Rotary kiln temperature field prediction and optimization method and system based on machine learning and CFD coupling
Through the coupling method of machine learning and CFD, combined with decision makers' preferences, real-time accurate prediction and optimization of the temperature field of the rotary kiln is achieved, solving the problem of insufficient real-time and accuracy of temperature field prediction and optimization in the existing technology, and reducing the Energy consumption and improve product quality and extend the life of the rotary kiln.
Patent Information
- Application Number
- CN202411054174.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-02
- Publication Date
- 2024-12-03
AI Technical Summary
The prediction and optimization of the rotary kiln temperature field has problems such as high labor costs and poor real-time tracking performance. Traditional mathematical modeling and machine learning models have insufficient accuracy. CFD simulation computing resources are high and not real-time, and it fails to meet the needs of industrial production. The optimization model is optimized. The failure to consider decision makers' opinions leads to unsatisfactory in the diversity of optimization results.
Using a method based on machine learning and CFD coupling, we can use data preparation, training multiple machine learning models, construct objective function models, and evaluate the Pareto solution set by using multi-objective decision-making methods. We can select the best optimization solution based on decision makers' preferences to achieve Real-time accurate prediction and optimization of the rotary kiln temperature field.
Real-time online prediction and optimization of the rotary kiln temperature field is achieved, which shortens the prediction optimization time, improves accuracy, reduces energy consumption and product quality decline, extends the life of the rotary kiln, provides optimized solutions that are satisfactory to decision makers, and has scalable Sex and practicality.
Smart Images

Figure CN119066953A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of temperature field prediction and optimization, and specifically relates to a rotary kiln temperature field prediction and optimization method and system based on machine learning and CFD coupling. Background Art
[0002] As an important industrial equipment, rotary kiln is widely used in metallurgy and mining industries. However, rotary kiln is a nonlinear, multivariable, strongly coupled system, which is a typical "black box" reactor. The temperature field in the kiln is easily affected by factors such as material reaction thermodynamics, heat transfer, mass transfer and material movement state, which leads to the formation of rings, resulting in increased energy consumption and reduced product quality. Therefore, it is urgent to study advanced temperature sensing and optimization technologies to extend the service life of rotary kilns and improve product quality.
[0003] Regarding temperature sensing technology, it can be mainly divided into two categories: 1) hard measurement based on instrumentation; 2) soft measurement based on mathematical modeling, simulation and algorithm. In terms of traditional hard measurement, manual timed measurement of the shell temperature of the rotary kiln is one of the main methods for rotary kiln temperature detection, but this method has great disadvantages, consumes a lot of labor costs, and has poor real-time tracking performance. In addition, there are two popular hard measurement methods, namely infrared temperature measurement and thermocouple temperature measurement. In terms of soft measurement technology, general soft measurement is divided into three categories: mathematical modeling, CFD simulation and machine learning model establishment. The traditional mathematical modeling method analyzes the heat transfer mechanism of the rotary kiln through a large number of experiments, and then establishes relevant mathematical equations to obtain effective mathematical models such as process heat transfer simulation and optimization strategy. However, the results of model establishment through mathematical formulas are not accurate, and the results corresponding to different working conditions may be different, so there are accidents and errors. The data-driven model based on the machine learning model provides new ideas, but the machine learning-based method is a black box mechanism. The accuracy of the results depends greatly on the big data stored in the industrial operation. The actual industrial process is extremely complex, and the data is missing, abnormal, and redundant, which makes the final prediction of the model inaccurate. With the improvement of computer capabilities, many scholars have begun to use computational fluid dynamics (CFD) to simulate the temperature field of the rotary kiln, but it requires extremely high computing resources and cannot meet the needs of real-time detection in industrial production. Therefore, coupling CFD with machine learning is an important idea for the current research on industrial prediction variables. This hybrid method can combine the fast real-time performance of machine learning with the accuracy of CFD.
