Method for controlling the screening performance of an air flow microcellular screen

Abstract

The present application relates to a kind of airflow micro-hole screen screening performance control method, belong to screening efficiency control technical field, solve the calculation method of airflow micro-hole screen screening efficiency in prior art, high cost, poor efficiency, low accuracy and cannot be applied to the problem of complex working condition.If the first predicted screening efficiency of the airflow micro-hole screen to be measured is not in line with the expected target, the initial design parameters are optimized using genetic algorithm;During the operation of the airflow micro-hole screen to be measured, for each time, the following operations are performed: if the difference between the second predicted screening efficiency and the actual screening efficiency is greater than the preset threshold, the control parameters of the feed speed controller and the vibration frequency controller are optimized based on the second predicted screening efficiency and the actual screening efficiency, so that the airflow micro-hole screen to be measured can achieve the second predicted screening efficiency.A kind of airflow micro-hole screen screening performance control method with high precision, high efficiency, low cost and wide application range is realized.

Description

Technical Field

[0001] This invention relates to the field of screening efficiency control technology, and in particular to a method for controlling the screening performance of an airflow microporous sieve. Background Technology

[0002] Airflow microporous sieve is a device that achieves fine classification of materials based on the principle of airflow dynamics. It abandons the traditional gravity potential energy operation principle and uses the kinetic energy of high-speed airflow to fully diffuse powder particles and spray them onto the sieve with sufficient kinetic energy, thereby achieving rapid classification. It is especially suitable for screening ultrafine powders, lightweight or easily agglomerated materials.

[0003] In existing technologies, the main methods for studying the sieving efficiency of airflow microporous sieves include experimental testing, numerical simulation, and theoretical analysis. Experimental testing involves operating actual equipment, measuring the particle size distribution of materials before and after sieving, and calculating the sieving efficiency. However, this method suffers from high cost, long cycle time, and poor repeatability (significantly affected by material characteristics such as humidity and static electricity). Numerical simulation utilizes computational fluid dynamics (CFD) and discrete element method (DEM) to simulate the airflow sieving process, analyzing particle trajectory, airflow distribution, and sieving efficiency. However, this method suffers from simplified models, difficulty in fully replicating actual working conditions (such as particle shape and electrostatic effects), and high computational complexity. Theoretical analysis involves establishing mathematical models based on sieving dynamics and fluid dynamics theories to predict sieving efficiency. However, this method suffers from limited accuracy, difficulty in comprehensively considering complex working conditions (such as particle agglomeration and screen clogging), and a narrow applicability (the model is usually specific to a particular material or equipment).

[0004] Therefore, there is a need to provide a method for controlling the screening performance of airflow microporous sieves that is highly accurate, efficient, low-cost, and widely applicable. Summary of the Invention

[0005] Based on the above analysis, the present invention aims to provide a method for controlling the screening performance of an airflow microporous sieve, in order to solve the problems of high cost, poor efficiency, low accuracy, and inapplicability to complex working conditions in existing methods for calculating the screening efficiency of airflow microporous sieves.

[0006] This invention provides a method for controlling the sieving performance of an airflow microporous sieve, comprising:

[0007] Obtain the initial design parameters of the airflow microporous sieve to be tested, and input the initial design parameters into the target screening performance prediction model to obtain the first predicted screening efficiency corresponding to the initial design parameters;

[0008] If the first predicted screening efficiency meets the expected target, the initial design parameters are used as the operating parameters of the airflow microporous sieve to be tested. If it does not meet the expected target, the initial design parameters are optimized using a genetic algorithm, and the optimized design parameters are used as the operating parameters of the airflow microporous sieve to be tested.

[0009] During the operation of the microporous airflow sieve under test, the following operations are performed at each moment: the feed rate and vibration frequency in the operating parameters at the current moment are obtained; the feed rate, the vibration frequency, and other parameters in the operating parameters are input into the target screening performance prediction model to obtain the second predicted screening efficiency corresponding to that moment; and the actual screening efficiency at that moment is calculated based on the current actual screening result. If the difference between the second predicted screening efficiency and the actual screening efficiency is greater than a preset threshold, the control parameters of the feed rate controller and the vibration frequency controller are optimized based on the second predicted screening efficiency and the actual screening efficiency so that the microporous airflow sieve under test can achieve the second predicted screening efficiency.

[0010] Based on further improvements to the above method, the target screening performance prediction model is trained on a historical operating parameter sample dataset. Each data point in the historical operating parameter sample dataset includes particle size distribution ratio, moisture content, screen aperture size, screen aperture shape, screen surface inclination angle, airflow velocity, vibration frequency and amplitude, feed rate, toothed blade spacing, toothed blade length, tail blade spacing, tail blade length, and screening efficiency.

