fluoride addition amount.

[0013] Further, the human-computer interaction module further comprises: a multi-level permission management unit for setting and distinguishing administrator configuration, operator viewing, and patrol alarm receiving permission; a data interface unit for providing an API for external MES / DCS systems to call the prediction result; a historical data tracing unit for supporting molecular ratio fluctuation trend analysis and early warning record query.

[0014] Further, the control execution module is provided with an abnormal response mechanism: When the predicted value makes the molecular ratio exceed the preset threshold for two consecutive rounds, a model re-calculation verification process is triggered, and the prediction model is re-called for verification; When the verification result is still abnormal, an alarm information is automatically pushed to a specified terminal device; The control execution module also communicates with a production line PLC system in real time, and synchronously starts an audible and visual alarm device when the alarm information is triggered.

[0015] Further, the prediction model supports collaborative prediction and centralized scheduling of multiple electrolytic cells, and the control execution module realizes automatic management of the addition task of the whole production line through a unified control center.

[0016] Scheme two A molecular ratio optimization prediction method under a complex electrolyte system, which applies a molecular ratio optimization control model under a complex electrolyte system as described in scheme one to perform molecular ratio optimization prediction, and comprises the following steps: Collecting tank control daily reports and inspection data, and generating basic data after pretreatment as a training data set; Using a random forest algorithm to screen core features, and constructing a regression prediction model fused with a self-attention mechanism; Setting a molecular ratio threshold through a visual interface, and outputting a predicted value of the aluminum fluoride addition amount in real time; Converting the predicted value into a control instruction and issuing it to an electrolytic cell control system to dynamically adjust the addition strategy; When the predicted value makes the molecular ratio exceed the preset threshold for two consecutive rounds, an alarm mechanism is started, and the processing result is closed-loop tracked.

[0017] The working principle and advantages of the present application are that: The application discloses a complex electrolyte system under a molecular ratio optimization control model and a prediction method, which can accurately and efficiently predict the aluminum fluoride addition amount, realizes intelligent and automatic control of the aluminum fluoride addition process, and helps improve the overall technical and economic indicators and operation stability of electrolytic production. First, the scheme constructs a multi-level optimization architecture for a complex electrolyte system, which can maintain high prediction accuracy and efficiency under a complex electrolyte system.

[0018] Specifically, the scheme first realizes feature dimension reduction and importance screening of high-dimensional industrial data through a random forest algorithm, effectively overcoming the overfitting and low computational efficiency problems caused by redundant input parameters in traditional control models. Moreover, the feature selection unit adopts a random forest instead of a conventional correlation coefficient method, which can better handle the nonlinear relationship between industrial parameters and form a match with the gradient boosting decision tree with the subsequent fusion attention mechanism. Further, the scheme innovatively fuses the gradient boosting decision tree with the self-attention mechanism, which enables the model to have dual capabilities: the gradient boosting decision tree captures local nonlinear relationships through iterative optimization of residuals, while the self-attention mechanism models global feature dependencies through dynamic weight distribution, compensating for the shortcomings of tree models in modeling feature interaction relationships, and the two work together to significantly enhance the model's representation ability for the multi-variable strong coupling characteristics of the electrolyte system.

[0019] At the control execution layer, the system realizes fast response from prediction to execution through a real-time data flow closed-loop feedback mechanism (including threshold trigger re-computation verification, PLC linkage response, etc.), solving the lagging defects of traditional manual control. The formed man-machine cooperation framework not only guarantees the automation degree of control decision, but also retains key manual intervention nodes through a visual interface, improving accuracy while ensuring the operation safety of the industrial system. In addition, the model provided by the scheme is deployed on the production line, which can automatically execute the aluminum fluoride addition task, greatly reducing manual intervention and improving production automation level and management efficiency.

[0020] Second, although random forest, gradient boosting decision tree and self-attention mechanism are known algorithms, their combination application in the field of electrolyte molecular ratio control and solution to specific technical problems has significant creativity.

