Rotary furnace parameter optimization recommendation method based on data analysis
By using a data analysis-based rotary furnace parameter optimization method, which employs differential evolution algorithm and environmental correction coefficients to optimize rotary furnace parameters, the problem of low control precision in rotary furnaces is solved, and optimal parameter recommendation and finished product quality improvement are achieved under complex environments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGZHOU SOUTHSTAR MACHINE FACILITIES
- Filing Date
- 2026-04-08
- Publication Date
- 2026-05-12
AI Technical Summary
Existing rotary oven control technology lacks the ability to sense and adaptively compensate for environmental disturbances, and cannot determine the optimal parameters in complex industrial baking scenarios, resulting in low control accuracy and affecting the quality of finished products.
By using a data analysis-based rotary furnace parameter optimization recommendation method, the maximum allowable variation of each control parameter is obtained. Combined with differential evolution algorithm and environmental correction coefficient, adaptive cross-operation is generated to focus on key parameters that have a significant impact on the quality of finished products and optimize and recommend rotary furnace parameters.
It improves the accuracy of rotary furnace control and the convergence speed of the iterative process, ensuring the recommendation of optimal process parameters in complex environments, thereby enhancing the stability of finished product quality and production efficiency.
Smart Images
Figure CN122018610A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of parameter control technology, and in particular to a method for optimizing and recommending parameters of a rotary furnace based on data analysis. Background Technology
[0002] Food baking is a crucial step in modern food manufacturing. Rotary ovens, through the coordinated control of multiple parameters such as temperature, humidity, and airflow, can complete the baking process for pasta products. As consumer demands for food quality continue to rise, ensuring the color and taste of each batch of finished products while simultaneously improving production efficiency has become a key indicator for evaluating the performance of baking control systems.
[0003] Existing rotary oven control technology mainly adopts a parameter control mode based on fixed recipes. In actual operation, operators typically manually set the target temperature, humidity, and wind speed values for each baking stage according to a preset standard baking process sheet. After receiving the instructions, the control system drives the heating components, humidifiers, and fans to operate according to the target temperature, humidity, and wind speed values for the corresponding baking stage, gradually bringing the oven environment closer to the set values and maintaining it until baking is complete.
[0004] However, fixed-formula control lacks the ability to sense and adaptively compensate for environmental temperature and humidity, and cannot adapt to external environmental interference. In addition, the sensitivity contribution of each control parameter to the quality of the finished product is different, and the adjustment range of each control parameter directly affects the improvement effect of the quality of the finished product. Ignoring the different effects of each control parameter on the quality of the finished product, it is impossible to determine the optimal parameters of the rotary oven in complex industrial baking scenarios, resulting in low control accuracy of the rotary oven. Therefore, in order to ensure the quality of the finished product, it is necessary to focus on the key parameters that have a significant impact on the quality of the finished product during the control process of the rotary oven. Summary of the Invention
[0005] To address the technical problem of low control precision in rotary kilns, this application provides a data analysis-based method for optimizing and recommending rotary kiln parameters. This method can focus on key parameters that significantly affect the quality of the finished product during the iteration process, ensuring the accuracy of optimal process parameter recommendations in complex industrial baking scenarios.
[0006] In a first aspect, this application provides a data analysis-based method for optimizing and recommending parameters for a rotary furnace. The method includes: obtaining the maximum permissible variation of each control parameter of the rotary furnace; generating an initial population containing multiple parameter vectors; and executing an iterative process of a differential evolution algorithm, including: for any target vector, selecting parent vectors from the population to calculate a differential vector; limiting the scaling degree of the differential vector according to the maximum permissible variation to obtain an effective mutation scaling factor for each control parameter, and generating a mutation vector; obtaining the cumulative sensitivity of each control parameter, whereby the cumulative sensitivity characterizes the degree of influence of the control parameter on the finished product quality score; calculating the adaptive crossover probability of each control parameter based on the cumulative sensitivity; performing a crossover operation between the mutation vector and the target vector to generate an experimental vector; inputting the experimental vector into a pre-constructed finished product score prediction model to obtain the finished product score; updating the cumulative sensitivity of each control parameter based on the improvement of the finished product score relative to the target vector, and updating the population based on the finished product score; iterating multiple times until a preset termination condition is reached, and outputting the optimal recommended parameters to control the rotary furnace.