[0011] Based on a further improvement of the above method, the optimization of the control parameters of the feed speed controller and the vibration frequency controller based on the second predicted screening efficiency and the actual screening efficiency includes: if the difference is greater than a preset critical value, the vibration frequency is adjusted based on the improved Grey Wolf algorithm while the feed speed remains unchanged; if the difference is less than the preset critical value, the feed speed is adjusted based on the improved Grey Wolf algorithm while the vibration frequency remains unchanged; wherein, the preset critical value is greater than a preset threshold.

[0012] Based on a further improvement of the above method, the improved gray wolf algorithm includes:

[0013] A1: Initialize the wolf pack, set the population size and the first preset iteration count, and define the size of each individual in the population {K}. p K i K d} represents a possible solution, and initial values ​​are set for the convergence factor a, coefficient vector A, and coefficient vector C, where K p K i K d These are the control parameters for the feed speed controller or the vibration frequency controller.

[0014] A2: Calculate the fitness function value for each wolf, where the fitness function is:

[0015] Fitness = (STEF(t) - SSEF(t)) 2 ,

[0016] Where STEF(t) is the third predicted sieving efficiency at time t, SSEF(t) is the second predicted sieving efficiency at time t, and t is the operating time of the microporous sieve to be tested.

[0017] The third predicted screening efficiency refers to the difference between the control parameter corresponding to the wolf and the second predicted screening efficiency and the actual screening efficiency at time t, which is input to the feed rate controller to obtain the feed rate or input to the vibration frequency controller to obtain the vibration frequency. Then, the feed rate or vibration frequency and other parameters in the operating parameters at time t are input to the target screening performance prediction model.

[0018] A3: Sort the wolf pack according to their fitness values ​​and select the top three as the leader wolves;

[0019] A4: In each iteration, the positions of the non-leader wolves are updated using the leader wolf's position, the convergence factor, and the coefficient vectors A and C; whereby the convergence factor a is obtained as follows:

[0020]

[0021] Where a1 and a2 are constants, σ ​​is an adjustment factor, max(Fitness) is the maximum value of the fitness function, and std(Fitness) is the standard deviation of the fitness function values ​​of all wolves in this iteration; the coefficient vector A is obtained as follows:

[0022] A=2a*r1-a+γ*(max(Fitness)-avg(Fitness)),

[0023] Where r1 is a randomly generated vector in the range [0, 1], γ is the learning rate, and avg(Fitness) is the average value of the fitness function; the coefficient vector C is obtained as follows:

[0024]

[0025] Where d is the current iteration number, D is the first preset iteration number, and β is a constant;

[0026] A5: Calculate the fitness function value of the new position and compare it with the original position. If the new position is better than the original position, update the position of the non-leader wolf; otherwise, retain the position of the non-leader wolf.

[0027] A6: Repeat A2-A5 until the first preset number of iterations is reached, and set α_wolf at this time as the PID control parameter of the microporous sieve to be tested.

[0028] Based on the further improvement of the above method, after obtaining the historical operating parameter sample dataset, it is also necessary to perform data cleaning operations on the historical operating parameter sample dataset. The data cleaning operations include: removing outliers using the dynamic threshold method, supplementing missing values ​​using a small BPNN auxiliary model, or normalizing the data using a sliding window.

[0029] Based on further improvements to the above method, the initial design parameters include: particle size distribution ratio, moisture content, sieve aperture size, sieve aperture shape, sieve surface inclination angle, airflow velocity, vibration frequency and amplitude, feed velocity, toothed blade spacing, toothed blade length, tail blade spacing, and tail blade length.

[0030] The genetic algorithm optimizes the initial design parameters, including:

[0031] B1: Initialize the population, set the population size N, and each individual in the population includes: particle size distribution ratio, moisture content, sieve aperture size, sieve aperture shape, sieve surface inclination angle, airflow velocity, vibration frequency and amplitude, feed rate, toothed blade spacing, toothed blade length, tail blade spacing, and tail blade length.

[0032] The particle size distribution ratio, moisture content, sieve aperture size, sieve aperture shape, sieve surface inclination angle, airflow velocity, toothed blade spacing, toothed blade length, tail blade spacing, and tail blade length are encoded in binary code, while the vibration frequency, amplitude, and feed rate are encoded in Gray code.