[0021] First, industrial electrolyte systems face unique constraints such as high data acquisition noise and complex variable coupling in high-temperature, highly corrosive environments. This solution creatively embeds a self-attention mechanism into the tree model output layer rather than the original input layer. By first extracting high-order feature representations through a gradient boosting decision tree and then redistributing feature dimension weights through the attention mechanism, this design effectively mitigates the interference of raw industrial data noise on the attention weight calculation. Secondly, to address the real-time requirements of aluminum fluoride addition control, this solution incorporates a dual-threshold mechanism for abnormal response (two consecutive limit violations trigger verification). This strategy differs from simple single-threshold alarms and fixed-period verification by dynamically adjusting verification intensity based on system confidence, resulting in higher reliability. Furthermore, traditional process control models typically optimize either algorithm accuracy or execution efficiency separately. This solution, by synergizing a feature selection unit with a lightweight attention mechanism, achieves a dual-effect balance while ensuring prediction accuracy while meeting the time constraints of real-time industrial control. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic diagram of the model structure of an embodiment of a molecular ratio optimization control model and prediction method for a complex electrolyte system according to the present invention; Figure 2 The figure is a flow chart of a method embodiment of a molecular ratio optimization control model and prediction method in a complex electrolyte system according to the present invention. DETAILED DESCRIPTION

[0023] The following is a further detailed description through specific implementation methods: The embodiment is basically as shown in the attached Figure 1 Shown: A molecular ratio optimization control model for a complex electrolyte system, including: The data acquisition module is used to collect daily cell control reports and inspection data from the aluminum electrolysis production process as basic data. The daily cell control reports contain cell control parameters, which include at least current intensity, voltage (including cell operating voltage, anode effect voltage, inter-electrode voltage, and electrolyte resistance voltage), mean square error (i.e., the mean square error of anode current per cell), temperature (including electrolyte temperature), aluminum level, quality level (i.e., electrolyte level), alumina feed rate, and furnace bottom pressure drop. The inspection data includes at least calcium fluoride content, sodium oxide content, potassium fluoride content, iron content, silicon content, and molecular weight ratio.

[0024] Preprocessing module: The preprocessing module is used to perform standardization processing, outlier removal and missing value filling operations on the collected basic data to ensure the accuracy and completeness of the data.

[0025] When performing outlier removal, the preprocessing module uses the interquartile range method to remove outliers, including the following operations: S1, calculate the 25th percentile Q1 and 75th percentile Q3 of each type of data, and determine the interquartile range IQR=Q3-Q1; S2, set the outlier judgment interval to [Q1-1.5×IQR, Q3+1.5×IQR] and remove data points outside this interval; S3, repeat S1-S2 until there are no abnormal points in the basic data. At the same time, the missing data are supplemented by linear interpolation algorithm.

[0026] By eliminating outliers through multiple iterations of the interquartile range method, extreme values ​​in the data can be gradually and accurately identified and removed, avoiding anomalies that may be missed in a single processing; at the same time, the linear interpolation algorithm fills in the missing data, which not only retains the original distribution characteristics of the data but also ensures the continuity of the data, and can provide a high-quality, complete and reliable data foundation for subsequent model training, thereby effectively improving the stability of the model and the prediction accuracy.

[0027] The feature selection module is used to calculate the feature importance score based on the basic data through the random forest algorithm and filter out a preset number of core features.

[0028] The number of core features screened by the feature selection module is the top 10 features ranked by feature importance score, including calcium fluoride content, aluminum level, quality level, current intensity, voltage and temperature; and the hyperparameters of the random forest algorithm are tuned by grid search or Bayesian optimization method.

[0029] Specifically, the feature selection module takes basic data as input and uses the random forest algorithm to calculate the feature importance score. : ; in, is the total number of decision trees, is the Gini impurity before node splitting; Weighted Gini impurity for the child nodes after splitting; right Perform sliding window normalization and obtain .

[0030] Pick The features are selected as candidate features. When the number of candidate features is greater than 10, press Take the first 10 items in descending order. In this embodiment, The value is 0.8.

[0031] When tuning the hyperparameters of the random forest algorithm, we first search for the optimal tree depth and number of split features through Bayesian optimization; then, we select the parameter combination that minimizes the MAE of the molecular ratio prediction based on the time series cross-validation method.

