A decision analysis method and system for AI big data modeling applied to aluminum smelting
Through AI big data modeling and reinforcement learning algorithms, the smelting parameters in the aluminum smelting process are optimized in real time, solving the problem of inflexible parameter adjustment in traditional aluminum smelting, improving smelting efficiency and quality, and achieving accurate reducing agent control.
Patent Information
- Application Number
- CN202411026892.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-30
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-07-30
AI Technical Summary
In the traditional aluminum smelting process, it is difficult to adjust the smelting parameters in real time, resulting in poor process adaptability, high energy consumption, large fluctuations in product quality, and lack of precise control in the existing technology, especially in terms of reducing agent dosage and temperature control.
Using decision-making analysis method of AI big data modeling, smelting parameters are collected in real time through sensors, feature filtering algorithms, multivariable regression algorithms and genetic algorithms are used to build a production parameter optimization model, combined with reinforcement learning algorithms to achieve reducing agent delivery control, and real-time monitoring and optimization Smelting process.
It improves the efficiency and quality of the aluminum smelting process, enhances the adaptability and flexibility of the process, and realizes precise control of the amount of reducing agent output and time point, ensuring the stability and efficiency of the smelting process.
Smart Images

Figure CN118552016B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of industrial generation and data analysis, and in particular relates to a decision analysis method and system for AI big data modeling applied to aluminum smelting. Background Art
[0002] With the urgent global demand for energy conservation, emission reduction, and improved production efficiency, the aluminum industry, as an energy-intensive industry, faces unprecedented challenges and opportunities. Traditional aluminum smelting processes rely on experienced engineers manually adjusting smelting parameters. This is not only inefficient but also difficult to precisely control, leading to high energy consumption, wide fluctuations in product quality, and severe environmental pollution. Existing technologies are particularly deficient in areas such as raw material ratios, smelting temperature control, and reducing agent dosage, all of which are directly related to the output, quality, and production cost of aluminum products.
[0003] The above existing technologies have the following problems: 1) Smelting parameters (such as temperature, time, raw material ratio, etc.) are usually pre-set and difficult to dynamically adjust according to real-time working conditions, resulting in poor process adaptability; 2) The reduction process of alumina by temperature and pressure parameters is a key step in aluminum smelting, but the amount of reducing agent added in the existing process is often estimated based on experience and lacks precise calculation, resulting in low reduction efficiency. In order to solve the above problems, the present invention provides a decision analysis method and system based on AI big data modeling. Summary of the Invention
[0004] In response to the shortcomings of the existing technology, the present invention proposes a decision analysis method for AI big data modeling applied to aluminum smelting. The method collects process parameters in the smelting process in real time through sensors, and obtains effective feature parameters through feature filtering algorithm processing. Secondly, a production parameter optimization model is constructed using a multivariate regression algorithm and a genetic algorithm to obtain initial production process parameters. At the same time, a feed control model is constructed based on a reinforcement learning algorithm to achieve intelligent control of the reducing agent feed amount and time point. The reduction efficiency of the smelting process is monitored in real time, and comparison and adjustment are made according to thresholds to ensure smelting efficiency and quality. The present invention realizes intelligent decision analysis of the aluminum smelting process through big data modeling and AI technology, thereby improving smelting efficiency and product quality.
[0005] To achieve the above object, the present invention provides the following technical solutions:
[0006] A decision analysis method for AI big data modeling applied to aluminum smelting, including:
[0007] Step S1: collecting production process parameters and production efficiency parameters of the smelting process in real time through sensors, and filtering the collected parameters using a feature filtering algorithm built into the sensor to obtain irrelevant valid feature parameters;
[0008] Step S2: constructing a multivariable production process parameter optimization model based on a multivariable regression algorithm and a genetic algorithm, inputting the obtained irrelevant effective characteristic parameters into the multivariable production process parameter optimization model to obtain the initial production process parameters of the smelting process;
[0009] Step S3: constructing a production process delivery control model based on a reinforcement learning algorithm, and deploying the constructed delivery control model to the smelting process control platform. At the same time, the obtained initial production process parameters are input into the smelting process control platform to control the delivery amount and timing of the reducing agent through the constructed delivery control model;
[0010] Step S4: Setting a reduction efficiency threshold, monitoring the reduction efficiency during the smelting process in real time through an integrated sensor, and comparing the monitored reduction efficiency with the threshold. If the reduction efficiency is greater than the reduction efficiency threshold, maintaining the current smelting process state;
[0011] Step S5: If the reduction efficiency is less than the reduction efficiency threshold, the monitored reduction efficiency threshold is fed back to the multivariable production process parameter optimization model and the dosage control model to optimize and adjust the initial production process parameters and the dosage and time point of the reducing agent so that the reduction efficiency is greater than the reduction efficiency threshold.