[0033] B2: Calculate the fitness value of each individual based on the fitness function, wherein the fitness function is:

[0034]

[0035] y z The third screening efficiency is obtained by inputting the target screening performance prediction model for the z-th individual. Let z be the screening efficiency corresponding to the initial design parameters, where z = 1, 2, 3, ..., N;

[0036] B3: Selection operation, which selects individuals with high fitness from the current population as parents based on the tournament algorithm to generate the next generation;

[0037] B4: Interleaving operation. For binary encoded parameter groups, a single-point interleaving method is used; for Gray code encoded parameter groups, a multi-point interleaving method is used.

[0038] B5: Mutation operation. Select the 10% of individuals with the highest fitness from the population for local search. For each selected individual, generate several neighborhood solutions in its vicinity. Search for a better solution than the current solution in the neighborhood solutions. If a better solution is found, replace the current solution and put the individual after the local search back into the population.

[0039] B6: Generate a new population, merge the parent and offspring individuals, select a preset number of individuals as the new generation population based on the fitness value, determine whether the iteration stopping condition has been met, and if so, use the best individual in the current population as the optimized running parameters.

[0040] Based on a further improvement to the above method, the neighborhood solution is generated in the following manner:

[0041]

[0042] Where, x new_p For the newly generated neighborhood solution, x old For the current individual, This indicates rounding up, where q is the current iteration number and Q is the second preset iteration number.

[0043] Based on a further improvement of the above method, the target screening performance prediction model is a BP neural network model, and a multi-stage training strategy is used to train the BP neural network model; the activation function of the BP neural network model is:

[0044] f(x)=max(0,x)+αmin(0,x),

[0045] Where α is an adjustable parameter, 0 < α < 0.1, and the loss function is:

[0046] L=λ*MSE+(1-λ)*HuberLoss,

[0047] Where MSE is the mean squared error, HuberLoss is the Huber loss, and 0 < λ < 1.

[0048] Based on a further improvement of the above method, the learning rate of each iteration in the warm-up phase of the multi-stage training strategy increases by 0.0001 compared to the previous iteration.

[0049] Based on the further improvement of the above method, intermediate data of the operation process of the microporous sieve of the airflow to be tested are collected. After the screening process is completed, the target screening performance prediction model is retrained based on the intermediate data to improve the stability of the target screening performance prediction model.

[0050] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0051] 1. This invention provides a method for controlling the screening performance of an airflow microporous sieve. Based on the predicted screening efficiency and the actual screening efficiency, the output of the feed speed PID controller and the vibration frequency PID controller are controlled. This combines the rapid prediction advantage of the target screening performance prediction model with the real-time adjustment advantage of PID control. It not only achieves accurate prediction of the screening efficiency of the airflow microporous sieve by the target screening performance prediction model, but also uses PID control to adjust parameters in real time according to the prediction results, thus providing a screening performance control method with high accuracy and high efficiency.

[0052] 2. This invention provides a method for controlling the screening performance of an airflow microporous sieve. Based on an improved Grey Wolf algorithm, it optimizes the PID control parameters, not only improving the accuracy of obtaining PID control parameters but also optimizing the adaptive adjustment mechanism of the PID controller when facing complex scene changes, reducing costs and energy consumption, and improving efficiency. Furthermore, by optimizing the convergence factor and coefficient vector in the Grey Wolf algorithm, the algorithm can explore the solution space more extensively in the early stages, avoiding premature entrapment in local optima and increasing the chance of finding the global optimum. During the search process, it can more efficiently approach the optimal solution, reducing unnecessary computation and thus accelerating the convergence speed. In addition, the optimized parameters allow the algorithm to perform more stably when facing different problems, reducing dependence on initial conditions or problem characteristics and improving the algorithm's robustness. Therefore, the Grey Wolf algorithm can find the optimal solution faster, more accurately, and more stably.

[0053] 3. This invention provides a method for controlling the screening performance of an airflow microporous sieve. After calculating the screening efficiency based on the current initial design parameters, if the screening efficiency requirement is not met, a genetic algorithm is used to obtain initial design parameters that meet the screening efficiency. Therefore, this invention reduces the difficulty of adjusting the initial design parameters, eliminating the need for professional personnel to adjust them based on their expertise. The provided genetic algorithm alone can quickly obtain initial design parameters that match the current airflow microporous sieve, and this method is applicable to any complex scenario. Furthermore, by combining neighborhood solution optimization with the mutation operation of the genetic algorithm, high-quality solutions can be quickly found through efficient local search. This not only improves the algorithm's search efficiency and solution quality but also enhances the algorithm's robustness and global search capability, thus exhibiting better performance when solving complex optimization problems.

