A method for processing intelligent model data applied to metal smelting

Abstract

The present invention discloses a method for processing intelligent model data applied to metal smelting, and relates to the technical field of metal smelting. The specific steps of the method are: S100 real-time collection of metal smelting data, S200 establishment based on different sub-models, S300 each sub-model outputs prediction results and performs uncertainty assessment, S400 intelligently integrates the output results to obtain the most comprehensive and optimal decision output, S500 continuously monitors the actual data in the production process, and adjusts the equipment operation parameters. The present invention designs a multi-objective model output fusion mechanism that can simultaneously consider product quality, energy consumption and production efficiency in the metal smelting process. Through the weighted average method, the model prediction results corresponding to different objectives are intelligently integrated to obtain the most comprehensive and optimal decision output, so that the metal smelting process can achieve balance and optimization among multiple key indicators, and meet the needs of enterprises for maximizing comprehensive benefits.

Description

Technical Field

[0001] The present invention relates to the technical field of metal smelting, and in particular to a method for processing intelligent model data applied to metal smelting. Background Art

[0002] In today's industrial production field, metal smelting, as a basic and key industry, plays an important supporting role in economic development. With the increasingly fierce market competition and the continuous improvement of the requirements for sustainable development, metal smelting companies are facing many challenges and opportunities. On the one hand, customers have increasingly stringent quality requirements for metal products, not only requiring them to have good physical properties and chemical composition stability, but also expecting products to meet higher standards in terms of microstructure and surface quality to meet the precision needs of the high-end equipment manufacturing industry. On the other hand, the continuous rise in energy costs requires companies to attach great importance to the control of energy consumption and the improvement of production efficiency in order to reduce production costs and reduce the impact on the environment.

[0003] However, the existing metal smelting data processing technology has obvious limitations in dealing with these complex demands. In terms of data fusion and model construction, traditional methods mostly only optimize for a single goal. For example, when only focusing on improving product quality, they often ignore the sharp increase in energy consumption and the decline in production efficiency. They fail to fully explore the inherent correlation and mutual influence mechanism between various goals, and lack effective multi-objective optimization strategies. At the same time, the existing technology also has shortcomings in dealing with data uncertainty issues. Uncertain factors are not reasonably incorporated into the model decision-making process, making it difficult for the model to make robust and flexible decisions when facing complex and changeable smelting conditions, and cannot meet the urgent needs of modern metal smelting enterprises for refined and intelligent production management.

[0004] In summary, current metal smelting urgently needs an innovative data processing method that can comprehensively and effectively integrate multi-source data, establish accurate intelligent models, and use multi-objective optimization mechanisms while fully considering data uncertainty to enhance the adaptability of the model and the reliability of decision-making. Summary of the invention

[0005] The purpose of the present invention is to make up for the shortcomings of the prior art and to provide a method for processing intelligent model data applied to metal smelting, which can bring comprehensive benefit improvement to metal smelting enterprises through a multi-objective optimization intelligent model output fusion mechanism. By accurately integrating model prediction results in multiple aspects of product quality, energy consumption and production efficiency, enterprises can significantly reduce energy consumption while ensuring high-quality product output.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for processing intelligent model data applied to metal smelting, wherein the specific steps of the method are:

[0007] S100, collect data related to product quality, energy consumption, and production efficiency in all links of metal smelting in real time, and clean and remove noise from the collected data;

[0008] S200, establishing different sub-models for product quality, energy consumption and production efficiency targets in the metal smelting process, wherein the sub-models include a sub-model based on neural network prediction for product quality targets, a sub-model based on linear regression algorithm for energy consumption targets, and a sub-model based on decision tree evaluation for production efficiency targets, and independently training each sub-model using historical data and actual production data, so that each sub-model has the ability to predict and analyze the corresponding targets;

[0009] S300, each sub-model performs prediction calculation based on the real-time collected data, outputs its own prediction results on product quality, energy consumption and production efficiency, and performs uncertainty assessment on the output results of each sub-model, where:

[0010] The uncertainty evaluation index of the product quality is set as the variance of the prediction result

[0011] The uncertainty evaluation index of energy consumption is set as the variance of the residual of the prediction result

[0012] The uncertainty evaluation index of the production efficiency is set as the mean value A of the prediction result accuracy pe ;

[0013] S400, intelligently integrating the output results of each sub-model by weighted average to obtain the comprehensive optimal decision output;

[0014] The weighted average fusion method includes assigning weights to the output results of each sub-model according to the importance of each target in actual production, and multiplying the output results of each sub-model by its corresponding weight and adding them together to obtain the weighted average fusion result. The fusion result is used as the final decision output to guide the adjustment of equipment operating parameters in the metal smelting process.