[0004] In addition, during the operation of the rotary kiln, it is particularly important to optimize the distribution of the temperature field in the kiln. However, the current main optimization model does not take into account the opinions and ideas of decision makers. The optimization results are diverse, and it is impossible to screen out an optimization method that satisfies the decision makers. Therefore, by improving the optimization method and integrating the ideas of decision makers, decision makers can use their preferences to re-select a key solution space. This improved optimization method provides new ideas for the rotary kiln temperature optimization plan, thereby obtaining an optimization solution that satisfies the decision makers. Summary of the invention
[0005] In order to solve the problems existing in the prior art, the main purpose of the present invention is to propose a rotary kiln temperature field prediction and optimization method and system based on machine learning and CFD coupling.
[0006] To solve the above technical problems, according to one aspect of the present invention, the present invention provides the following technical solutions:
[0007] A method for predicting and optimizing the temperature field of a rotary kiln based on machine learning and CFD coupling includes the following steps:
[0008] S1. In the data preparation stage, a large number of data sets generated by experimental design are stored in the database and the data sets are standardized;
[0009] S2. Train a variety of different machine learning models and select the method that performs better in predicting the temperature field of the rotary kiln as the model for predicting the overall temperature field of the rotary kiln;
[0010] S3, using the selected model for predicting the overall temperature field of the rotary kiln to construct two objective function models, and then using the model as an input of the objective function to obtain a parameter optimization solution;
[0011] S4. Evaluate the Pareto solution set based on the multi-objective decision-making method and select the best solution from many solutions to ultimately achieve the optimization of the rotary kiln temperature field.
[0012] As a preferred solution of the rotary kiln temperature field prediction and optimization method based on machine learning and CFD coupling described in the present invention, wherein: in the step S1, the data set is generated by CFD simulation to establish a database of the global temperature field distribution in the rotary kiln.
[0013] As a preferred scheme of the rotary kiln temperature field prediction and optimization method based on machine learning and CFD coupling described in the present invention, wherein: in the step S2, the machine learning models are RF (random forest), CNN (convolutional neural network), XGBoost (extreme gradient boosting) and BPNN (back propagation neural network); the coal powder injection amount, secondary air temperature, secondary air speed, volatile matter and fixed carbon are set as learning features of the machine learning model, and then the machine learning model with the best performance is selected after hyperparameter tuning to perform real-time and accurate prediction of the global temperature field of the rotary kiln.
[0014] As a preferred solution of the rotary kiln temperature field prediction and optimization method based on machine learning and CFD coupling described in the present invention, wherein: in the step S2, based on the values of MSE and MAPE, it is judged whether the machine learning model performs well in the rotary kiln temperature field prediction.
[0015] As a preferred solution of the rotary kiln temperature field prediction and optimization method based on machine learning and CFD coupling described in the present invention, wherein: in the step S2, the machine learning model with the best performance is RF.
[0016] As a preferred embodiment of the rotary kiln temperature field prediction and optimization method based on machine learning and CFD coupling described in the present invention, in which: in the step S3, the two objective functions are the maximum temperature of the rotary kiln along the axis and the width of the roasting zone of the rotary kiln along the axis, respectively, and the machine learning model is used in the fitness objective function of the optimization algorithm, and the five decision variables of coal powder injection amount, secondary air temperature, secondary air speed, volatile matter and fixed carbon are used as objective function inputs, which are used to predict the two targets respectively.
[0017] As a preferred solution of the rotary kiln temperature field prediction and optimization method based on machine learning and CFD coupling described in the present invention, wherein: in the step S4, when evaluating the Pareto solution, the two key performance indicators are weighed by introducing the decision maker's preference, so as to objectively determine the best optimization solution in the Pareto solution set with preferences.
[0018] As a preferred solution of the rotary kiln temperature field prediction and optimization method based on machine learning and CFD coupling described in the present invention, wherein: in the step S4, the multi-objective decision-making method is an improved R-NSGA-II algorithm.