[0032] The feature selection effect can be further optimized by adjusting the model hyperparameters, thereby providing reliable data support for subsequent aluminum fluoride addition amount prediction, and improving the model generalization ability and prediction accuracy.

[0033] The model training and prediction module is internally provided with a prediction model; the prediction model adopts a regression prediction model fused with a self-attention mechanism, takes the screened core features as input, and outputs the aluminum fluoride addition amount prediction value through dynamic weighting feature dependency. In the embodiment, the regression prediction model is a gradient boosting decision tree model; specifically, an XGBoost model, which has better efficiency and flexibility and is suitable for complex industrial data analysis scenarios.

[0034] Further, the self-attention mechanism of the prediction model is implemented in the following manner: Receiving the feature vector generated by the gradient boosting decision tree, which corresponds to the feature vector in the subsequent steps ; For the feature vector, the correlation weight between pairs of features is calculated in the feature dimension or time dimension, that is, the correlation weight between different time steps or feature dimensions is calculated, which corresponds to step (3) in the subsequent steps; according to the correlation weight, a dynamic weighted feature expression is generated through weighted fusion, which corresponds to step (4) in the subsequent steps, thereby improving the fitting ability of the model to complex nonlinear relationships and further improving the prediction accuracy; Mapping the weighted feature expression to the aluminum fluoride addition amount prediction value, which corresponds to steps (5)-(7) in the subsequent steps.

[0035] Specifically, the 10 core features screened are input into the XGBoost model and generate a regression decision tree, in this embodiment, the default is 100. Each tree outputs a leaf node index ,

[0036] The feature vector ; is represented by a real number set, and the dimension of the real number matrix is d, where is the number of trees, and d is the total number of leaf nodes, that is, the feature dimension corresponding to the output of each tree (for example, the length of the vector after encoding the leaf node index of each tree).

[0037] . Among them, represents vector splicing, and the leaf node index is one-hot encoded, which can convert the decision logic of the tree model into numerical features and retain fine-grained decision information.

[0038] Again based on the feature vector Perform multi-head attention calculation and concatenate the multi-head output along the feature dimension as the final output feature vector Z, including the following sub-steps: (1) The feature vector is taken as input and fed into a multi-head attention layer.

[0039] (2) For each attention head, the query vector Q, the key vector K and the value vector V are linearly mapped from the original feature vector , , by a learnable weight matrix : ; K ; V .

[0040] The purpose of this step is to map the original feature vector to different representation spaces, preparing for the subsequent calculation of feature correlation.

[0041] (3) Calculate the attention weight matrix A, whose each element represents the correlation between the i-th feature and the j-th feature: ; where “ ” in the matrix multiplication, the similarity between the query vector Q and the key vector K is calculated by matrix multiplication, and each element in the resulting matrix reflects the degree of association between different features (represented by the dimensions corresponding to the query and the key). is the dimension of the key vector; softmax is a normalization function used to normalize the similarity matrix, so that each attention weight is between 0 and 1, and the sum of each row weight is 1.

[0042] (4) Weighted sum the value vector V using the attention weight matrix A to get the output of this attention head. The purpose of this step is to aggregate the value vector according to the correlation between features (attention weights).

[0043] (5) Concatenate the output vectors , , of all attention heads (H is the number of attention heads), and project them through a final linear layer to get the final output feature vector Z of the self-attention mechanism: .

[0044] in, Represents the concatenation operation. Through the concatenation operation, the diverse feature information captured by different attention heads can be integrated. Combined with the linear layer for linear projection, the concatenated high-dimensional vector can be mapped to the appropriate dimension to obtain the final feature representation.

[0045] (6) The feature vector Z rich in global dependency information is input into a fully connected regression layer.

[0046] (7) The fully connected regression layer uses a learnable weight vector and bias , perform a linear transformation on the eigenvector Z and map it to a scalar value, which is the final predicted value of ammonium fluoride addition : .

[0047] In the above steps 、 、 、 、 and Both are obtained by iterative optimization during the model training phase by minimizing the loss function (such as mean square error loss MSE) between the predicted value and the true value.