[0007] By obtaining the maximum allowable variation of each control parameter of the rotary furnace to limit the scaling of the difference vector, and calculating the adaptive crossover probability based on the cumulative sensitivity, the limited search computing power is concentrated on key parameters while preventing the generation of invalid instructions that exceed the equipment's capabilities. This improves the accuracy and convergence speed of parameter optimization and determines the values of each control parameter in each baking stage.
[0008] Preferably, the process of constructing the finished product score prediction model includes: acquiring historical data of the rotary furnace, the historical data including multiple sets of historical process parameters and corresponding finished product quality scores; preprocessing the historical process parameters as input features, using the quality scores as output labels, training the finished product score prediction model, establishing a mapping relationship between input features and output labels, and obtaining the finished product score prediction model.
[0009] Preferably, the finished product scoring prediction model is a BP neural network or a support vector regression model.
[0010] Preferably, generating an initial population containing multiple parameter vectors includes: obtaining the environmental correction coefficient for the current environment; obtaining the reference parameter vector and range vector of the rotary furnace, wherein the reference parameter vector includes the reference values of each control parameter in each baking stage, and the range vector includes the allowable adjustment range of each control parameter in each baking stage; calculating the product of the random gain and the environmental correction coefficient, multiplying the minimum value of the product and 1 by the range vector to obtain the bias vector, and using the sum of the bias vector and the reference parameter vector as the parameter vector; obtaining the parameter vectors corresponding to multiple random gains to obtain the initial population.
[0011] The environmental correction coefficient is obtained, and the product of the random gain, range vector and environmental correction coefficient is used as the bias vector to generate the initial population. Environmental conditions directly affect the initial thermal state and energy loss of the rotary furnace. The bias vector can be used to adjust the search starting point to a region that is closer to the optimal solution under the current operating conditions.
[0012] Preferably, the calculation process of the environmental correction coefficient is as follows: collecting the real-time room temperature and input voltage of the environment in which the rotary furnace is located; calculating the temperature difference between the real-time room temperature and the preset reference room temperature, determining the temperature correction amount based on the temperature difference, wherein the temperature correction amount is positively correlated with the temperature difference; calculating the voltage difference between the input voltage and the preset reference voltage, determining the voltage correction amount based on the voltage difference, wherein the voltage correction amount is positively correlated with the voltage difference; and determining the environmental correction coefficient based on the preset reference coefficient, the temperature correction amount, and the voltage correction amount.
[0013] By quantifying external disturbances such as real-time room temperature and input voltage, dynamic compensation for the thermal load capacity of the equipment is achieved, enhancing the environmental adaptability of the control system.
[0014] Preferably, obtaining the effective variation scaling factor for each control parameter includes: for any control parameter, calculating the product of the basic scaling factor and the variation amplitude of the control parameter in the difference vector to obtain the theoretical variation; calculating the cutoff coefficient of the control parameter, which is the ratio of the maximum allowable variation to the theoretical variation; using the minimum value between 1 and the cutoff coefficient as a correction coefficient; and using the product of the basic scaling factor and the correction coefficient as the effective variation scaling factor for the control parameter.
[0015] The truncation factor can be used to forcibly limit the range of variation to a range achievable by the device.
[0016] Preferably, calculating the adaptive crossover probability of each control parameter based on the cumulative sensitivity includes: obtaining the maximum and minimum values of the cumulative sensitivity of each control parameter in the current population, and using the difference between the maximum and minimum values as the sensitivity range; using the difference between the cumulative sensitivity of any control parameter and the minimum value as the relative sensitivity; and performing linear interpolation on the preset minimum crossover probability and maximum crossover probability based on the ratio of the relative sensitivity to the sensitivity range to obtain the adaptive crossover probability of the control parameter.