[0054] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0055] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0056] Figure 1 This is an example diagram illustrating a method for controlling the sieving performance of an airflow microporous sieve according to an embodiment of the present invention. Detailed Implementation

[0057] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0058] A specific embodiment of the present invention discloses a method for controlling the sieving performance of an airflow microporous sieve, such as... Figure 1 As shown, it includes:

[0059] S1: Obtain the initial design parameters of the microporous sieve to be tested, and input the initial design parameters into the target screening performance prediction model to obtain the first predicted screening efficiency corresponding to the initial design parameters.

[0060] The factors affecting the screening efficiency of airflow microporous sieves mainly include the following aspects: (1) material characteristics, including particle shape, particle size distribution, moisture content, friction characteristics and flowability, etc.; (2) equipment parameters, including sieve hole size and shape, sieve opening ratio, airflow velocity, equipment motion state, toothed blade parameters, tail blade parameters, etc.; (3) operating conditions, including material layer thickness, feeding uniformity, environmental factors, etc.; (4) process parameters, including airflow velocity, feeding method, screening time, etc. The design parameters finally selected by this invention based on production experience and feasibility include: particle size distribution ratio, moisture content, sieve hole size, sieve hole shape, sieve surface inclination angle, airflow velocity, vibration frequency and amplitude, feeding speed, toothed blade spacing, toothed blade length, tail blade spacing, and tail blade length.

[0061] The target screening performance prediction model is a BP neural network model, and a multi-stage training strategy is used to train the BP neural network model; the activation function of the BP neural network model is:

[0062] f(x)=max(0,x)+αmin(0,x),

[0063] Where α is an adjustable parameter used to enhance the model's adaptability to different data patterns, 0 < α < 0.1. For example, the value of α is 0.01, and the loss function is:

[0064] L=λ*MSE+(1-λ)*HuberLoss,

[0065] Where MSE is the mean squared error and Huber Loss is the Huber loss, 0 < λ < 1. MSE measures the average error between the model's predicted value and the true value, while Huber loss is more robust to outliers. For example, λ is 0.5.

[0066] The target screening performance prediction model is trained based on a historical operating parameter sample dataset. Each data point in the historical operating parameter sample dataset includes particle size distribution ratio, moisture content, screen aperture size, screen aperture shape, screen surface inclination angle, airflow velocity, vibration frequency and amplitude, feed rate, toothed blade spacing, toothed blade length, tail blade spacing, tail blade length, and screening efficiency. After obtaining the historical operating parameter sample dataset, data cleaning is required. The data cleaning operation includes one or more of the following: removing outliers using a dynamic threshold method, supplementing missing values ​​using a small BPNN auxiliary model, or normalizing data using a sliding window.

[0067] For example, the collected historical operating parameter sample data is cleaned, and outliers are identified and removed using a dynamic thresholding method based on statistical distribution and equipment physical characteristics. Missing values ​​are initially inferred based on the equipment's physical principles and operating modes, and then accurately filled in using a small BPNN-assisted model. A dynamic adaptive normalization method is employed, using a sliding window technique to set different normalization ranges for data from different time periods according to different stages and operating conditions of the equipment, thus completing data preprocessing. Specifically, high-precision pressure sensors, flow sensors, and other measuring devices are installed at key measuring points of the airflow microporous screen to detect airflow pressure, gas flow rate, and screen inclination angle. The data acquisition frequency is set according to the equipment's operating characteristics to ensure that subtle changes during equipment operation are captured and data is transmitted in real time. Outliers are identified using a dynamic thresholding method based on statistical distribution and equipment physical characteristics. The normal range of values ​​for each parameter is determined based on the equipment's historical operating data and physical principles. For example, based on the design specifications and actual operating experience of the airflow microporous screen, the normal range for the screen inclination angle is determined to be [-4°, 4°]. Then, by calculating the mean and standard deviation of the data and combining them with a dynamic threshold factor, the threshold range is adjusted in real time. Data points exceeding the threshold range are identified as outliers and marked. Finally, median filtering or neighborhood-based interpolation is used to replace outliers, ensuring data accuracy. Missing values ​​are initially inferred based on the equipment's physical principles and operating modes. If feeding speed data is missing, a preliminary estimate can be made based on feeding speeds in preceding and following time periods and the equipment's operating status (e.g., whether it's in a stable operating phase). Then, a small BPNN auxiliary model is used for precise imputation. This auxiliary model takes other relevant parameters (e.g., perturbation frequency, screen inclination angle, etc.) as input and the parameters corresponding to the missing values ​​as output. Training on a large amount of historical data enables the auxiliary model to accurately predict missing values. During training, mean squared error is used as the loss function, and the Adam optimization algorithm is used for model training until convergence. A dynamic adaptive normalization method is employed, using a sliding window technique to set different normalization ranges for data in different time periods based on different equipment operating stages and conditions. Different normalization parameters are set during the equipment startup and stable operation phases due to differences in parameter variation ranges and characteristics. Specifically, for each parameter, its maximum and minimum values ​​are calculated within a sliding window, and then the data is normalized to the interval [0, 1]. The formula is: Where x represents the original data, x min and x max These are the minimum and maximum values ​​within the sliding window, respectively, x norm This is the normalized data.