[0012] Specifically, the steps of obtaining irrelevant valid feature parameters in step S1 include:
[0013] S101. Setting a collinearity correlation threshold, preprocessing the collected smelting process production process parameters, performing collinearity diagnosis on the preprocessed process parameters using a correlation algorithm, obtaining collinearity correlation coefficients between various variables, and comparing the obtained collinearity correlation coefficients;
[0014] S102, setting the reduction efficiency as the target variable and the remaining process parameters as the decision variables, inputting the target variable and the remaining process parameters as the decision variables into the factor analysis, and obtaining the factor loading matrix of the target variable and the decision variables;
[0015] S103, setting a factor load threshold, comparing the load correlation coefficient between the decision variable and the target variable in the obtained factor load matrix with the factor load threshold, and if it is greater than the factor load threshold, it is a valid characteristic parameter;
[0016] S104. Compare the collinear correlation coefficient between the obtained valid feature parameters with the collinear correlation threshold. If it is greater than the collinear correlation threshold, retain the decision variable with the larger load correlation coefficient among the two decision variables with collinearity, and delete the other decision variable to obtain irrelevant valid feature parameters.
[0017] Specifically, the process of constructing the multivariable production process parameter optimization model in step S2 specifically includes:
[0018] S201. Using the production efficiency parameter among the irrelevant effective characteristic parameters as the target variable and the production process parameters among the irrelevant effective characteristic parameters as the decision variables, a multivariate production process parameter optimization model is constructed through a multivariate regression function, wherein the multivariate regression function formula is: ,in, represents the target variable in the production efficiency parameter, represents the decision variable in the nth production process parameter, is the weight coefficient of the corresponding decision variable.
[0019] Specifically, the process of constructing the multivariable production process parameter optimization model in step S2 further includes:
[0020] S202, based on the constructed multiple regression function, using the weighted coefficient vector in the multiple regression function as an input chromosome of the genetic algorithm, and using the input chromosome to randomly generate a coefficient vector as an initialization population;
[0021] S203, constructing a genetic algorithm multi-objective fitness function F and a constraint condition set S using the production efficiency parameters and production process parameters in the unrelated effective characteristic parameters;
[0022] S204. Set the maximum number of iterations and rate of change stop threshold , iterate the genetic algorithm and calculate the Pareto front change rate , the specific formula is Among them, Indicates the first The Pareto frontier solution set of the generation, Indicates that x is the first A solution in the Pareto frontier solution set of the first generation, y represents the obtained A solution in the Pareto front solution set of the generation, represents the 2-norm symbol;
[0023] S205, iteratively calculate the process of S201-S204 based on the maximum number of iterations and the change rate stop threshold set in S204. When , the iteration stops, and the weighted coefficient vector solution of a set of optimal decision variables of the multivariate regression function in S201 obtained by the search is output, and the obtained weighted coefficient vector solution is fed back to the multivariate regression function to initialize the production process parameters to obtain the initial production process parameters.
[0024] Specifically, the multi-objective fitness function F and the constraint condition set S in S203 are:
[0025] ,
[0026]
[0027] in, Indicates the actual reduction efficiency of the product during the smelting process, Indicates the product reduction efficiency threshold during the smelting process, Indicates the total energy consumption to complete a smelting cycle. Indicates the quality of the output product after completing a smelting cycle. Indicates the amount of pollution emitted during a smelting cycle. Indicates the process score error of the output product after completing a smelting cycle; It represents the penalty function for the corresponding variable in the fitness function that violates the sth constraint in the constraint set, It represents the expected energy consumption per unit time. Indicates the expected energy consumption fluctuation tolerance, It represents the average concentration of pollutants emitted per unit time during the smelting process. It represents the average concentration threshold of pollutants emitted per unit time during the smelting process. Represents the process scoring error threshold, which is a fixed constant. It represents the total mass of the corresponding smelting material minerals put into the smelting process per unit time. Indicates the mineral smelting efficiency threshold, Indicates the pressure inside the smelting equipment at time t, It indicates the maximum pressure that the smelting equipment can withstand during operation. is the weighted coefficient of the corresponding fitness term.
[0028] Specifically, the specific steps of constructing the delivery control model in step S3 include:
[0029] S301. Set the following control constraints based on the production efficiency parameters and initial production process parameters among the obtained irrelevant effective characteristic parameters, specifically:
[0030] ,
[0031] in, Indicates the smelting temperature inside the equipment at time t during the smelting process, In turn, they represent the minimum and maximum smelting temperatures in the equipment during the smelting process. Indicates the pressure inside the equipment at time t during the smelting process, In turn, they represent the minimum and maximum values of the smelting pressure during the smelting process. Indicates the concentration of reducing agent in the equipment at time t during the smelting process, In turn, they represent the minimum and maximum values of reducing agent concentration during the smelting process. Indicates the actual reduction efficiency of the product at time t during the smelting process;
[0032] S302: Setting the following trigger conditions according to the control constraint information in S301, specifically including:
[0033] When the temperature condition in the constraint condition is not met, triggering the first execution information;
[0034] When the pressure condition in the constraint condition is not satisfied, the second execution information is triggered;
[0035] When the reducing agent concentration condition in the constraint condition is not satisfied, triggering the third execution information;
[0036] When two or more of the temperature, pressure or reducing agent concentration conditions in the constraint conditions are not satisfied, triggering the fourth execution information;
[0037] When the restoration efficiency condition in the constraint condition is not satisfied, the fifth execution information is triggered, and the first to fourth execution information are jointly executed.