[0015] S500. Apply the fused decision output to the actual production process of metal smelting, continuously monitor the actual data in the production process, compare and analyze the actual production results with the predicted results of the fusion model, and make corresponding adjustments and optimizations to the equipment operating parameters and process control parameters to achieve continuous optimization and closed-loop control of the metal smelting process.

[0016] Furthermore, the data related to the product quality in S100 include the hardness value A1 of the finished metal product, the strength value A2 of the finished metal product, the purity value A3 of the finished metal product, the number of internal defects of the finished metal product A4, the type of internal defects of the finished metal product A5, and the surface roughness value A6 of the finished metal product;

[0017] The energy consumption related data include the power consumption value B1 of each stage of the smelting process, the fuel consumption flow value B2 of each stage of the smelting process, and the thermal energy utilization rate value B3 of the smelting process;

[0018] The production efficiency-related data include the metal output value per unit time C1, the equipment operation time ratio C2, the residence time value of each link in the production process C3, the number of equipment failure shutdowns C4, and the equipment failure shutdown duration C5.

[0019] Furthermore, the S200 establishes a sub-model based on neural network prediction for the product quality target, the model includes an input layer, a hidden layer and an output layer, wherein:

[0020] The input layer is composed of neurons related to product quality parameter data, and all collected product quality parameter data are used as input values, that is, the input layer neurons include A1, A2, A3, A4, A5 and A6;

[0021] The hidden layer is provided with n neurons, which extract features from the data of the input layer. The input layer and the hidden layer are connected through a weight matrix W1, whose dimension is n×p, where p is the number of neurons in the input layer and each element represents the connection weight between the i-th neuron in the input layer and the j-th neuron in the hidden layer. At the same time, the hidden layer also has a bias vector b1, whose dimension is n×1, and each element represents the bias value of the jth neuron in the hidden layer. The activation function of the hidden layer neuron is f1(·), that is, f1(x)=max(0,x), and its output h j The calculation formula is: where j = 1, 2, ..., n, x i is the value of the i-th neuron in the input layer, i.e., the parameter data values ​​A1, A2, A3, A4, A5 and A6 of the corresponding product quality;

[0022] The output layer is responsible for outputting the final product quality prediction value, which is connected to the hidden layer through the weight matrix W2, with a dimension of 1×n, and each element represents the connection weight between the jth neuron in the hidden layer and the neurons in the output layer. The output layer also has a bias vector b2. The activation function of the neurons in the output layer is f2(·), that is, f2(x) = x. The output of the neurons in the output layer is used as the product quality prediction value y pqThe calculation formula is: The weight matrices W1, W2 and bias vectors b1, b2 are adjusted by inputting product quality parameter data, and the model parameters are continuously optimized iteratively, so that the model can accurately predict the quality of metal products based on the input quality-related parameters, thereby providing a reliable decision-making basis for quality control and optimization in the metal smelting process.

[0023] Furthermore, the S200 establishes a sub-model based on a linear regression algorithm for the energy consumption target. By linearly combining the parameters related to energy consumption, a quantitative relationship model between energy consumption and these parameters is established to achieve accurate prediction of energy consumption, providing effective support for energy management and cost control. The model uses B1, B2, and B3 as independent variables and converts the energy consumption prediction value y er As the dependent variable, the constructed linear regression equation is: er =w4B1+w5B2+w6B3+w7, where w 4、 w 5、 w6 is the regression coefficient, which reflects the influence of each independent variable on energy consumption; w7 is the intercept, which represents the basic value of energy consumption when the independent variable is zero. The linear regression equation is fitted by the least squares method, the purpose of which is to minimize the difference between the actual energy consumption value and the predicted value y er The sum of squared errors is used to find the best regression coefficient and intercept value so that the predicted value of the model is close to the actual value. The calculation formula of the sum of squared errors is: Among them, q is the number of samples, y is erk , B 1k , B 2k , B 3k are the actual energy consumption value of the kth sample and the corresponding independent variable value, respectively. Using historical data and energy consumption-related data samples collected in the actual production process, the estimated values ​​of regression coefficients w4, w5, w6 and intercept w7 are iteratively updated by solving the problem of minimizing the sum of squared errors, so that the model can accurately predict energy consumption based on the input energy consumption-related data.