[0017] Different control parameters have different weights influencing the quality of the finished product. The greater the cumulative sensitivity, the greater the exploration probability required for the corresponding control parameter. This achieves a non-uniform search strategy, avoids wasting computing power on non-critical parameters with low cumulative sensitivity, and improves the efficiency of finding the global optimal solution.
[0018] Preferably, updating the cumulative sensitivity of each control parameter based on the improvement of the finished product score relative to the target vector includes: selecting test vectors whose finished product scores are higher than the target vector as the winning set; in response to the non-empty winning set, for any control parameter, calculating the parameter change magnitude and score improvement of each test vector relative to the target vector; calculating the weight coefficient of each test vector, and weighting and summing the score improvement based on the weight coefficient, wherein the weight coefficient is the ratio of the parameter change magnitude to the value range of the corresponding control parameter; using the ratio of the weighted summation result to the sum of the score improvement values as the sensitivity contribution value; and using a forgetting factor to weight and sum the cumulative sensitivity of the previous generation with the sensitivity contribution value to obtain the updated cumulative sensitivity.
[0019] By extracting key factors that lead to quality improvement from successful evolutionary history, we have achieved self-learning and dynamic updating of accumulated sensitivity, making subsequent iterations more targeted.
[0020] Preferably, updating the cumulative sensitivity of each control parameter further includes: in response to the winning set being empty, using the cumulative sensitivity of the previous generation as the updated cumulative sensitivity.
[0021] Preferably, the control parameters include the temperature, humidity, and wind speed of the rotary furnace.
[0022] The technical solution of this application has the following beneficial technical effects: By introducing the maximum allowable variation constraint of the equipment and the initialization strategy based on the environmental correction coefficient into the differential evolution algorithm, and combining the cumulative sensitivity of the control parameters to perform adaptive crossover operations, the search starting point can be automatically adjusted according to environmental changes while ensuring that the process parameters meet the equipment response limits. During the iteration process, the algorithm focuses on key parameters that have a significant impact on the quality of the finished product, ensuring the accuracy of the optimal process parameter recommendation in complex industrial baking scenarios, and improving the convergence efficiency of the iteration process. Attached Figure Description
[0023] Figure 1 This is a three-dimensional structural diagram of a rotary furnace according to an embodiment of this application.
[0024] Figure 2 This is a flowchart of a data analysis-based rotary furnace parameter optimization and recommendation method according to an embodiment of this application.
[0025] Figure 3 This is a graph showing the change in the quality score of the finished product with the number of iterations according to an embodiment of this application.
[0026] Reference numerals: 10. Furnace body; 11. Insulated furnace door; 12. Intelligent control box; 13. Rotary drive mechanism; 14. Rotary hanger. Detailed Implementation
[0027] According to the first aspect of this application, this application provides a data analysis-based method for optimizing and recommending rotary oven parameters, applicable to food baking and processing scenarios, and enabling adaptive control of baking bread, cakes and other pastry products using a rotary oven. Figure 1 This is a three-dimensional structural diagram of a rotary oven according to an embodiment of this application, including an oven body 10, an insulated oven door 11, an intelligent control box 12, a rotary drive mechanism 13, and a rotary rack 14. The insulated oven door 11 is movably connected to the oven body 10. By opening the insulated oven door 11, pastries are placed on the rotary rack 14. The intelligent control box 12 is located on the oven body 10 and on one side of the insulated oven door 11. The intelligent control box 12 is equipped with a programmable logic controller, a frequency converter, and a temperature control module, which are used to receive sensor data and send commands to the motor, heating tube, and fan. The rotary drive mechanism 13 is located on the top of the oven body 10 and is connected to the rotary rack 14 inside the oven via a drive shaft or chain. It drives the rotary rack 14 to rotate 360 degrees at a uniform speed, ensuring that each tray of pastries placed on the rotary rack 14 is alternately exposed to the heat flow blown from the hot air nozzles, ensuring that the product is heated evenly.