[0038] Specifically, the steps for building the S3 delivery control model include:
[0039] S303: Setting the current input state of the reinforcement learning sub-model in the delivery control model according to the production efficiency parameter and the initial production process parameter in the obtained irrelevant effective feature parameters , perform the action and the reward function , specifically:
[0040] , , ,
[0041] in, Indicates the action that triggers the execution of the first execution information. Indicates the action that triggers the execution of the second execution information. Indicates the action that triggers the execution of the third execution information. Indicates the action that triggers the execution of the fourth execution information. Indicates the action that triggers the execution of the fifth execution information. Indicates the threshold value of reducing agent concentration during smelting process, It indicates the quality of the expected smelting product per unit mass of mineral input. Represents the corresponding weight coefficient in the reward function.
[0042] A decision analysis system for AI big data modeling applied to aluminum smelting, including: a data preprocessing module, a parameter optimization module, and a delivery control module;
[0043] The data preprocessing module is used to collect the production process parameters of the smelting process in real time and preprocess the data to extract effective features; the data preprocessing module includes a data acquisition unit and a feature filtering unit;
[0044] The data acquisition unit is used to collect various production parameters and production efficiency parameters in the smelting process in real time through sensors; the feature filtering unit is used to process the collected data through the feature filtering algorithm built into the sensor to extract irrelevant valid feature parameters;
[0045] The delivery control module is used to build a delivery control model and control the delivery amount and time point of the reducing agent according to the initial production process parameters and the delivery control model.
[0046] Specifically, the parameter optimization module includes a model building unit and a parameter output unit;
[0047] The model construction unit is used to construct and train a multivariable production process parameter optimization model through a multivariable regression algorithm and a genetic algorithm; the parameter output unit is used to initialize and optimize the obtained irrelevant effective feature parameters through the multivariable production process parameter optimization model to obtain initial production process parameters.
[0048] A computer-readable storage medium stores computer instructions, which, when executed, execute a decision analysis method for AI big data modeling applied to aluminum smelting.
[0049] Compared with the prior art, the present invention has the following beneficial effects:
[0050] 1. This invention addresses the shortcomings of existing technologies by collecting production process parameters and production efficiency parameters during the smelting process in real time and using a built-in feature filtering algorithm to filter parameters, thereby achieving accurate extraction of key parameters. Secondly, this method combines a multivariate regression algorithm with a genetic algorithm to construct a multivariate production process parameter optimization model. This model can dynamically adjust smelting parameters such as temperature, pressure, and reduction efficiency based on real-time data, thereby improving the adaptability and flexibility of the aluminum smelting process. This dynamic adjustment strategy makes the smelting process more consistent with actual working conditions and effectively improves smelting efficiency.
[0051] 2. In response to the shortcomings of the existing technology, the present invention constructs a production process feeding control model through a reinforcement learning algorithm, thereby achieving precise control of the reducing agent feeding amount and timing. Secondly, by setting a reduction efficiency threshold and real-time monitoring of the reduction efficiency during the smelting process, closed-loop control of the aluminum smelting process is achieved, ensuring the stability and efficiency of the aluminum smelting process. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] Figure 1 This is a flow chart of a decision analysis method for AI big data modeling applied to aluminum smelting in Example 1 of the present invention;
[0053] Figure 2 This is a flow chart of the construction of a multivariable production process parameter optimization model according to Example 1 of the present invention;
[0054] Figure 3 This is a module architecture diagram of a decision analysis system for AI big data modeling applied to aluminum smelting according to Example 2 of the present invention. DETAILED DESCRIPTION
[0055] Example 1
[0056] See also Figure 1 The present invention provides an embodiment of a decision analysis method for AI big data modeling applied to aluminum smelting, which specifically includes the following steps:
[0057] Step S1, collect the production process parameters and production efficiency parameters of the smelting process in real time through the sensor, and filter the collected parameters using the feature filtering algorithm built into the sensor to obtain irrelevant valid feature parameters; further, in order to better illustrate the method steps of this embodiment, this embodiment takes the aluminum smelting production process as an example for illustration, and the production process parameters of the aluminum smelting process include: temperature, pressure, reducing agent concentration, reducing agent input amount, raw material ratio, power consumption and carbon dioxide emissions in the smelting equipment; production efficiency parameters include the quality of aluminum products output by the aluminum smelting process, aluminum smelting Reduction efficiency and aluminum product process score; further, the specific reasons for selecting the above production process parameters in this embodiment are: temperature directly affects the melting rate of aluminum, the chemical reaction rate, and the quality of the final product; pressure control is crucial to improving reaction efficiency and product purity when using high-pressure aluminum electrolysis; the type and input amount of reducing agent affect the reduction efficiency of aluminum ore; the ratio of raw materials such as alumina and coke materials affects the final aluminum output and quality; in electrolytic aluminum smelting, electricity consumption is the main cost and needs to be monitored to optimize energy efficiency; the efficiency of aluminum reduction from raw materials is used to measure production efficiency;
[0058] Step S2: constructing a multivariable production process parameter optimization model based on a multivariable regression algorithm and a genetic algorithm, inputting the obtained irrelevant effective characteristic parameters into the multivariable production process parameter optimization model to obtain the initial production process parameters of the smelting process;
[0059] Step S3: constructing a production process delivery control model based on a reinforcement learning algorithm, and deploying the constructed delivery control model to the smelting process control platform. At the same time, the obtained initial production process parameters are input into the smelting process control platform to control the delivery amount and timing of the reducing agent by constructing the delivery control model;
[0060] Step S4: Setting a reduction efficiency threshold, monitoring the reduction efficiency during the smelting process in real time through an integrated sensor, and comparing the monitored reduction efficiency with the threshold. If the reduction efficiency is greater than the reduction efficiency threshold, maintaining the current smelting process state;
[0061] Step S5: If the reduction efficiency is less than the reduction efficiency threshold, the monitored reduction efficiency threshold is fed back to the multivariable production process parameter optimization model and the dosage control model to optimize and adjust the initial production process parameters and the dosage and time point of the reducing agent so that the reduction efficiency is greater than the reduction efficiency threshold.
[0062] Furthermore, the step of obtaining irrelevant valid characteristic parameters in step S1 includes:
[0063] S101. Setting a collinearity correlation threshold, pre-processing the collected smelting process production process parameters, performing collinearity diagnosis on the pre-processed process parameters using a correlation algorithm, obtaining collinearity correlation coefficients between various variables, and comparing the obtained collinearity correlation coefficients; further, this embodiment uses a correlation algorithm to calculate correlation coefficients between any two smelting process production process parameters, and sets a collinearity correlation threshold of 0.5. When the calculated correlation between any two variables is greater than 0.5, it indicates that collinearity exists between the two calculated variables; the correlation algorithm in this embodiment includes a regression algorithm and a correlation matrix;
[0064] S102, setting the reduction efficiency as the target variable and the remaining process parameters as the decision variables, inputting the target variable and the remaining process parameters as the decision variables into the factor analysis, and obtaining the factor loading matrix of the target variable and the decision variables;
[0065] S103, setting a factor load threshold, comparing the load correlation coefficient between the decision variable and the target variable in the obtained factor load matrix with the factor load threshold, and if it is greater than the factor load threshold, it is a valid feature parameter; in this embodiment, the factor load threshold is set to 0.8;
[0066] S104. Compare the collinear correlation coefficient between the obtained valid characteristic parameters with the collinear correlation threshold. If it is greater than the collinear correlation threshold, retain the decision variable with the larger load correlation coefficient among the two decision variables with collinearity, delete the other decision variable, and obtain irrelevant valid characteristic parameters; for example, assume that both temperature and air supply volume are valid characteristic parameters of aluminum reduction efficiency, but the collinear correlation between temperature and air supply volume is 0.6, and the load correlation coefficient between temperature and aluminum reduction efficiency is 0.9; the load correlation coefficient between air supply volume and aluminum reduction efficiency is 0.8, then the air supply volume is discarded.