[0024] Furthermore, the S200 establishes a sub-model based on decision tree evaluation for the production efficiency target. The model is constructed based on the decision tree algorithm. The decision tree is a method for constructing a tree structure model by recursively dividing data features. It can effectively handle complex classification and regression problems. In the present invention, it is used to evaluate and predict production efficiency based on multiple production efficiency-related parameters in the metal smelting process, and provide an intuitive, easy-to-understand and decision-making guiding model output for the optimization and management of the production process. The sub-model uses C1, C2, C3, C4 and C5 as splitting features of the decision tree, calculates the information gain of each feature, and uses information entropy as an indicator to measure the uncertainty of the data set. The larger the information gain, the more the uncertainty of the data set can be reduced by using the feature for division, and the more conducive it is to the classification or prediction task, and the data set is set to S, feature C i The value set is {c i1 ,c i2 ,…,c ir}, i = 1, 2, 3, 4, 5, according to feature C i Divide the data set S into r subsets S ij , j = 1, 2, ..., r, then the feature C i The information gain IG(C i )for: Among them, Entropy(S) is the information entropy of the data set S, that is, Among them, u is the number of production efficiency categories, p t is the probability of category t in the data set S. By calculating the information gain of each feature, the feature with the largest information gain is selected as the root node for splitting, and the data set is divided into different subsets. For each subset, the above process is repeated recursively, and the feature with the largest information gain is continued to be split until the stopping condition is met. The stopping condition is: the information gain of all features is less than the threshold information gain IG yz , thus constructing a complete decision tree sub-model;

[0025] When the parameter values ​​of production efficiency related data are input, starting from the root node of the decision tree, according to the characteristic judgment conditions of each node, judgment is made step by step along the branches of the decision tree. Each internal node of the decision tree represents a test condition of a feature, the branch represents the different values ​​of the feature, and the leaf node represents the final predicted value, which is the predicted result y of production efficiency. pe .

[0026] Furthermore, in the prediction sub-model of metal smelting product quality based on neural network, in the model training stage, the output values ​​of some hidden layer neurons are randomly set to 0, and the probability is set to r glIn the prediction stage, T forward propagations are performed to obtain T different product quality prediction results. Calculate the predicted result y pq The mean and variance The variance As an uncertainty measure of product quality prediction results, a large variance indicates high uncertainty.

[0027] Furthermore, the S300 estimates the uncertainty of the prediction sub-model for metal smelting energy consumption based on linear regression by using the variance of the residual, that is, by calculating the residual e of each sample k =y erk -(w4B 1k +w5B 2k +w6B 3k +w7), where k = 1, 2, ..., q, q is the number of samples, and the variance of the residual is calculated As an uncertainty measure of energy consumption prediction results, the variance is large and the uncertainty is high.

[0028] Furthermore, the S200 establishes a sub-model based on decision tree evaluation for the production efficiency target. When constructing the decision tree, for each sample in the data set S, one-third of the data that does not participate in the construction of the leaf node where the sample is located is used as external data. The sub-model is verified using the external data. For each sample, its prediction accuracy a on the external data is calculated. pe , the accuracy is low, indicating that the prediction result of the sample is highly uncertain. The average prediction accuracy of all samples is As the overall uncertainty measure of the production efficiency prediction result, s is the number of samples in the data set S, 1-A pe As a quantitative indicator of the degree of uncertainty, the larger the value, the higher the uncertainty.

[0029] Furthermore, the S400 determines the initial weight of product quality based on the importance of each target in the metal smelting production process: The initial weight of energy consumption is The initial weight of production efficiency is And meet Taking into account the impact of the uncertainty of the output results of each sub-model on the decision, the initial weight is dynamically adjusted, that is, the output result of product quality is y pq , the uncertainty measure is the variance The output result of energy consumption is y er , the uncertainty measure is the variance The output result of production efficiency is y pe, the uncertainty measure is 1-A pe , using the weight adjustment formula: in, Represents the weights of product quality, energy consumption and production efficiency set at the initial stage, is the uncertainty measure of the output results of each sub-model, i∈{pq,er,pe};

[0030] By adjusting the weights, the output results of the sub-models with low uncertainty have a relatively large influence in the final fusion decision, making the decision stable. The output results of each sub-model are multiplied by their corresponding adjusted weights and then added to obtain the weighted average fusion result y fusion , that is, y fusion =w pq y pq +w er y er +w pe y pe ,The fusion result is used as the final decision output to guide the adjustment of equipment operating parameters in the metal smelting process, to achieve precise control of the metal smelting process and effective balance and optimization among multiple objectives.

[0031] Compared with the prior art, the processing method of intelligent model data applied to metal smelting has the following beneficial effects:

[0032] 1. The present invention designs a multi-objective model output fusion mechanism that can simultaneously consider product quality, energy consumption and production efficiency in the metal smelting process. Through the weighted average method, the model prediction results corresponding to different objectives are intelligently integrated to obtain the comprehensive optimal decision output. This innovative fusion mechanism overcomes the limitations of traditional single-objective optimization models, enables the metal smelting process to achieve balance and optimization among multiple key indicators, meets the company's needs for maximizing comprehensive benefits, and improves the sustainability and competitiveness of the production process.

[0033] Second, the present invention has strong adaptability and decision reliability. Considering the complexity and uncertainty of the metal smelting process, the present invention evaluates the uncertainty of the output results of each sub-model and incorporates it into the fusion decision-making process, so that the final decision can be more robust to cope with various emergencies and changes in working conditions.