[0028] Figure 2 This is a flowchart of a rotary furnace parameter optimization and recommendation method based on data analysis, according to an embodiment of this application. For example... Figure 2 As shown, the data analysis-based rotary furnace parameter optimization recommendation method includes steps S101 to S104, which are described in detail below.
[0029] S101, obtain the maximum allowable variation of each control parameter of the rotary furnace.
[0030] In one embodiment, control parameters refer to the adjustable process variables of the rotary furnace during operation, specifically including the temperature, humidity, and air velocity of the rotary furnace. The maximum permissible variation refers to the maximum adjustment range of the control parameters that the rotary furnace can achieve per unit time.
[0031] Understandably, the maximum permissible variation of each control parameter is related to the model of the rotary furnace and can be preset by technicians.
[0032] Furthermore, in order to accurately obtain the finished product score during the iterative process of differential evolution, it is necessary to construct a finished product score prediction model. The construction process of the finished product score prediction model includes: acquiring historical data of the rotary furnace, which includes multiple sets of historical process parameters and corresponding finished product quality scores; preprocessing the historical process parameters as input features, using the quality scores as output labels, training the finished product score prediction model, establishing the mapping relationship between input features and output labels, and obtaining the finished product score prediction model.
[0033] For example, a set of historical process parameters includes historical values of each control parameter at each baking stage, and the quality score is the quality of the finished product obtained after baking in a rotary oven under these historical process parameters. The pretreatment includes standardization to eliminate dimensional differences between the control parameters.
[0034] Preferably, the finished product score prediction model is a backpropagation (BP) neural network, which utilizes its nonlinear mapping capability to fit the complex relationship between baking processes and quality. In another embodiment, the finished product score prediction model can also be a support vector regression model.
[0035] Thus, by constructing a finished product score prediction model, a reliable evaluation method is provided for subsequent parameter optimization, enabling accurate and rapid acquisition of finished product scores.
[0036] S102 generates an initial population containing multiple parameter vectors.
[0037] In one embodiment, the initial population refers to the set of candidate baking schemes generated by the differential evolution algorithm before the start of iteration. Each parameter vector corresponds to a baking scheme, and the parameter vector includes the parameter values of each control parameter at each baking stage. The environmental correction coefficient is a coefficient used to compensate for the impact of ambient temperature and humidity on equipment performance.
[0038] Before generating an initial population containing multiple parameter vectors, it is necessary to obtain the environmental correction coefficient for the current environment. The calculation process of the environmental correction coefficient is as follows: collect the real-time room temperature and input voltage of the rotary furnace environment; calculate the temperature difference between the real-time room temperature and the preset reference room temperature, and determine the temperature correction amount based on the temperature difference, wherein the temperature correction amount is positively correlated with the temperature difference; calculate the voltage difference between the input voltage and the preset reference voltage, and determine the voltage correction amount based on the voltage difference, wherein the voltage correction amount is positively correlated with the voltage difference; determine the environmental correction coefficient based on the preset reference coefficient, the temperature correction amount, and the voltage correction amount.
[0039] Environmental correction factor Satisfying the relation:
[0040] in, and These are the first preset coefficient and the second preset coefficient, respectively. Real-time room temperature, Input voltage, Based on room temperature, As the reference voltage, This serves as a baseline coefficient. When the ambient temperature is too low or the voltage is insufficient, the correction amount is increased, thus increasing the environmental correction coefficient. This allows for upward adjustment of parameter setting values during subsequent parameter determination to compensate for the issue. The baseline coefficient... The value of is 1.
[0041] It should be noted that, in order to avoid inconsistencies in the dimensions of temperature difference, pressure difference, and various control parameters, the data is standardized after being collected.