[0048] In this embodiment, the set model training stage specifically includes: Based on the training dataset, the dataset was labeled with the actual amount of aluminum fluoride added and divided into a training set, a validation set, and a test set (in this example, the ratio was set to 6:2:2). A mean squared error function was used as the loss function, and the prediction model was trained iteratively over multiple rounds using the training set data. Each round included the following sub-steps: forward propagation, backpropagation, and parameter update. The specific steps of forward propagation, backpropagation, and parameter update were based on existing techniques and are not detailed here.

[0049] Monitor the training process based on the validation set loss. When the loss stops decreasing, stop training and save the model parameters with the best performance on the validation set. Use the saved model to predict the test set and evaluate the final generalization performance of the model. Once it meets the standards, deploy it to the model training and prediction module.

[0050] Among them, the compliance criteria are: the mean absolute error (MAE) on the test set is lower than the preset threshold of 1; the coefficient of determination on the test set ( ) is higher than the preset threshold value 2. The threshold value 1 and the threshold value 2 are set according to the precision requirements of the aluminum electrolysis production process for the addition amount. In this embodiment, it can be set to , .

[0051] Through the above settings, the prediction model can optimize the model parameters through the gradient boosting decision tree, and use the self-attention mechanism to enhance the ability to capture long-distance dependence between features, thereby improving the prediction accuracy. Moreover, the weight parameter matrix is continuously iteratively optimized in the training process of the prediction model to ensure accurate prediction of the daily aluminum fluoride addition amount.

[0052] The control execution module is configured to convert the addition amount prediction value into an aluminum fluoride addition instruction and issue it to the production line control system to realize automatic execution of the addition task.

[0053] Specifically, in the embodiment, the aluminum fluoride addition instruction includes the addition amount, the feeding rate, and the feeding duration.

[0054] When converting the addition amount prediction value, the control execution module includes the following steps: The addition amount prediction value is calculated into three-dimensional control parameters, namely, the addition amount D (in kg), the feeding rate V (in kg / h), and the feeding duration (in h).

[0055] The addition amount D is calculated as follows: That is, the mass value predicted by the model is directly adopted.

[0056] The feeding rate V needs to be set according to the actual application of the electrolytic cell type and the rated conveying capacity of the feeding execution mechanism (for example, the maximum conveying capacity corresponding to the maximum speed of the feeding motor).

[0057] The feeding duration t is calculated as follows: ; n is a compensation coefficient, which is obtained by calibration according to the actual application of the electrolytic cell structure and the feeding execution mechanism, and is used to correct the actual feeding efficiency (such as pipeline resistance and material flowability) to ensure accurate addition amount.

[0058] After calculating the three-dimensional control parameters, they are packaged into a standard data packet and transmitted to the PLC controller of the feeding execution mechanism (such as the aluminum fluoride feeding machine) of the production line control system to realize automatic execution of the addition task.

[0059] The control execution module is provided with an abnormal response mechanism: When the prediction value exceeds the preset threshold for two consecutive rounds, the model re-computation verification process is triggered, and the prediction model is re-called for verification. For example, the preset threshold is set to 2.2±0.1 (which can be adjusted according to actual application requirements), and when the molecular ratio detection value of a certain electrolytic cell exceeds this range (i.e. <2.1 or >2.3) for two consecutive rounds (each round interval is 2 hours), the model re-computation verification process is automatically triggered.

[0060] Preferably, each round of molecular ratio detection value needs to meet: the chemical analysis confirmation data uploaded by the laboratory information management system (LIMS), and the associated tank control parameters in this period have no major adjustment (such as current intensity fluctuation ≤±50A, alumina discharge amount deviation ≤±2%), otherwise it is judged as invalid threshold value, and the process is not triggered.

[0061] When the verification result is still abnormal, automatically push the alarm information to the specified terminal device, such as pushing the alarm information to the terminal device of the on-duty personnel through SMS, email and other ways, to realize real-time response and closed-loop management.

[0062] The control execution module also communicates with the production line PLC system in real time, and synchronously starts the sound and light alarm device set in the production line when the alarm information is triggered. Specifically, in the embodiment, the control execution module is connected with the production line PLC system through industrial Ethernet or Modbus protocol.