[0067] Further, see Figure 2 The construction process of the multivariable production process parameter optimization model in step S2 specifically includes:
[0068] S201. Using the production efficiency parameter among the irrelevant effective characteristic parameters as the target variable and the production process parameters among the irrelevant effective characteristic parameters as the decision variables, a multivariate production process parameter optimization model is constructed through a multivariate regression function, wherein the multivariate regression function formula is: ,in, represents the target variable in the production efficiency parameter, represents the decision variable in the nth production process parameter, is the weight coefficient of the corresponding decision variable;
[0069] S202, based on the constructed multiple regression function, the weighted coefficient vector in the multiple regression function is used as the input chromosome of the genetic algorithm, and the input chromosome is used to randomly generate the coefficient vector as the initialization population ;
[0070] S203. Utilize the production efficiency parameters and production process parameters in the unrelated effective characteristic parameters to construct a genetic algorithm multi-objective fitness function F and a constraint condition set S. The multi-objective fitness function F and the constraint condition set S are specifically:
[0071] ,
[0072]
[0073] in, Indicates the actual reduction efficiency of the product during the smelting process, Indicates the product reduction efficiency threshold during the smelting process. Further, in this embodiment, The historical reduction efficiency of the aluminum smelting industry during the smelting process is calculated by statistical methods to obtain the average value of the historical reduction efficiency; Indicates the total energy consumption to complete a smelting cycle. Indicates the quality of the output product after completing a smelting cycle. Indicates the amount of pollution emitted during a smelting cycle. Indicates the process scoring error of the output product after completing a smelting cycle process; further, the process scoring in this embodiment is scored by a comprehensive evaluation model built into the integrated sensor, and the input variables include the chemical composition, physical properties, and surface quality of the produced aluminum product. The chemical composition includes aluminum purity, impurity element content, oxide and nitride content; physical properties include density, tensile strength, and elongation; surface quality includes defect distribution and surface roughness; the comprehensive evaluation model includes a grey system theory model, random forest, hierarchical analysis method, and fuzzy comprehensive evaluation method; this embodiment The aluminum process score and aluminum process score threshold are obtained from real-time production. The aluminum process score threshold is set by technicians in this field according to production requirements. It represents the penalty function for the variable in the fitness function that violates the sth constraint in the constraint set. When the constraint is violated, =1, otherwise 0; It represents the expected energy consumption per unit time. Indicates the expected energy consumption fluctuation tolerance, which is a fixed value and is set by those skilled in the art based on production requirements. In this embodiment, it is set to 0.05; Indicates the average concentration of pollutants discharged per unit time during the smelting process. In this embodiment, the pollutants discharged are the average concentration of toxic substances in the wastewater discharged during the electrolytic aluminum process; It represents the average concentration threshold of pollutants emitted per unit time during the smelting process. This threshold is set by the specific regulations on pollutant emissions from the aluminum smelting industry. It represents the process scoring error threshold, which is a fixed constant and is set by technicians in this field according to the specific requirements of aluminum production. The smaller it is, the higher the level of production technology; It represents the total mass of the corresponding smelting material minerals put into the smelting process per unit time. Indicates the mineral smelting efficiency threshold, which is set by those skilled in the art based on the input bauxite and the average output rate of the industry; Indicates the pressure inside the smelting equipment at time t, It indicates the maximum pressure that the smelting equipment can withstand during operation. is the weighted coefficient corresponding to the fitness term;
[0074] S204. Set the maximum number of iterations and rate of change stop threshold , iterate the genetic algorithm and calculate the Pareto front change rate , the specific formula is Among them, Indicates the first The Pareto frontier solution set of the generation, Indicates that x is the first A solution in the Pareto frontier solution set of the first generation, y represents the obtained A solution in the Pareto front solution set of the generation, represents the 2-norm symbol;
[0075] S205, iteratively calculate the process of S201-S204 based on the maximum number of iterations and the change rate stop threshold set in S204. When , the iteration stops, and the weighted coefficient vector solution of a set of optimal decision variables of the multivariate regression function in S201 obtained by the search is output, and the obtained weighted coefficient vector solution is fed back to the multivariate regression function to initialize the production process parameters to obtain the initial production process parameters.
[0076] Furthermore, the specific steps of constructing the delivery control model in step S3 include:
[0077] S301. Set the following control constraints based on the production efficiency parameters and initial production process parameters among the obtained irrelevant effective characteristic parameters, specifically:
[0078] ,
[0079] in, Indicates the smelting temperature in the equipment at time t during the smelting process, In turn, they represent the minimum and maximum smelting temperatures in the equipment during the smelting process. Indicates the pressure inside the equipment at time t during the smelting process, In turn, they represent the minimum and maximum values of the smelting pressure during the smelting process. Indicates the concentration of reducing agent in the equipment at time t during the smelting process, In turn, they represent the minimum and maximum values of reducing agent concentration during the smelting process. Indicates the actual reduction efficiency of the product at time t during the smelting process; in this embodiment , and The prediction is obtained through support vector machine based on the historical maximum and minimum temperature, reducing agent concentration and pressure data corresponding to the aluminum production process;
[0080] S302: Setting the following trigger conditions according to the control constraint information in S301, specifically including:
[0081] When the temperature condition in the constraint condition is not met, the first execution information is triggered. The action to be executed corresponding to the first execution information is to control the furnace temperature control subsystem to adjust the temperature by generating a specific control signal through the input control model. When the temperature is too high, the cooling equipment is controlled to reduce the temperature.