[0042] Further, generating an initial population containing multiple parameter vectors includes: obtaining a reference parameter vector and a range vector for the rotary oven, wherein the reference parameter vector includes the reference values of each control parameter in each baking stage, and the range vector includes the allowable adjustment range of each control parameter in each baking stage; calculating the product of the random gain and the environmental correction coefficient, multiplying the minimum value of the product and 1 by the range vector to obtain the bias vector, and using the sum of the bias vector and the reference parameter vector as the parameter vector; obtaining parameter vectors corresponding to multiple random gains to obtain the initial population.
[0043] Wherein, parameter vector Satisfying the relation:
[0044] in, As the baseline parameter vector, For random gain, The range vector is used; the random gain is a random number between 0 and 1. The initial search center is biased using an environmental correction coefficient. In an ideal environment, i.e., the real-time room temperature equals the reference room temperature and the input voltage equals the reference voltage, the environmental correction coefficient... At this point, all control parameters in the initial population are generated around the baseline value without any offset; when environmental conditions are poor, i.e., when the ambient temperature or input voltage is low, the environmental correction coefficient... , The compensation term is positive, causing the values of all control parameters in the initial population to shift upwards as a whole; when environmental conditions are better than the baseline environment, i.e., the real-time room temperature and input voltage both reach the baseline values, the environmental correction coefficient... , In the initial population, the values of all control parameters are shifted downwards to avoid overcompensation. For example, when the environmental correction coefficient... When, the bias vector is .
[0045] It should be noted that, to avoid excessive compensation, the maximum and minimum values of the bias vector also need to be set. The maximum value of the bias vector is... The maximum value of the bias vector is When the bias vector exceeds the maximum and minimum values, it is directly truncated.
[0046] For example, the baking stage includes an expansion period, a setting period, and a coloring period. Each baking stage corresponds to three control parameters: temperature, humidity, and wind speed. Therefore, the reference parameter vector, range vector, and parameter vector are all 9-row, 1-column column vectors. Each control parameter in each baking stage corresponds to a preset value range. The center point of the preset value range is used as the reference value for each control parameter in each baking stage, thus obtaining the reference parameter vector. The distance between the reference value and the endpoint of the preset value range is used as the allowable adjustment range for each control parameter in each baking stage, thus obtaining the range vector.
[0047] In one embodiment, the first preset coefficient Second preset coefficient The value is determined through historical data fitting and calibration, specifically including the following steps: collecting multiple sets of data from the rotary furnace during its historical operation, each set of data including real-time room temperature. Input voltage And the parameter vector consisting of each control parameter at each baking stage used to ensure the finished product quality is qualified under these environmental conditions. Calculate the first [parameter] based on the reference parameter vector and the range vector. Target values of environmental correction factors corresponding to the set of data :
[0048] in, For the Euclidean norm; As the dependent variable, relative temperature deviation and voltage relative deviation As the independent variable, a linear regression model is established: The regression model was fitted using the least squares method to obtain... and The estimated value.
[0049] In this way, by using the environmental correction coefficient to adjust the bias of the initial population, adaptive initialization of the initial population is achieved, ensuring that the initial population is in a better starting position in the early stage of the search, which significantly shortens the time to find the optimal solution.
[0050] S103 executes the iterative process of the differential evolution algorithm.
[0051] In one embodiment, during the iterative process of the differential evolution algorithm, for any target vector, parent vectors are selected from the population to calculate the differential vector. The scaling degree of the differential vector is limited according to the maximum allowable change to obtain an effective mutation scaling factor and generate a mutation vector.
[0052] The target vector is any parameter vector in the current population, with the target vector as the parameter vector. For example, from the parameter vector in the current population In addition, two parent vectors are randomly selected and denoted as the first parent vector. Second parent vector Then the difference vector is .
[0053] In traditional differential evolution algorithms, the scaling of the difference vector is determined by a random step size. However, a random step size may cause excessively large parameter mutations, exceeding the maximum allowable change of each control parameter. Therefore, it is necessary to limit the scaling of the difference vector.