[0019] To solve the above technical problem, according to another aspect of the present invention, the present invention provides the following technical solution:
[0020] A rotary kiln temperature field prediction and optimization system based on machine learning and CFD coupling, including:
[0021] The data set storage and processing module uses the large amount of data sets generated by the experimental design in the data preparation stage to store them in the database and standardize the data sets;
[0022] The machine learning model training module trains a variety of different machine learning models and selects the method with better performance in the prediction of the rotary kiln temperature field as the model for predicting the overall temperature field of the rotary kiln;
[0023] The parameter optimization solution obtaining module uses the selected model for predicting the overall temperature field of the rotary kiln to construct two objective functions, and then uses the model as the input of the objective function to obtain the parameter optimization solution;
[0024] The optimization module of the rotary kiln temperature field evaluates the Pareto solution set based on the multi-objective decision-making method, selects the best solution from many solutions, and finally realizes the optimization of the rotary kiln temperature field.
[0025] The beneficial effects of the present invention are as follows:
[0026] The present invention proposes a method and system for predicting and optimizing the temperature field of a rotary kiln based on machine learning and CFD coupling, which can optimize the width of the roasting zone in the rotary kiln and the maximum temperature of the region, thereby optimizing the temperature field of the entire rotary kiln, and finding the best process parameter combination for optimizing the temperature field. The optimization effect is significant, and the solution is searched according to the reference points and weights provided by the decision maker, and the Pareto preferred solution set space is reconstructed, which effectively controls the diversity of the Pareto solution set. The screened Pareto solution set is sorted through multi-objective decision-making, and a single optimal solution is searched, which saves the decision maker's time and trial and error costs, and has guiding significance for actual industrial production. The optimization effect is scalable, and has a positive effect on product quality control and the extension of the life of the rotary kiln. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] 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 description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.
[0028] Figure 1 Flow chart of the method of the present invention.
[0029] Figure 2 This is the CFD geometric model diagram of the rotary kiln.
[0030] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0031] The following will be described clearly and completely in conjunction with the technical solutions in the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0032] The present invention proposes a method and system for predicting and optimizing the temperature field of a rotary kiln based on machine learning and CFD coupling. The performance of the rotary kiln temperature field is predicted by a machine learning method, and the data of the global temperature field can be accurately predicted in real time. At the same time, the temperature field of the rotary kiln is optimized globally by an improved multi-objective optimization algorithm, and a multi-objective decision method is used to evaluate the preferred Pareto frontier to determine the best optimization solution. The prediction and optimization time of the present invention is greatly shortened compared to the CFD simulation time, and the accuracy is high. It can realize real-time online prediction of various field information such as temperature field and can timely optimize and adjust the temperature field, thereby improving the quality of pellet production, reducing the increase in energy consumption and the decline in product quality caused by ring formation, and providing an effective and time-saving tool for industrial temperature prediction and optimization.
[0033] The main problem to be solved by the present invention is the prediction and optimization of the temperature field inside the rotary kiln, and the accurate and real-time acquisition of the distribution values of the global temperature field during the pellet production process, which cannot be measured by manual operation, thereby providing a reference for the optimization of the temperature field inside the rotary kiln, and being able to accurately optimize the temperature field in a timely manner, thereby reducing the increase in energy consumption and the decline in product quality caused by the formation of rings, extending the working life of the rotary kiln, and improving product quality.
[0034] According to one aspect of the present invention, the present invention provides the following technical solution:
[0035] like Figure 1 As shown, a rotary kiln temperature field prediction and optimization method based on machine learning and CFD coupling includes the following steps:
[0036] S1. Expand the data set to improve the accuracy of prediction. In the data preparation stage, a large number of data sets generated by experimental design are used for database storage to ensure the accuracy of subsequent predictions and optimization models, and the data sets are standardized.