[0082] When the pressure condition in the constraint condition is not met, the second execution information is triggered. The action executed by the second execution information is to control the valve of the smelting equipment to adjust the pressure by releasing gas through the specific control signal generated by the control model to achieve the target pressure;
[0083] When the reducing agent concentration condition in the constraint condition is not met, the third execution information is triggered. The action executed by the third execution information is to automatically adjust the amount of reducing agent added by using a specific control signal generated by the delivery control model. For example, the required quality of reducing agent is calculated based on the deviation between the current concentration and the set concentration value. According to the required quality of reducing agent obtained by calculation, the reducing agent adding device is controlled. According to the reduction reaction rate, temperature and pressure monitored by the sensor, the reducing agent is added at a constant speed. The concentration of the reducing agent during the addition process is monitored in real time by the sensor to precisely control the time point of adding the reducing agent. Furthermore, the reducing agent is added at a constant speed. The calculation formula is: ,in, is the speed adjustment coefficient, To calculate the required mass of reducing agent, Indicates the planned addition time; , represents the benchmark adjustment factor, represents the reaction rate as a function of temperature, represents the reaction rate as a function of pressure, and According to specific chemical kinetics techniques, it is obtained by fitting the experimental process;
[0084] When two or more of the temperature, pressure, or reducing agent concentration conditions in the constraint conditions are not met, the fourth execution information is triggered to jointly execute the first to third execution information;
[0085] When the restoration efficiency condition in the constraint condition is not satisfied, the fifth execution information is triggered, and the first to fourth execution information are jointly executed.
[0086] Furthermore, the specific steps of constructing the S3 delivery control model include:
[0087] S303: Setting the current input state of the reinforcement learning sub-model in the delivery control model according to the production efficiency parameter and the initial production process parameter in the obtained irrelevant effective feature parameters , perform the action and the reward function , specifically:
[0088] , , ,
[0089] in, Indicates the action that triggers the execution of the first execution information. Indicates the action that triggers the execution of the second execution information. Indicates the action that triggers the execution of the third execution information. Indicates the action that triggers the execution of the fourth execution information. Indicates the action that triggers the execution of the fifth execution information. Indicates the threshold value of reducing agent concentration during smelting process, It indicates the quality of the expected smelting product per unit mass of mineral input. Represents the corresponding weight coefficient in the reward function;
[0090] S304, using the reinforcement algorithm to build a reinforcement learning sub-model, and obtain the current input state , perform the action and the reward function Input into the reinforcement learning sub-model for training, and obtain the trained reinforcement learning sub-model;
[0091] S305: construct a control sub-model using a control algorithm, and initialize the parameters of the control sub-model using the temperature and pressure error values to obtain an initialized control sub-model;
[0092] S306: Integrate the trained reinforcement learning sub-model and control sub-model into the aluminum smelting control system, and accurately control the aluminum smelting process based on the real-time monitoring data of the aluminum smelting process and the control signal corresponding to the production process. Further, generate the corresponding control signal The specific formula is:
[0093] ,
[0094] in, Indicates the signal data generated at time t-1. In this embodiment, this parameter is added to maintain the continuity of the instruction. In turn, 、 、 、 and The corresponding evaluation value is obtained by calculating the reinforcement learning sub-model; express The j-th evaluation value in is calculated by the reinforcement learning sub-model when the action that triggers the execution of the j-th execution information is executed; It indicates the difference between the real-time temperature value and the expected temperature of the smelting equipment monitored at time t. It represents the difference between the real-time pressure value and the expected pressure in the smelting equipment monitored at time t. Indicates the length of a cycle of aluminum smelting, They represent the proportional coefficient, integral coefficient and differential coefficient in the control sub-model respectively.
[0095] Example 2
[0096] See also Figure 3 , another embodiment provided by the present invention: a decision analysis system for AI big data modeling applied to aluminum smelting, comprising: a data preprocessing module, a parameter optimization module, a delivery control module, a monitoring feedback module and an interaction module;
[0097] The data preprocessing module is used to collect the production process parameters of the smelting process in real time and preprocess the data to extract effective features; the data preprocessing module includes a data acquisition unit and a feature filtering unit;
[0098] The data acquisition unit is used to collect various production parameters and production efficiency parameters in the smelting process in real time through sensors; the feature filtering unit is used to process the collected data through the feature filtering algorithm built into the sensor to extract irrelevant valid feature parameters;
[0099] The parameter optimization module is used to build a multivariable production process parameter optimization model and optimize the irrelevant effective characteristic parameters to obtain the initial production process parameters; the parameter optimization module includes a model construction unit and a parameter output unit;
[0100] A model building unit is used to build and train a multivariable production process parameter optimization model through a multivariable regression algorithm and a genetic algorithm; a parameter output unit is used to initialize and optimize the obtained irrelevant effective characteristic parameters through the multivariable production process parameter optimization model to obtain initial production process parameters;
[0101] A dosing control module is used to build a dosing control model and control the dosing amount and timing of the reducing agent according to the initial production process parameters and the dosing control model;
[0102] The monitoring and feedback module is used to monitor the reduction efficiency during the smelting process in real time and use the monitoring results to feedback and optimize the feeding amount and timing. The monitoring and feedback module includes a monitoring and feedback unit and a parameter optimization unit.