[0054] Specifically, the scaling factor of the difference vector is limited based on the maximum permissible change, and the effective variation scaling factor of each control parameter is obtained by: for any control parameter, calculating the product of the basic scaling factor and the change amplitude of the control parameter in the difference vector to obtain the theoretical variation; calculating the cutoff coefficient of the control parameter, which is the ratio of the maximum permissible change to the theoretical variation; using the minimum value between 1 and the cutoff coefficient as the correction coefficient; and using the product of the basic scaling factor and the correction coefficient as the effective variation scaling factor of the control parameter.
[0055] Control parameters Effective variation scaling factor Satisfying the relation:
[0056] in, Based on the scaling factor, For control parameters The maximum allowable variation, Control parameters in the difference vector The range of change, A preset positive number is used to prevent the denominator from being 0; its value can be set to 1. When the theoretical variation exceeds the physical boundary, the correction coefficient is less than 1, forcibly reducing the scaling factor and adjusting the control parameter. The range of variation is limited to the achievable range of the device.
[0057] After obtaining the effective variation scaling factors for each control parameter, a variation vector is generated. The variation vector... Satisfying the relation:
[0058] in, For the target vector, and These are the first and second parent vectors selected from the population, respectively. This is the effective variation scaling factor vector, which includes the effective variation scaling factors for each control parameter.
[0059] Subsequently, the cumulative sensitivity of each control parameter is obtained; based on the cumulative sensitivity, the adaptive crossover probability of each control parameter is calculated, and the mutation vector is crossed with the target vector to generate the test vector.
[0060] Specifically, calculating the adaptive crossover probability of each control parameter based on the cumulative sensitivity includes: obtaining the maximum and minimum values of the cumulative sensitivity of each control parameter in the current population, and using the difference between the maximum and minimum values as the sensitivity range; using the difference between the cumulative sensitivity of any control parameter and the minimum value as the relative sensitivity; and performing linear interpolation on the preset minimum and maximum crossover probabilities based on the ratio of the relative sensitivity to the sensitivity range to obtain the adaptive crossover probability of the control parameter.
[0061] Control parameters Adaptive crossover probability Satisfying the relation:
[0062] in, For control parameters Cumulative sensitivity, and These represent the minimum and maximum cumulative sensitivity values for each control parameter. and These are the pre-set minimum crossover probability and maximum crossover probability, respectively; This is a preset positive number used to prevent the denominator from being 0; its value can be set to 1.
[0063] Understandably, the higher the cumulative sensitivity of a control parameter, the greater its influence on the finished product quality score, and the greater its corresponding crossover probability. This means that the control parameter is more likely to be updated in the next generation, thus prioritizing the updating of control parameters that have a greater impact on the finished product quality score and improving the convergence speed.
[0064] After obtaining the adaptive crossover probability of each control parameter, the process of generating the test vector is as follows: For any control parameter, generate a random number between 0 and 1. If the random number is less than or equal to the adaptive crossover probability corresponding to the control parameter, or if the control parameter is randomly selected as a mandatory parameter, then the value of the control parameter in the test vector comes from the mutation vector; otherwise, the value of the control parameter in the test vector retains the value of the target vector. In this way, each parameter vector in the current population will generate a corresponding test vector.
[0065] Next, the experimental vector is input into the pre-built product score prediction model to obtain the product score. The cumulative sensitivity of each control parameter is updated based on the improvement of the product score relative to the target vector, and the population is updated based on the product score.
[0066] Specifically, updating the cumulative sensitivity of each control parameter based on the improvement of the finished product score relative to the target vector includes: selecting test vectors whose finished product scores are higher than the target vector as the winning set; in response to the non-empty winning set, for any control parameter, calculating the parameter change magnitude and score improvement of each test vector relative to the target vector; calculating the weight coefficient of each test vector, and weighting and summing the score improvement based on the weight coefficient, wherein the weight coefficient is the ratio of the parameter change magnitude to the corresponding control parameter value range; using the ratio of the weighted summation result to the sum of all score improvement values as the sensitivity contribution value; and using a forgetting factor to weight and sum the previous generation's cumulative sensitivity with the sensitivity contribution value to obtain the updated cumulative sensitivity.