[0063] The prediction model supports collaborative prediction and centralized scheduling of multiple electrolytic cells. The control execution module realizes the automatic management of adding tasks of the whole production line through the unified control center, has good scalability and compatibility, and can adapt to multiple types of electrolytic cells and industrial control system platforms. Moreover, in actual application, the unified control center is used to realize the automatic issuance and centralized monitoring of aluminum fluoride addition tasks. After applying the prediction model, the fluctuation range of electrolyte molecular ratio is significantly reduced, and the reduction is 50% compared with manual strategy adjustment, which significantly improves the current efficiency and achieves the goal of energy saving and consumption reduction.

[0064] The human-computer interaction module is used to provide a visual interface to display the running state of the electrolytic cell, the addition amount prediction value and the molecular ratio threshold alarm information in real time. The real-time visual interface enables the operator to intuitively master the state of the electrolytic cell, the prediction data and the alarm information, and facilitates timely decision-making.

[0065] The human-computer interaction module further comprises: The multi-level permission management unit is used to set and distinguish administrator configuration, operator viewing and patrol alarm receiving permission; The data interface unit is used to provide API for external MES / DCS system to call the prediction result. Through the integration of API and external MES / DCS system, data interconnection between production management systems can be realized, and higher level of automatic control can be supported.

[0066] The historical data tracing unit is used to support molecular ratio fluctuation trend analysis and early warning record query; can provide data support for process optimization and fault troubleshooting, and form a closed-loop management system from real-time monitoring to historical analysis.

[0067] By integrating visual display and multi-dimensional functional units, the efficiency and safety of electrolytic cell operation management can be further improved.

[0068] As Figure 2 shown, the embodiment also provides a method for predicting the optimization of the molecular ratio under a complex electrolyte system, which applies the above-mentioned control model for optimizing the molecular ratio under a complex electrolyte system to predict the optimization of the molecular ratio, including the following steps: Step 1, collect the daily reports and inspection data of the tank control, and generate the basic data after preprocessing as the training data set.

[0069] Step 2, use the random forest algorithm to screen the core features, and construct a regression prediction model integrating the self-attention mechanism.

[0070] Specifically, the preprocessed data set is input into the random forest feature selection unit, and the feature importance score of each parameter on the molecular ratio fluctuation is calculated; The top 10 features (including calcium fluoride content, aluminum level, quality level, average current intensity, voltage variance, temperature gradient value) with the highest scores are selected to generate a feature subset; The feature subset is input into the regression prediction model integrating the self-attention mechanism: the gradient boosting decision tree first generates a high-order feature vector representation; then based on the multi-head self-attention layer, the correlation matrix in the feature dimension is calculated to generate a weight distribution coefficient; the feature vector is weighted and fused according to the weight coefficient, and the predicted value of the aluminum fluoride addition amount is output.

[0071] Step 3, set the molecular ratio threshold through the visualization interface, and output the predicted value of the aluminum fluoride addition amount in real time.

[0072] Specifically, the predicted addition amount and the real-time molecular ratio trend curve are displayed on the visualization interface; the target molecular ratio threshold interval set by the operator is received .

[0073] Step 4, convert the predicted value into a control instruction and issue it to the electrolytic tank control system to dynamically adjust the addition strategy.

[0074] Specifically, the approved addition amount prediction value is converted into a PLC control instruction; the control instruction includes three-dimensional parameters of the additive amount, the feeding rate and the feeding duration.

[0075] Step 5, when the predicted value makes the molecular ratio exceed the preset threshold for two consecutive rounds, start the alarm mechanism and close-loop track the processing results.

[0076] When the molecular ratio corresponding to the predicted addition amount exceeds , highlight the warning area and start the abnormal response mechanism.

[0077] Preferably, when the real-time detected molecular ratio deviates from for two consecutive periods, trigger the alarm signal and freeze the automatic addition program, and switch to the manual review mode.

[0078] The embodiment provides a molecular ratio optimization control model and a prediction method under a complex electrolyte system, key process parameters and test data in aluminum electrolysis production are collected, a high-precision prediction model is constructed, visual scheduling and real-time deployment are combined, intelligent and automatic control of an aluminum fluoride adding process can be realized, prediction precision and response speed are significantly improved, electrolyte molecular ratio fluctuation is effectively reduced, current efficiency and energy utilization rate are improved, the model and the method have good application and popularization value, and are suitable for energy saving and consumption reduction requirements in modern aluminum electrolysis production.