[0103] The monitoring and feedback unit is used to monitor the reduction efficiency of the smelting process in real time through integrated sensors, and use the set threshold to judge and feedback the monitored reduction efficiency; the parameter optimization unit is used to adjust the production process parameters and the input control strategy according to the feedback information to improve the reduction efficiency;
[0104] The interactive module is used to provide an interactive interface for setting smelting process parameters and visualizing data.
[0105] Example 3
[0106] A computer-readable storage medium stores computer instructions, which, when executed, execute a decision analysis method for AI big data modeling applied to aluminum smelting.
[0107] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific embodiments. The above-mentioned specific embodiments are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also change, modify, replace and modify the above-mentioned embodiments without departing from the scope of protection of the purpose of the present invention and the claims, and all of these are protected by the present invention.
Claims
1. A decision analysis method for AI big data modeling applied to aluminum smelting, characterized in that: include: Step S1, collecting the production process parameters and production efficiency parameters of the smelting process in real time through sensors, and filtering the collected parameters using a feature filtering algorithm built into the sensor to obtain irrelevant effective feature parameters; Step S2: construct a multivariate production process parameter optimization model based on a multivariate regression algorithm and a genetic algorithm, input the obtained irrelevant effective characteristic parameters into the multivariate production process parameter optimization model, and obtain the initial production process parameters of the smelting process; Step S3: construct a production process delivery control model based on a reinforcement learning algorithm, and deploy the constructed delivery control model to the smelting process control platform. Meanwhile, input the obtained initial production process parameters into the smelting process control platform, and control the delivery amount and time point of the reducing agent through the constructed delivery control model; Step S4, setting a reduction efficiency threshold, monitoring the reduction efficiency in the smelting process in real time through an integrated sensor, and comparing the monitored reduction efficiency with the threshold. If the reduction efficiency is greater than the reduction efficiency threshold, maintaining the current smelting process state; Step S5: If the reduction efficiency is less than the reduction efficiency threshold, the monitored reduction efficiency is fed back to the multivariable production process parameter optimization model and the delivery control model, and the initial production process parameters and the delivery amount and time point of the reducing agent are optimized and adjusted so that the reduction efficiency is greater than the reduction efficiency threshold; The specific steps of constructing the delivery control model in step S3 include: S301, setting the following control constraints according to the production efficiency parameters and initial production process parameters in the obtained irrelevant effective characteristic parameters, specifically: , in, Indicates the smelting temperature in the equipment at time t during the smelting process, It represents the minimum and maximum smelting temperature in the equipment during the smelting process. It represents the pressure inside the equipment at time t during the smelting process. They represent the minimum and maximum values of the smelting pressure during the smelting process, respectively. It indicates the concentration of reducing agent in the equipment at time t during the smelting process. They represent the minimum and maximum values of reducing agent concentration in the smelting process, respectively. It indicates the actual reduction efficiency of the product at time t during the smelting process. Indicates the reduction efficiency threshold of products during the smelting process; S302: according to the control constraint conditions in S301, the following trigger conditions are set, specifically including: When the temperature condition in the constraint condition is not met, triggering the first execution information; When the pressure condition in the constraint condition is not met, the second execution information is triggered; When the reducing agent concentration condition in the constraint condition is not met, triggering the third execution information; When two or more of the temperature, pressure or reducing agent concentration conditions in the constraint conditions are not met, the fourth execution information is triggered; When the restoration efficiency condition in the constraint condition is not met, the fifth execution information is triggered, and the first execution information to the fourth execution information are jointly executed; S303: Setting the current input state of the reinforcement learning sub-model in the delivery control model according to the production efficiency parameters and the initial production process parameters in the obtained irrelevant effective feature parameters , perform the action And the reward function , specifically: , , , in, Indicates the action that triggers the execution of the first execution information. Indicates the action that triggers the execution of the second execution information. Indicates the action that triggers the execution of the third execution information. Indicates the action that triggers the execution of the fourth execution information. Indicates the action that triggers the execution of the fifth execution information. Indicates the threshold value of reducing agent concentration in the smelting process, It indicates the quality of the expected smelting product per unit mass of mineral input. represents the corresponding weighted coefficient in the reward function, Indicates the quality of the output product after completing a smelting cycle.
2. The decision analysis method for AI big data modeling applied to aluminum smelting according to claim 1, characterized in that: The step of acquiring irrelevant valid characteristic parameters in step S1 includes: S101, setting a collinear correlation threshold, preprocessing the collected smelting process production process parameters, performing collinearity diagnosis on the preprocessed process parameters through a correlation algorithm, obtaining collinear correlation coefficients between various variables, and comparing the obtained collinear correlation coefficients; S102, setting the reduction efficiency as the target variable and the remaining process parameters as the decision variables, inputting the target variable and the remaining process parameters as the decision variables into the factor analysis, and obtaining the factor loading matrix of the target variable and the decision variables; S103, setting a factor load threshold, comparing the load correlation coefficient between the decision variable and the target variable in the obtained factor load matrix with the factor load threshold, and if it is greater than the factor load threshold, it is a valid characteristic parameter; S104. Compare the collinear correlation coefficient between the obtained valid characteristic parameters with the collinear correlation threshold. If it is greater than the collinear correlation threshold, retain the decision variable with the larger load correlation coefficient among the two decision variables with collinearity, and delete the other decision variable to obtain irrelevant valid characteristic parameters.