[0067] Control parameters Updated cumulative sensitivity Satisfying the relation:
[0068] in, This is the forgetting factor, which can be set to 0.3. For the winning group; For the test vector The increase in ratings For the test vector Control parameters The value, Control parameters in the target vector The value, For control parameters The range of parameter changes, For control parameters The allowable adjustment range; For control parameters The cumulative sensitivity of the previous generation. If the change in the control parameter is large and the corresponding score improvement is also significant, the sensitivity of that control parameter will increase. Furthermore, in response to an empty winner set, the cumulative sensitivity of the previous generation will be used as the updated cumulative sensitivity.
[0069] In this way, by performing a differential evolution iteration process, truncating the degree of mutation by combining the maximum allowable change, and using sensitivity to guide non-uniform crossover, the limited computing power can be focused on key parameters that have a significant impact on quality, while ensuring that the process parameters meet the equipment response capabilities, thus greatly improving optimization efficiency and convergence accuracy.
[0070] S104 iterates multiple times until the preset termination condition is reached, and outputs the optimal recommended parameters to control the rotary furnace.
[0071] In one embodiment, the preset termination condition includes reaching a preset maximum number of iterations or the highest finished product score in the current population reaching a preset threshold; the maximum number of iterations is 50; the preset threshold is the minimum score corresponding to when the finished product is qualified, which is set to 9.6 in this embodiment.
[0072] If the preset termination condition is not met, the updated population and cumulative sensitivity are used as input for the next generation, and step S103 is repeated; through continuous iteration, the survival of the fittest mechanism causes the parameter vectors in the population to gradually cluster towards the high-score region; please refer to Figure 3 The graph shows the change of the finished product quality score with the number of iterations according to the embodiments of this application. As can be seen from the graph, the finished product quality score gradually increases with the continuous increase of the number of iterations, and the iteration stops when the preset threshold is reached.
[0073] If the preset termination condition is met, the iteration stops, and the experimental vector with the highest product score is selected from the current population and output as the optimal recommended parameter.
[0074] It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the scope of protection of this application. Therefore, the scope of protection of this patent application shall be determined by the appended claims.
Claims
1. A method for optimizing and recommending parameters of a rotary furnace based on data analysis, characterized in that, The recommended method includes: obtaining the maximum permissible variation of each control parameter of the rotary furnace; Generate an initial population containing multiple parameter vectors; The iterative process of executing the differential evolution algorithm includes: for any target vector, selecting parent vectors from the population to calculate the difference vector; limiting the scaling degree of the difference vector according to the maximum allowable change to obtain the effective mutation scaling factor of each control parameter and generating a mutation vector; obtaining the cumulative sensitivity of each control parameter, wherein the cumulative sensitivity characterizes the degree of influence of the control parameter on the finished product quality score; calculating the adaptive crossover probability of each control parameter based on the cumulative sensitivity; performing a crossover operation between the mutation vector and the target vector to generate an experimental vector; inputting the experimental vector into a pre-constructed finished product score prediction model to obtain the finished product score; updating the cumulative sensitivity of each control parameter according to the improvement of the finished product score relative to the target vector, and updating the population according to the finished product score; The process iterates multiple times until a preset termination condition is met, then outputs the optimal recommended parameters to control the rotary furnace.
2. The method for optimizing and recommending rotary furnace parameters based on data analysis according to claim 1, characterized in that, The process of constructing the finished product scoring prediction model includes: Historical data of the rotary furnace is acquired, including multiple sets of historical process parameters and corresponding quality scores of finished products; the historical process parameters are preprocessed as input features, and the quality scores are used as output labels to train a finished product score prediction model, establishing a mapping relationship between input features and output labels to obtain the finished product score prediction model.