[0037] S2. Train a variety of different machine learning models and select the method that performs better in predicting the temperature field of the rotary kiln as the model for predicting the overall temperature field of the rotary kiln;
[0038] S3, using the selected model for predicting the overall temperature field of the rotary kiln to construct two objective function models, and then using the model as an input of the objective function to obtain a parameter optimization solution;
[0039] S4. Evaluate the Pareto solution set based on the multi-objective decision-making method and select the best solution from many solutions to ultimately achieve the optimization of the rotary kiln temperature field.
[0040] Preferably, in step S1, the data set is generated by CFD simulation to establish a database of the global temperature field distribution in the rotary kiln; the rotary kiln model fully considers the complex combustion process in the rotary kiln, such as turbulent particle flow and mass and momentum equations, so the CFD model has high reliability and can be used for subsequent research work.
[0041] Preferably, in step S2, the machine learning models are RF, CNN, XGBoost and BPNN respectively; the coal powder injection amount, secondary air temperature, secondary air speed, volatile matter and fixed carbon are set as learning features of the machine learning model, and then the machine learning model with the best performance is selected after hyperparameter tuning to perform real-time and accurate prediction of the global temperature field of the rotary kiln.
[0042] Preferably, in step S2, whether the machine learning model performs well in the prediction of the rotary kiln temperature field is determined based on the values of MSE and MAPE.
[0043] Preferably, in step S2, the machine learning model with the best performance is RF.
[0044] Preferably, in step S3, the two objective functions are the maximum temperature of the rotary kiln along the axis and the width of the roasting zone of the rotary kiln along the axis, and the roasting zone of the rotary kiln is not a fixed section of the kiln space, but refers to the space where the material temperature is above 1350°C, and the length of the roasting zone changes dynamically according to the temperature field. Therefore, the length of the roasting zone of the same rotary kiln is different under different working conditions. These two indicators can comprehensively evaluate the temperature field optimization of the rotary kiln, and then can be used to evaluate the quality of pellet roasting; the machine learning model is used in the fitness objective function of the optimization algorithm, and the five decision variables of coal powder injection amount, secondary air temperature, secondary air speed, volatile matter and fixed carbon are used as objective function inputs, which are used to predict the two goals respectively.
[0045] Preferably, in step S4, when evaluating the Pareto solution, two key performance indicators are weighed by introducing the decision maker's preference, so as to objectively determine the best optimization solution in the Pareto solution set with preferences.
[0046] Preferably, in step S4, the multi-objective decision-making method is an improved R-NSGA-II algorithm.
[0047] According to another aspect of the present invention, the present invention provides the following technical solution:
[0048] A rotary kiln temperature field prediction and optimization system based on machine learning and CFD coupling, including:
[0049] The data set storage and processing module expands the data set to improve the accuracy of prediction. In the data preparation stage, a large number of data sets generated by experimental design are used for database storage, which ensures the accuracy of subsequent predictions and optimization models, and standardizes the data set.
[0050] The machine learning model training module trains a variety of different machine learning models and selects the method with better performance in the prediction of the rotary kiln temperature field as the model for predicting the overall temperature field of the rotary kiln;
[0051] The parameter optimization solution obtaining module uses the selected model for predicting the overall temperature field of the rotary kiln to construct two objective functions, and then uses the model as the input of the objective function to obtain the parameter optimization solution;
[0052] The optimization module of the rotary kiln temperature field evaluates the Pareto solution set based on the multi-objective decision-making method, selects the best solution from many solutions, and finally realizes the optimization of the rotary kiln temperature field.
[0053] The technical solution of the present invention is further described below in conjunction with specific embodiments.
[0054] Example
[0055] Create a two-dimensional plane model file of a rotary kiln with a length of 40m and a cylinder diameter of 6m. The CFD model of the rotary kiln is shown in the figure below. Figure 2 As shown, the drawn model file is then imported into the ANSYS mesh generation software for mesh division. In order to speed up the convergence of the model, this embodiment uses hexahedral meshes instead of tetrahedral meshes for division. Due to the strong turbulence and fluctuation of the flow field near the burner, the surrounding area is meshed. In addition, more model details are shown in Table 1, and the set initial operating conditions are shown in Table 2.