[0034] Other advantages, objectives and features of the present invention will be set forth in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0036] Figure 1 An operation diagram of a method for processing intelligent model data applied to metal smelting;

[0037] Figure 2 This is a flowchart of the steps of user classification in Example 2. DETAILED DESCRIPTION

[0038] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation mode, structure, characteristics and effects of the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments.

[0039] Embodiment 1

[0040] like Figure 1 As shown, this embodiment aims to demonstrate the specific application process of a method for processing intelligent model data applied to metal smelting in a steel smelting plant. By collecting, processing and analyzing data related to product quality, energy consumption and production efficiency in the smelting process of the plant, and utilizing the established multi-objective intelligent model and fusion mechanism, precise control and optimization of the smelting process can be achieved, thereby effectively improving the overall benefits.

[0041] On the production line, data closely related to product quality, energy consumption, and production efficiency are comprehensively collected. Data related to product quality include the hardness value A1 of the finished metal product, the strength value A2 of the finished metal product, the purity value A3 of the finished metal product, the number of internal defects of the finished metal product A4, the type of internal defects of the finished metal product A5, and the surface roughness value A6 of the finished metal product; data related to energy consumption include the power consumption value B1 of each stage of the smelting process, the fuel consumption flow value B2 of each stage of the smelting process, and the thermal energy utilization value B3 of the smelting process; data related to production efficiency include the metal output value C1 per unit time, the proportion of equipment operation time C2, the residence time value C3 of each link of the production process, the number of equipment failure shutdowns C4, and the equipment failure shutdown duration C5. The collected raw data often contains noise and outliers. These interference factors will mislead the training and decision-making of subsequent models. Therefore, the data is cleaned and denoised to ensure the accuracy and reliability of the data and provide a high-quality data source for subsequent model training.

[0042] Different sub-models are established for the product quality, energy consumption and production efficiency goals in the metal smelting process, and each sub-model is independently trained using historical data and actual production data, so that each sub-model has the ability to predict and analyze the corresponding goals. A sub-model based on neural network prediction is established for the product quality goal. The input layer of the model is composed of neurons with parameter data related to product quality. The collected A1, A2, A3, A4, A5 and A6 are used as input values. The hidden layer has n neurons, which are used to extract complex nonlinear features from the data in the input layer. The input layer and the hidden layer are connected through the weight matrix W1. Each element of the weight matrix represents the connection weight between the i-th neuron in the input layer and the j-th neuron in the hidden layer. It determines the scaling and weighting degree of the input data when it is passed to the hidden layer. At the same time, the hidden layer is also equipped with a bias vector b1, which provides an initial offset for the hidden layer neurons to increase the expressive power of the model. The activation function of the hidden layer neurons is f1(x)=max(0,x), which accelerates the convergence of model training and introduces nonlinear factors so that the model can learn the complex relationship between the input data. Its output h j The calculation formula is: where j = 1, 2, ..., n, x i is the value of the i-th neuron in the input layer, that is, the parameter data value of the corresponding product quality. The output layer is responsible for outputting the final product quality prediction value and is connected to the hidden layer through the weight matrix W2. Each element represents the connection weight between the jth neuron in the hidden layer and the neurons in the output layer. The output layer also has a bias vector b2. The activation function of the neurons in the output layer is f2(x)=x. The output of the neurons in the output layer is used as the product quality prediction value y. pq The calculation formula is: pq =f2(∑ i＝1 (W 2j h j +b2)), through a large amount of historical data and actual production data, the weight matrix W1, W2 and the bias vector b1, b2 are continuously adjusted so that the model can accurately predict the quality of steel products based on the input product quality related parameters. A sub-model based on the linear regression algorithm is established for the energy consumption target. The model is based on B 1、 B 2、 B3 is used as the independent variable, and the energy consumption forecast value y er As the dependent variable, the constructed linear regression equation is: er =w4B1+w5B2+w6B3+w7, where w 4、 w 5、w6 is the regression coefficient, which reflects the influence of each independent variable on energy consumption. For example, w4 represents the change in the predicted energy consumption value for each unit change in the power consumption value; w7 is the intercept, which represents the basic value of energy consumption when the independent variable is zero, including the basic energy consumption of the equipment, the energy consumption caused by factors such as environmental heat dissipation, etc. The linear regression equation is fitted by the least squares method, and its purpose is to minimize the difference between the actual energy consumption value and the predicted value y er The sum of squared errors between Among them, q is the number of samples, y is erk , B 1k , B 2k , B 3k are the actual energy consumption value of the kth sample and the corresponding independent variable value. In the actual training process, the regression coefficient w is iteratively updated by minimizing the sum of squared errors by using historical data and energy consumption-related data samples collected in the actual production process. 4、 w 5、 The estimated values ​​of w6 and intercept w7 enable the model to accurately predict energy consumption based on the input energy consumption related data; a sub-model based on decision tree evaluation is established for the production efficiency target. This sub-model uses C1, C2, C3, C4 and C5 as the splitting features of the decision tree. First, the information gain of each feature is calculated, and the information entropy is used as an indicator to measure the uncertainty of the data set. Suppose the data set is S, and the feature C i The value set of (i=1,2,3,4,5) is {c i1 ,c i2 ,…,c ir}, according to feature C i Divide the dataset S into r subsets S ij (j=1,2,…,r), then feature C i The information gain IG(C i )for: Among them, Entropy(S) is the information entropy of the data set S, that is, u is the number of production efficiency categories, p t is the probability of category t in the data set S. By calculating the information gain of each feature, the feature with the largest information gain is selected as the root node for splitting, and the data set is divided into different subsets. For example, if it is found that the information gain of the metal output value C1 per unit time is the largest, then C1 is used as the root node for splitting, and the data set is divided into different branches according to its different values. For each subset, the above process is repeated recursively, and the feature with the largest information gain is continued to be selected for splitting until the stopping condition is met, that is, the information gain of all features is less than the threshold information gain IG yz, thus constructing a complete decision tree sub-model. When the parameter values ​​of production efficiency-related data are input, starting from the root node of the decision tree, according to the characteristic judgment conditions of each node, judgment is made step by step along the branches of the decision tree. Each internal node of the decision tree represents a test condition of a feature, the branch represents the different values ​​of the feature, and the leaf node represents the final predicted value, which is the predicted result y of production efficiency. pe .