3. The method for optimizing and recommending rotary furnace parameters based on data analysis according to claim 2, characterized in that, The finished product scoring prediction model is a BP neural network or a support vector regression model.
4. The method for optimizing and recommending rotary furnace parameters based on data analysis according to claim 1, characterized in that, The generation of the initial population containing multiple parameter vectors includes: Obtain the environmental correction factor for the current environment; Obtain the reference parameter vector and range vector of the rotary oven. The reference parameter vector includes the reference values of each control parameter in each baking stage, and the range vector includes the allowable adjustment range of each control parameter in each baking stage. Calculate the product of random gain and environmental correction coefficient. Multiply the minimum value of the product and 1 by the range vector to obtain the bias vector. Use the sum of the bias vector and the reference parameter vector as the parameter vector. Obtain the parameter vectors corresponding to multiple random gains to obtain the initial population.
5. The method for optimizing and recommending rotary furnace parameters based on data analysis according to claim 4, characterized in that, The calculation process for the environmental correction factor is as follows: Collect real-time room temperature and input voltage of the environment in which the rotary furnace is located; Calculate the temperature difference between the real-time room temperature and the preset reference room temperature, and determine the temperature correction amount based on the temperature difference. The temperature correction amount is positively correlated with the temperature difference. Calculate the voltage difference between the input voltage and the preset reference voltage, and determine the voltage correction amount based on the voltage difference. The voltage correction amount is positively correlated with the voltage difference. The environmental correction factor is determined based on the preset reference factor, temperature correction factor, and voltage correction factor.
6. The method for optimizing and recommending rotary furnace parameters based on data analysis according to claim 1, characterized in that, The effective variation scaling factor for each control parameter is obtained by: for any control parameter, calculating the product of the basic scaling factor and the variation amplitude of the control parameter in the difference vector to obtain the theoretical variation; calculating the cutoff coefficient of the control parameter, which is the ratio of the maximum allowable variation to the theoretical variation; using the minimum value between 1 and the cutoff coefficient as the correction coefficient; and using the product of the basic scaling factor and the correction coefficient as the effective variation scaling factor of the control parameter.
7. The method for optimizing and recommending rotary furnace parameters based on data analysis according to claim 1, characterized in that, The adaptive crossover probability of each control parameter is calculated based on the cumulative sensitivity, including: Obtain the maximum and minimum cumulative sensitivity of each control parameter in the current population, and use the difference between the maximum and minimum values as the sensitivity range; The difference between the cumulative sensitivity of any control parameter and the minimum value is used as the relative sensitivity. The adaptive crossover probability of the control parameter is obtained by linearly interpolating the preset minimum crossover probability and maximum crossover probability based on the ratio of relative sensitivity to sensitivity range.
8. The method for optimizing and recommending rotary furnace parameters based on data analysis according to claim 1, characterized in that, The method of updating the cumulative sensitivity of each control parameter based on the improvement of the finished product score relative to the target vector includes: Select the experimental vectors whose finished product scores are higher than the target vector as the winning set; In response to the non-empty set of winners, for any control parameter, the parameter change magnitude and score improvement of each experimental vector relative to the target vector are calculated; the weight coefficient of each experimental vector is calculated, and the score improvement is weighted and summed according to the weight coefficient, wherein the weight coefficient is the ratio of the parameter change magnitude to the value range of the corresponding control parameter; the ratio of the weighted summation result to the sum of the score improvement is used as the sensitivity contribution value. The cumulative sensitivity of the previous generation is weighted and summed with the sensitivity contribution value using a forgetting factor to obtain the updated cumulative sensitivity.
9. The method for optimizing and recommending rotary furnace parameters based on data analysis according to claim 8, characterized in that, Updating the cumulative sensitivity of each control parameter also includes: in response to the winner set being empty, using the cumulative sensitivity of the previous generation as the updated cumulative sensitivity.
10. The method for optimizing and recommending rotary furnace parameters based on data analysis according to claim 1, characterized in that, The control parameters include the temperature, humidity, and wind speed of the rotary furnace.