[0056] Table 1 Rotary kiln geometric parameter settings
[0057]
[0058] Table 2 Initial operating conditions of rotary kiln
[0059]
[0060] Before predicting and optimizing the temperature field, it is necessary to build a database, simulate the actual pellet production of a steel plant through CFD simulation, and normalize the data. All processed data are input into the temperature field prediction and optimization database for subsequent training to generate the proxy model, which will help the subsequent optimization of the rotary kiln temperature field.
[0061] Subsequently, the temperature field of the rotary kiln is predicted and optimized, and the performance of the rotary kiln temperature field is predicted by machine learning methods, and the data of the global temperature field is accurately predicted in real time; at the same time, the temperature field of the rotary kiln is optimized globally by an improved multi-objective optimization algorithm, and the multi-objective decision-making method is used to evaluate the preferred Pareto frontier to determine the best optimization solution. The prediction and optimization time of this embodiment is greatly shortened compared to the CFD simulation time, and the accuracy is high. It can realize real-time online prediction of various field information such as temperature field and can timely optimize and adjust the temperature field, thereby improving the quality of pellet production and reducing the increase in energy consumption and the decline in product quality caused by ring formation.
[0062] Taking pulverized coal injection rate, secondary air temperature, secondary air speed, volatile matter and fixed carbon as the learning features of the machine learning model, several different machine learning models are used to predict the global temperature field of the rotary kiln. The accuracy of the final prediction results is shown in Table 3 below.
[0063] Table 3. Accuracy of prediction results of machine learning models
[0064]
[0065] In the data set distribution of this study, the MSE and MAPE values are the smallest in the RF model. This shows that for the prediction of rotary kiln temperature, the use of the random forest model for feature extraction in this data set is more effective, and can achieve accurate measurement and have more stable numerical performance. In order to comprehensively consider the stability and accuracy of subsequent prediction capabilities, the RF model is selected as the prediction model of the rotary kiln temperature field, which has important reference value for the prediction of rotary kilns in industrial sites.
[0066] Then, the improved R-NSGA-II algorithm was used to optimize the width of the kiln roasting zone and the highest temperature of the region, integrating the views and preferences of the decision makers. The optimized Pareto front solution set can better satisfy the decision makers. Then, 5 Pareto front solution sets were randomly selected, and the highest temperature was reduced from 1645.85℃ to 1575.70℃, and the roasting zone temperature was also optimized. In the above solution set, the width l of all roasting zones is distributed at 30m and above, and the highest temperature of the region is below 1580℃. The optimization effect is relatively satisfactory, and the prediction accuracy is also high. Three of the optimization schemes are shown in Table 4.
[0067] Table 4 Comparison of optimization scheme and baseline results
[0068]
[0069] Traditional rotary kiln temperature perception and optimization often have the disadvantages of low accuracy and lack of real-time performance. The method of the present invention can perceive the temperature field of the rotary kiln in real time with high efficiency. The temperature in the kiln is optimized by sensing the temperature field. The previous traditional optimization did not take into account the opinions and ideas of decision makers, and the optimization results were diverse, and it was impossible to screen out an optimization method that was satisfactory to the decision makers. The improved reference point-based NSGA-II method of the present invention provides a new idea for the rotary kiln temperature optimization solution. This improvement can integrate the ideas of decision makers, and decision makers can use preferences to re-select a key solution space, which can provide a solid theoretical basis and important reference for decision makers at industrial sites.
[0070] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. All equivalent structural changes made by using the contents of the present invention specification under the inventive concept of the present invention, or directly / indirectly applied in other related technical fields are included in the patent protection scope of the present invention.