[0079] In addition, in order to verify the application effect of the model and the method, the traditional method and the scheme are applied respectively for two months on the same aluminum electrolysis production line.

[0080] The scheme comprises the following steps: applying the molecular ratio optimization control model under the complex electrolyte system provided by the embodiment, automatically predicting an aluminum fluoride adding amount prediction value by a prediction model, converting the aluminum fluoride adding amount prediction value into an aluminum fluoride adding instruction containing an adding amount, an adding rate and an adding duration by a control execution module, and controlling an adding mechanism in a production line to add.

[0081] The traditional method comprises the following steps: taking an electrolyte sample once per period (for example, taking 4 hours as one period), detecting the molecular ratio, confirming the adding amount according to experience, for example, confirming the adding amount according to an experience formula: , and adding by an operator.

[0082] Obtained implementation data is shown in Table 1. Table 1

[0083] The molecular ratio fluctuation range is a standard deviation of the measured molecular ratio value in a statistical period, and is expressed as Data is derived from periodic chemical composition analysis results of a laboratory information management system (LIMS).

[0084] The molecular ratio qualified rate is a percentage of a number of samples, in which the measured molecular ratio value falls within a target control interval (for example, 2.2±0.1), in total sample numbers in a statistical period (two months). The calculation formula is as follows: .

[0085] The continuous stable duration is a continuous duration in which the measured molecular ratio value continuously falls within the target control interval, and an average value of each continuous qualified duration in the statistical period is taken. Data is derived from historical records with time stamps in the LIMS system.

[0086] From the implementation data, after applying the scheme, the control accuracy of the molecular ratio is significantly improved, the abnormal working conditions are obviously inhibited (the qualified rate of the molecular ratio is improved), the production stability is improved (the fluctuation range is small and stable). And compared with the traditional method, the control efficiency of the scheme is higher, without adding the amount of artificial experience and automatic closed-loop control can be realized.

[0087] The above results fully verify the effectiveness of the design of each module in the scheme. Through random forest feature selection (precise elimination of interference parameters) and XGBoost-self attention fusion model (tree model extracts local features + attention mechanism dynamically weights global dependence), the problem that the traditional artificial experience cannot capture the multivariate strong coupling relationship is solved, the prediction error of the molecular ratio is reduced, and the qualified rate of the molecular ratio is increased; in the control layer, three-dimensional instruction calculation (dose-rate-time) is detailed, combined with the abnormal response mechanism (continuous over-limit trigger verification), the effective control of the molecular ratio fluctuation is realized.

[0088] The above is only an embodiment of the present application, and the well-known specific structures and characteristics of the scheme are not described in detail, and the ordinary skilled person in the art knows all the ordinary technical knowledge in the technical field of the application before the application date or the priority date, can know all the prior art in the field, and has the ability to apply conventional experimental means before the date, and the ordinary skilled person in the art can improve and implement the scheme under the guidance of the present application, some typical known structures or known methods should not be an obstacle for the ordinary skilled person in the art to implement the present application. It should be pointed out that for those skilled in the art, without departing from the structure of the present application, a number of modifications and improvements can be made, which should be considered as the protection scope of the present application, which will not affect the effect and practicality of the patent.

Claims

1. A molecular ratio optimization control model for a complex electrolyte system, characterized in that: include: Data acquisition module, used to collect daily cell control and inspection data during the aluminum electrolysis production process as basic data; The preprocessing module is used to perform standardization, outlier removal, and missing value filling operations on the collected basic data; The feature selection module is used to calculate the feature importance score based on the basic data through the random forest algorithm and select a preset number of core features; A model training and prediction module includes a prediction model; the prediction model uses a regression prediction model integrated with a self-attention mechanism. The prediction model takes the screened core features as input and outputs a predicted value of the aluminum fluoride addition amount through a dynamic weighted feature dependency relationship; A control execution module is used to convert the addition amount prediction value into an aluminum fluoride addition instruction and send it to the production line control system to realize the automatic execution of the addition task; the aluminum fluoride addition instruction includes the addition amount, the addition rate and the addition duration; The human-computer interaction module is used to provide a visual interface to display the electrolytic cell operating status, addition amount prediction value and molecular ratio threshold alarm information in real time.