[0043] Each sub-model performs prediction calculations based on the real-time collected data, outputs its own prediction results on product quality, energy consumption and production efficiency, and performs uncertainty assessment on the output results of each sub-model. For the prediction sub-model of metal smelting product quality based on neural network, during the model training stage, some output values ​​of hidden layer neurons are randomly set to 0, and the probability is set to r gl This is to simulate the uncertainty of the model under different training conditions and increase the generalization ability of the model. In the prediction stage, T forward propagations are performed to obtain T different product quality prediction results. Calculate the predicted result y pq The mean and variance variance As a measure of uncertainty in product quality prediction results, the larger the variance, the greater the degree of dispersion of the prediction results in different forward propagation processes, that is, the higher the prediction uncertainty of the model for the input data. For example, if the variance is small, it means that the model is relatively stable in multiple predictions and the prediction reliability of the product quality is high; on the contrary, if the variance is large, further data analysis or model improvement is required; for the prediction sub-model of metal smelting energy consumption based on linear regression, the variance of the residual is used to estimate the uncertainty, and the residual e of each sample is calculated. k =y erk -(w4B 1k +w5B 2k +w6B 3k +w7), where k = 1, 2, ..., q, q is the number of samples, the residual reflects the deviation between the model prediction value and the actual value, and the variance of the residual is calculated variance As a measure of uncertainty in the energy consumption prediction results, the larger the variance, the greater the deviation fluctuation between the model's prediction of energy consumption and the actual value, and the higher the uncertainty; for the evaluation sub-model of metal smelting production efficiency based on the decision tree, when constructing the decision tree, for each sample, one-third of the data that does not participate in the construction of the leaf node where the sample is located is used as external data, and the sub-model is verified using the external data. For each sample, its prediction accuracy a on the external data is calculated. pe, the accuracy is low, indicating that the prediction result of the sample is highly uncertain. The average prediction accuracy of all samples is As the overall uncertainty measure of the production efficiency prediction results, 1-A pe As a quantitative indicator of the degree of uncertainty, the larger the value, the higher the uncertainty.

[0044] Based on the importance of each goal in the metal smelting production process, combined with the actual production needs and strategic priorities of the steel plant, the initial weight of product quality is determined as The initial weight of energy consumption is The initial weight of production efficiency is And meet Taking into account the impact of the uncertainty of the output results of each sub-model on the decision, the initial weights are dynamically adjusted. The output result of product quality is y pq , the uncertainty measure is the variance The output result of energy consumption is y er , the uncertainty measure is the variance The output result of production efficiency is y pe , the uncertainty measure is 1-A pe , using the weight adjustment formula In these formulas, the denominator It plays a normalization role to ensure that the sum of the adjusted weights is still 1. For the numerator, taking product quality as an example, Indicates that the weights are adjusted based on the uncertainty measure based on the initial weights. If the variance of the prediction results of the product quality sub-model is is small, indicating that its uncertainty is low, then Larger, the weight of product quality w will be increased when the weight is adjusted pq Relatively increase, that is, let the output results of the sub-models with low uncertainty have a relatively large influence in the final fusion decision, so that the decision tends to be stable, and the output results of each sub-model are multiplied by their corresponding adjusted weights and then added to obtain the weighted average fusion result y fusion , that is, y fusion =w pq y pq +w er y er +w pe y pe ,The fusion result is used as the final decision output to guide the adjustment of equipment operating parameters in the metal smelting process, such as adjusting the furnace temperature, feeding speed, and rolling pressure, so as to achieve optimized control of the steel smelting process.