Claims
1. A method for predicting and optimizing the temperature field of a rotary kiln based on machine learning and CFD coupling, characterized in that: The steps include: S1. In the data preparation stage, a large number of data sets generated by experimental design are stored in the database and the data sets are standardized; S2. Train a variety of different machine learning models and select the method that performs better in predicting the temperature field of the rotary kiln as the model for predicting the overall temperature field of the rotary kiln; S3, using the selected model for predicting the overall temperature field of the rotary kiln to construct two objective functions, and then using the model as an input of the objective function to obtain a parameter optimization solution; S4. Evaluate the Pareto solution set based on the multi-objective decision-making method and select the best solution from many solutions to ultimately achieve the optimization of the rotary kiln temperature field.
2. The rotary kiln temperature field prediction and optimization method based on machine learning and CFD coupling according to claim 1 is characterized in that: In step S1, the data set is generated by using CFD simulation to establish a database of the global temperature field distribution in the rotary kiln.
3. The rotary kiln temperature field prediction and optimization method based on machine learning and CFD coupling according to claim 1 is characterized in that: In step S2, the machine learning models are RF, CNN, XGBoost and BPNN respectively; the pulverized coal injection amount, secondary air temperature, secondary air speed, volatile matter and fixed carbon are set as learning features of the machine learning model, and then the machine learning model with the best performance is selected after hyperparameter tuning to perform real-time and accurate prediction of the global temperature field of the rotary kiln.
4. The rotary kiln temperature field prediction and optimization method based on machine learning and CFD coupling according to claim 3 is characterized in that: In step S2, whether the machine learning model performs well in the prediction of the rotary kiln temperature field is determined based on the values of MSE and MAPE.
5. The rotary kiln temperature field prediction and optimization method based on machine learning and CFD coupling according to claim 3 is characterized in that: In step S2, the best performing machine learning model is RF.
6. The rotary kiln temperature field prediction and optimization method based on machine learning and CFD coupling according to claim 1 is characterized in that: In step S3, the two objective functions are the maximum temperature of the rotary kiln along the axis and the width of the roasting zone of the rotary kiln along the axis, respectively. The machine learning model is used in the fitness objective function of the optimization algorithm, and the five decision variables of coal powder injection amount, secondary air temperature, secondary air speed, volatile matter and fixed carbon are used as objective function inputs to predict the two targets respectively.
7. The rotary kiln temperature field prediction and optimization method based on machine learning and CFD coupling according to claim 1 is characterized in that: In step S4, when evaluating the Pareto solution, the decision maker's preference is introduced to weigh the two key performance indicators, so as to objectively determine the best optimization solution in the Pareto solution set with preferences.
8. The rotary kiln temperature field prediction and optimization method based on machine learning and CFD coupling according to claim 1 is characterized in that: In step S4, the multi-objective decision-making method is an improved R-NSGA-II algorithm.
9. A rotary kiln temperature field prediction and optimization system based on machine learning and CFD coupling, characterized in that: The method for predicting and optimizing the temperature field of a rotary kiln based on machine learning and CFD coupling according to claim 1 comprises: The data set storage and processing module uses the large amount of data sets generated by the experimental design in the data preparation stage to store them in the database and standardize the data sets; The machine learning model training module trains a variety of different machine learning models and selects the method with better performance in the prediction of the rotary kiln temperature field as the model for predicting the overall temperature field of the rotary kiln; The parameter optimization solution obtaining module uses the selected model for predicting the overall temperature field of the rotary kiln to construct two objective functions, and then uses the model as the input of the objective function to obtain the parameter optimization solution; The optimization module of the rotary kiln temperature field evaluates the Pareto solution set based on the multi-objective decision-making method, selects the best solution from many solutions, and finally realizes the optimization of the rotary kiln temperature field.
Citation Information
Cited By
Activated carbon hazardous waste treatment method
CN121082666A
Multi-energy coupling temperature field control method and system based on intelligent optimization
CN121346356A
A Multi-Energy Coupled Temperature Field Control Method and System Based on Intelligent Optimization
CN121346356B