2. The molecular ratio optimization control model for a complex electrolyte system according to claim 1, characterized in that: The daily cell control report includes cell control parameters, which include at least current intensity, voltage, average setting difference, temperature, aluminum level, quality level, alumina discharge amount, and furnace bottom pressure drop; the inspection data include at least calcium fluoride content, sodium oxide content, potassium fluoride content, iron content, silicon content, and molecular ratio.

3. The molecular ratio optimization control model for a complex electrolyte system according to claim 1, characterized in that: When performing outlier removal, the preprocessing module uses the interquartile range method to remove outliers, including the following operations: S1, calculate the 25th percentile Q1 and 75th percentile Q3 of each type of data, and determine the interquartile range IQR=Q3-Q1; S2, set the outlier judgment interval to [Q1-1.5×IQR, Q3+1.5×IQR] and remove data points outside this interval; S3, repeat S1-S2 until there are no abnormal points in the basic data. At the same time, the missing data are supplemented by linear interpolation algorithm.

4. The molecular ratio optimization control model for a complex electrolyte system according to claim 1, characterized in that: The number of core features screened by the feature selection module is the top 10 features ranked by feature importance score, including calcium fluoride content, aluminum level, quality level, current intensity, voltage and temperature; and the hyperparameters of the random forest algorithm are tuned by grid search or Bayesian optimization method.

5. The molecular ratio optimization control model for a complex electrolyte system according to claim 1, characterized in that: The regression prediction model is a gradient boosting decision tree model; the self-attention mechanism of the prediction model is implemented in the following way: Receive the feature vector generated by the gradient boosting decision tree; For the feature vector, the correlation weights between paired features are calculated in the feature dimension or time dimension; according to the correlation weights, a dynamic weighted feature expression is generated through weighted fusion; The weighted feature expression is mapped to the predicted value of aluminum fluoride addition.

6. The molecular ratio optimization control model for a complex electrolyte system according to claim 1, characterized in that: The human-computer interaction module also includes: Multi-level authority management unit, used to set and distinguish the administrator configuration, operator viewing and inspector alarm receiving permissions; Data interface unit, used to provide API for external MES / DCS system to call prediction results; The historical data tracing unit is used to support molecular ratio fluctuation trend analysis and early warning record query.

7. The molecular ratio optimization control model for a complex electrolyte system according to claim 1, characterized in that: The control execution module is provided with an abnormal response mechanism: When the predicted value causes the molecular ratio to exceed the preset threshold for two consecutive rounds, the model recalculation and verification process is triggered, and the prediction model is re-called for verification; If the verification result is still abnormal, the alarm information will be automatically pushed to the designated terminal device; The control execution module also communicates with the production line PLC system in real time and synchronously starts the sound and light alarm device when the alarm information is triggered.

8. The molecular ratio optimization control model for a complex electrolyte system according to claim 1, characterized in that: The prediction model supports collaborative prediction and centralized scheduling of multiple electrolytic cells, and the control execution module realizes automated management of added tasks for the entire production line through a unified control center.

9. A molecular ratio optimization prediction method in a complex electrolyte system, characterized in that: Applying the molecular ratio optimization control model in a complex electrolyte system according to any one of claims 1 to 8 to perform molecular ratio optimization prediction comprises the following steps: Collect daily tank control and inspection data, generate basic data after preprocessing, and use it as a training data set; Use the random forest algorithm to screen core features and build a regression prediction model that integrates the self-attention mechanism; Set the molecular ratio threshold through the visual interface and output the predicted value of aluminum fluoride addition in real time; The predicted value is converted into a control instruction and sent to the electrolyzer control system to dynamically adjust the addition strategy; When the predicted value causes the molecular ratio to exceed the preset threshold for two consecutive rounds, an alarm mechanism is activated and the processing results are tracked in a closed loop.

Images