[0045] Finally, the fused decision output y fusionApplied to the actual production process of steel smelting, the equipment operating parameters and process control parameters are adjusted and optimized accordingly according to the decision results. At the same time, the actual data in the production process is continuously monitored, and the actual production results are compared and analyzed with the predicted results of the fusion model. Through continuous feedback and adjustment, the prediction accuracy of the model and the effectiveness of the decision are improved, and the continuous optimization and closed-loop control of the steel smelting process are achieved, ensuring that the production process is always in an efficient, stable and high-quality state.

[0046] In summary, this embodiment demonstrates in detail the entire application process of a method for processing intelligent model data applied to metal smelting in a steel smelting plant. By comprehensively acquiring key data and preprocessing it in the data collection stage, the foundation for model training is laid; in the model construction stage, suitable sub-models are established for different objectives and trained and optimized; in the model output and fusion stage, the prediction results and uncertainty assessment are accurately calculated, and multi-objective fusion decision-making is achieved through scientific weight adjustment; finally, in the decision execution and feedback adjustment stage, the decision is applied to production and the model is continuously improved. This method can effectively integrate multiple factors such as product quality, energy consumption and production efficiency, overcome the limitations of traditional methods, make robust and flexible decisions under complex and changeable steel smelting conditions, and significantly improve the comprehensive benefits and sustainability of the production process.

[0047] Embodiment 2

[0048] like Figure 2 As shown, this embodiment mainly demonstrates the application process of the intelligent model data processing method applied to metal smelting in production, applies decision-making to production to achieve continuous optimization, effectively improves the comprehensive benefits and competitiveness of smelting enterprises, and fully verifies the feasibility and practicality of this method in complex industrial scenarios.

[0049] First, enter the data collection stage (S100), use sensors distributed in various links of metal smelting to collect various data related to multiple goals such as product quality, energy consumption, and production efficiency, and perform pre-processing operations such as cleaning and denoising on the collected data to remove abnormal and invalid data to ensure the accuracy and reliability of the data, thereby providing a high-quality data source for subsequent model training and analysis.

[0050] Then, the multi-objective intelligent model establishment stage (S200) is entered. Corresponding sub-models are established for the goals of product quality, energy consumption and production efficiency in the metal smelting process. For example, a product quality prediction model is established using a neural network, energy consumption is analyzed using a linear regression model, and a production efficiency evaluation model is constructed using a decision tree algorithm. Each sub-model is independently trained according to its corresponding goal, and the complex relationship and rules between each goal and related input parameters are learned through a large amount of historical data and actual production data, so as to have the ability to predict and analyze their respective goals.

[0051] Subsequently, the model output and fusion preparation stage (S300) is entered, and each sub-model performs prediction calculations based on the real-time collected data to obtain respective prediction results on targets such as product quality, energy consumption and production efficiency. These prediction results are output in a standardized data format for subsequent fusion processing. At the same time, the output results of each sub-model are evaluated for uncertainty, for example, by calculating the variance and probability distribution of the prediction results, the reliability and uncertainty of each sub-model output are quantified, providing an important reference for subsequent fusion decisions.

[0052] Next, the multi-objective fusion decision-making stage (S400) is entered. According to the importance of each objective in actual production, corresponding weights are assigned to the output results of sub-models such as product quality, energy consumption and production efficiency. The output results of each sub-model are multiplied by their corresponding weights and then added to obtain the weighted average fusion result as the final decision output. This method can adjust the weight of each objective according to the actual needs of the enterprise and achieve emphasis and balance on different objectives.

[0053] Finally, the decision execution and feedback adjustment stage (S500) is entered, and the fused decision output is applied to the actual production process of metal smelting, and the equipment operating parameters, process control parameters, etc. are adjusted and optimized accordingly. At the same time, the actual data in the production process is continuously monitored, and the actual production results are compared and analyzed with the predicted results of the fusion model to evaluate the effect of decision execution. If it is found that there is a large deviation between the actual production results and the expectations, each sub-model is retrained and optimized according to the deviation information, and the model parameters, fusion weights or fuzzy rules are adjusted to improve the prediction accuracy of the model and the effectiveness of the decision, so as to achieve continuous optimization and closed-loop control of the metal smelting process.

[0054] To sum up, this embodiment demonstrates in detail the complete implementation process of the metal smelting intelligent model data processing method, realizes the effective optimization and comprehensive balance of product quality, energy consumption and production efficiency in the metal smelting process, and enhances the economic benefits and market competitiveness of the enterprise.