3. A decision analysis method for AI big data modeling applied to aluminum smelting as claimed in claim 2, characterized in that: The construction process of the multivariate production process parameter optimization model in step S2 specifically includes: S201, using the production efficiency parameter in the irrelevant effective characteristic parameters as the target variable, using the production process parameters in the irrelevant effective characteristic parameters as the decision variable, and constructing a multivariate production process parameter optimization model through a multivariate regression function, wherein the multivariate regression function formula is: ,in, represents the target variable in the production efficiency parameter, represents the decision variable in the nth production process parameter, is the weight coefficient of the corresponding decision variable.
4. The decision analysis method for AI big data modeling applied to aluminum smelting as claimed in claim 3 is characterized in that: The process of constructing the multivariate production process parameter optimization model in step S2 specifically includes: S202, based on the constructed multiple regression function, using the weighted coefficient vector in the multiple regression function as the input chromosome of the genetic algorithm, and using the input chromosome to randomly generate a coefficient vector as an initialization population; S203, constructing a genetic algorithm multi-objective fitness function F and a constraint condition set S using production efficiency parameters and production process parameters in unrelated effective characteristic parameters; S204, set the maximum number of iterations and the rate of change stop threshold , iterate the genetic algorithm and calculate the Pareto front change rate The specific formula is ,in, Indicates the first The Pareto frontier solution set of the generation, Indicates that x is the first A solution in the Pareto front solution set of the first generation, y represents the obtained A solution in the Pareto front solution set of the generation, represents the 2-norm symbol; S205, iteratively calculate the process of S201-S204 based on the maximum number of iterations and the change rate stop threshold set in S204. When , the iteration stops, and the weighted coefficient vector solution of a set of optimal decision variables of the multivariate regression function in S201 obtained by searching is output, and the obtained weighted coefficient vector solution is fed back to the multivariate regression function to initialize the production process parameters to obtain the initial production process parameters.
5. The decision analysis method for AI big data modeling applied to aluminum smelting according to claim 4, characterized in that: The multi-objective fitness function F and constraint condition set S in S203 are specifically: , ; in, Indicates the actual reduction efficiency of the product during the smelting process. Indicates the reduction efficiency threshold of products during the smelting process. It represents the total energy consumption to complete a smelting cycle. It indicates the amount of pollution emitted during a smelting cycle. It indicates the process score error of the output product after completing a smelting cycle; It represents the penalty function for the variable in the fitness function that violates the sth constraint in the constraint set. It represents the expected energy consumption per unit time. Indicates the expected energy consumption fluctuation tolerance, It represents the average concentration of pollutants emitted per unit time during the smelting process. It represents the average concentration threshold of pollutants emitted per unit time during the smelting process. Represents the process scoring error threshold, which is a fixed constant. It indicates the total mass of the corresponding smelting material mineral put into the smelting process per unit time. Indicates the mineral smelting efficiency threshold, It indicates the pressure inside the smelting equipment at time t, It indicates the maximum pressure that the smelting equipment can withstand during operation. is the weight coefficient corresponding to the fitness term.
6. A decision analysis system for AI big data modeling applied to aluminum smelting, which is implemented based on a decision analysis method for AI big data modeling applied to aluminum smelting according to any one of claims 1 to 5, characterized in that: include: Data preprocessing module, parameter optimization module, and delivery control module; The data preprocessing module is used to collect the production process parameters of the smelting process in real time and preprocess the data to extract effective features; The data preprocessing module includes a data acquisition unit and a feature filtering unit; The data acquisition unit is used to collect various production parameters and production efficiency parameters in the smelting process in real time through sensors; the feature filtering unit is used to process the collected data through a feature filtering algorithm built into the sensor to extract irrelevant effective feature parameters; The delivery control module is used to construct a delivery control model and control the delivery amount and time point of the reducing agent according to the initial production process parameters and the delivery control model.
7. A decision analysis system for AI big data modeling applied to aluminum smelting as claimed in claim 6, characterized in that: The parameter optimization module includes a model building unit and a parameter output unit; The model building unit is used to build and train a multivariable production process parameter optimization model through a multivariable regression algorithm and a genetic algorithm; the parameter output unit is used to initialize and optimize the obtained irrelevant effective characteristic parameters through the multivariable production process parameter optimization model to obtain initial production process parameters.
8. A computer-readable storage medium, characterized in that: Computer instructions are stored thereon, and when the computer instructions are executed, a decision analysis method for AI big data modeling applied to aluminum smelting as described in any one of claims 1-5 is executed.