[0055] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method for processing intelligent model data applied to metal smelting, characterized in that: The specific steps of this method are: S100, collect data related to product quality, energy consumption, and production efficiency in all links of metal smelting in real time, and clean and remove noise from the collected data; S200, establishing different sub-models for product quality, energy consumption and production efficiency targets in the metal smelting process, wherein the sub-models include a sub-model based on neural network prediction for product quality targets, a sub-model based on linear regression algorithm for energy consumption targets, and a sub-model based on decision tree evaluation for production efficiency targets, and independently training each sub-model using historical data and actual production data, so that each sub-model has the ability to predict and analyze the corresponding targets; S300, each sub-model performs prediction calculation based on the real-time collected data, outputs its own prediction results on product quality, energy consumption and production efficiency, and performs uncertainty assessment on the output results of each sub-model, where: The uncertainty evaluation index of the product quality is set as the variance of the prediction result The uncertainty evaluation index of energy consumption is set as the variance of the residual of the prediction result The uncertainty evaluation index of the production efficiency is set as the mean value A of the prediction result accuracy pe ; S400, intelligently integrating the output results of each sub-model by weighted average to obtain the comprehensive optimal decision output; The weighted average fusion method includes assigning weights to the output results of each sub-model according to the importance of each target in actual production, and multiplying the output results of each sub-model by its corresponding weight and adding them together to obtain the weighted average fusion result. The fusion result is used as the final decision output to guide the adjustment of equipment operating parameters in the metal smelting process. S500. Apply the fused decision output to the actual production process of metal smelting, continuously monitor the actual data in the production process, compare and analyze the actual production results with the predicted results of the fusion model, and make corresponding adjustments and optimizations to the equipment operating parameters and process control parameters to achieve continuous optimization and closed-loop control of the metal smelting process.

2. The method for processing intelligent model data applied to metal smelting according to claim 1, characterized in that: The data related to the S100 product quality include the hardness value A1 of the finished metal product, the strength value A2 of the finished metal product, the purity value A3 of the finished metal product, the number of internal defects of the finished metal product A4, the type of internal defects of the finished metal product A5, and the surface roughness value A6 of the finished metal product; The energy consumption related data include the power consumption value B1 of each stage of the smelting process, the fuel consumption flow value B2 of each stage of the smelting process, and the thermal energy utilization rate value B3 of the smelting process; The production efficiency-related data include the metal output value per unit time C1, the equipment operation time ratio C2, the residence time value of each link in the production process C3, the number of equipment failure shutdowns C4, and the equipment failure shutdown duration C5.

3. The method for processing intelligent model data applied to metal smelting according to claim 2, characterized in that: The S200 establishes a sub-model based on neural network prediction for the product quality target, the model includes an input layer, a hidden layer and an output layer, wherein: The input layer is composed of neurons related to product quality parameter data, and all collected product quality parameter data are used as input values, that is, the input layer neurons include A1, A2, A3, A4, A5 and A6; The hidden layer is provided with n neurons, which extract features from the data of the input layer. The input layer and the hidden layer are connected through a weight matrix W1, whose dimension is n×p, where p is the number of neurons in the input layer and each element represents the connection weight between the i-th neuron in the input layer and the j-th neuron in the hidden layer. At the same time, the hidden layer also has a bias vector b1, whose dimension is n×1, and each element represents the bias value of the jth neuron in the hidden layer. The activation function of the hidden layer neuron is f1(·), that is, f1(x)=max(0,x), and its output h j The calculation formula is: where j = 1, 2, ..., n, x i is the value of the i-th neuron in the input layer, i.e., the parameter data values ​​A1, A2, A3, A4, A5 and A6 of the corresponding product quality; The output layer is responsible for outputting the final product quality prediction value, which is connected to the hidden layer through the weight matrix W2, with a dimension of 1×n, and each element represents the connection weight between the jth neuron in the hidden layer and the neurons in the output layer. The output layer also has a bias vector b2. The activation function of the neurons in the output layer is f2(·), that is, f2(x) = x. The output of the neurons in the output layer is used as the product quality prediction value y pq The calculation formula is: The weight matrices W1, W2 and bias vectors b1, b2 are adjusted by inputting product quality parameter data, and the model parameters are continuously optimized iteratively.

4. The method for processing intelligent model data applied to metal smelting according to claim 2, characterized in that: The S200 establishes a sub-model based on a linear regression algorithm for the energy consumption target. The model uses B1, B2, and B3 as independent variables and converts the energy consumption prediction value y er As the dependent variable, the constructed linear regression equation is: er =w4B1+w5B2+w6B3+w7, where w 4、 w 5、 w6 is the regression coefficient, which reflects the influence of each independent variable on energy consumption; w7 is the intercept, which represents the basic value of energy consumption when the independent variable is zero. The linear regression equation is fitted by the least squares method, the purpose of which is to minimize the difference between the actual energy consumption value and the predicted value y er The sum of squared errors is used to find the best regression coefficient and intercept value so that the predicted value of the model is close to the actual value. The calculation formula of the sum of squared errors is: Among them, q is the number of samples, y is erk , B 1k , B 2k , B 3k are the actual energy consumption value of the kth sample and the corresponding independent variable value, respectively. Using historical data and energy consumption-related data samples collected in the actual production process, the estimated values ​​of regression coefficients w4, w5, w6 and intercept w7 are iteratively updated by solving the problem of minimizing the sum of squared errors, so that the model can accurately predict energy consumption based on the input energy consumption-related data.

5. The method for processing intelligent model data applied to metal smelting according to claim 2, characterized in that: The S200 establishes a sub-model based on decision tree evaluation for the production efficiency target. The sub-model uses C1, C2, C3, C4 and C5 as split features of the decision tree, calculates the information gain of each feature, uses information entropy as an indicator to measure the uncertainty of the data set, and sets the data set to S, feature C i The value set is {c i1 ,c i2 ,…,c ir }, i = 1, 2, 3, 4, 5, according to feature C i Divide the data set S into r subsets S ij , j = 1, 2, ..., r, then the feature C i The information gain IG(C i )for: Among them, Entropy(S) is the information entropy of the data set S, that is, Among them, u is the number of production efficiency categories, p t is the probability of category t in the data set S. By calculating the information gain of each feature, the feature with the largest information gain is selected as the root node for splitting, and the data set is divided into different subsets. For each subset, the above process is repeated recursively, and the feature with the largest information gain is continued to be split until the stopping condition is met. The stopping condition is: the information gain of all features is less than the threshold information gain IG yz , thus constructing a complete decision tree sub-model; When the parameter values ​​of production efficiency related data are input, starting from the root node of the decision tree, according to the characteristic judgment conditions of each node, judgment is made step by step along the branches of the decision tree. Each internal node of the decision tree represents a test condition of a feature, the branch represents the different values ​​of the feature, and the leaf node represents the final predicted value, which is the predicted result y of production efficiency. pe .

6. The method for processing intelligent model data applied to metal smelting according to claim 3 is characterized in that: The S300 predicts the quality of metal smelting products based on a neural network sub-model. During the model training phase, some output values ​​of hidden layer neurons are randomly set to 0, with a probability of r gl In the prediction stage, T forward propagations are performed to obtain T different product quality prediction results. Calculate the predicted result y pq The mean and variance The variance As an uncertainty measure of product quality prediction results, a large variance indicates high uncertainty.

7. The method for processing intelligent model data applied to metal smelting according to claim 4, characterized in that: The S300 predicts the energy consumption of metal smelting based on linear regression, and estimates the uncertainty by using the variance of the residual, that is, by calculating the residual e of each sample. k =y erk -(w4B 1k +w5B 2k +w6B 3k +w7), where k = 1, 2, ..., q, q is the number of samples, and the variance of the residual is calculated As an uncertainty measure of energy consumption prediction results, the variance is large and the uncertainty is high.

8. The method for processing intelligent model data applied to metal smelting according to claim 5, characterized in that: S200 establishes a sub-model based on decision tree evaluation for the production efficiency target. When constructing the decision tree, for each sample in the data set S, one-third of the data that does not participate in the construction of the leaf node where the sample is located is used as external data. The sub-model is verified using the external data. For each sample, its prediction accuracy a on the external data is calculated. pe , the accuracy is low, indicating that the prediction result of the sample is highly uncertain. The average prediction accuracy of all samples is As the overall uncertainty measure of the production efficiency prediction result, s is the number of samples in the data set S, 1-A pe As a quantitative indicator of the degree of uncertainty, the larger the value, the higher the uncertainty.

9. The method for processing intelligent model data applied to metal smelting according to claim 1, characterized in that: The S400 determines the initial weight of product quality based on the importance of each target in the metal smelting production process: The initial weight of energy consumption is The initial weight of production efficiency is And meet Taking into account the impact of the uncertainty of the output results of each sub-model on the decision, the initial weight is dynamically adjusted, that is, the output result of product quality is y pq , the uncertainty measure is the variance The output result of energy consumption is y er , the uncertainty measure is the variance The output result of production efficiency is y pe , the uncertainty measure is 1-A pe , using the weight adjustment formula: in, Represents the weights of product quality, energy consumption and production efficiency set at the initial stage, is the uncertainty measure of the output results of each sub-model, i∈{pq,er,pe}; By adjusting the weights, the output results of the sub-models with low uncertainty have a relatively large influence in the final fusion decision, making the decision stable. The output results of each sub-model are multiplied by their corresponding adjusted weights and then added to obtain the weighted average fusion result y fusion , that is, y fusion =w pq y pq +w er y er +w pe y pe ,The fusion result is used as the final decision output to guide the adjustment of equipment operating parameters in the metal smelting process.

Images