[0068] Optionally, a backpropagation (BP) neural network model is built using the TensorFlow deep learning framework in Python. The number of nodes in the input layer is determined based on the number of parameters of the airflow microporous sieve, totaling 12 nodes, corresponding to particle size distribution ratio, moisture content, sieve aperture size, sieve aperture shape, sieve surface inclination angle, airflow velocity, vibration frequency and amplitude, feed rate, toothed blade spacing, toothed blade length, tail blade spacing, and tail blade length. The hidden layers are initially set to 2 layers, each containing 10 nodes, which can be adjusted later according to model performance. The output layer is set to 1 node, using the Sigmoid function, and the output result is the screening efficiency. The weights and biases of the neural network model are adjusted by combining the Adam optimization algorithm and learning rate annealing technique until the model converges and reaches the expected prediction accuracy. During training, the loss value and accuracy of the model on the test set are continuously monitored. When the loss value converges to below 0.01 and the accuracy reaches above 97%, the model is considered to have reached the expected prediction accuracy.

[0069] Multi-stage training strategies improve training efficiency and generalization ability by adjusting the learning rate and model parameters in stages. This strategy typically includes three stages: warm-up, acceleration, and fine-tuning, each with its specific goals and methods. Specifically, the warm-up stage stabilizes model training, preventing gradient explosion or vanishing in the initial stage; the acceleration stage speeds up training and improves efficiency; and the fine-tuning stage performs final optimization to improve model performance on specific tasks. For example, in the warm-up stage, the learning rate is set to 0.01, the batch size to 16, and the number of training iterations to 200, allowing the model to initially adapt to the data features. In the acceleration stage, the learning rate is adjusted to 0.001, the batch size is increased to 32, and the number of training iterations is increased to 500, accelerating model convergence. In the fine-tuning stage, the learning rate is further reduced to 0.0001, the batch size remains at 32, and the number of training iterations is 300, finely adjusting the model to improve prediction accuracy.

[0070] For example, during the warm-up phase, the learning rate is increased by 0.0001 for each iteration compared to the previous iteration.

[0071] After obtaining the target screening performance prediction model, the initial design parameters of the airflow microporous sieve to be tested are input into the target screening performance prediction model to obtain the first predicted screening efficiency corresponding to the initial design parameters.

[0072] For example, to further improve the prediction accuracy and stability of the target screening performance prediction model, new operational data (i.e., intermediate data) of the microporous sieve under test is collected at regular intervals (e.g., 15 minutes) and added to the training dataset for retraining and optimization of the target screening performance prediction model. Specifically, new operational data is collected at fixed intervals and added to the training dataset, while the oldest data of the same quantity is deleted to ensure the timeliness and stability of the training dataset. After the screening process is completed, the target screening performance prediction model is retrained using the updated training dataset. During retraining, the model structure remains unchanged, and the same training strategy, loss function, and optimization algorithm as the initial training are used. Through retraining, the model can adapt to changes in equipment operating conditions, improving the model's prediction accuracy and stability.

[0073] S2: If the first predicted screening efficiency meets the expected target, the initial design parameters are used as the operating parameters of the airflow microporous sieve to be tested. If it does not meet the expected target, the initial design parameters are optimized using a genetic algorithm, and the optimized design parameters are used as the operating parameters of the airflow microporous sieve to be tested.

[0074] The initial design parameters include: particle size distribution ratio, moisture content, sieve aperture size, sieve aperture shape, sieve surface inclination angle, airflow velocity, vibration frequency and amplitude, feed velocity, toothed blade spacing, toothed blade length, tail blade spacing, and tail blade length.

[0075] The initial design parameters are optimized using a genetic algorithm, including:

[0076] B1: Initialize the population, set the population size N, and each individual in the population includes: particle size distribution ratio, moisture content, sieve aperture size, sieve aperture shape, sieve surface inclination angle, airflow velocity, vibration frequency and amplitude, feed velocity, toothed blade spacing, toothed blade length, tail blade spacing, and tail blade length.

[0077] The particle size distribution ratio, moisture content, sieve aperture size, sieve aperture shape, sieve surface inclination angle, airflow velocity, toothed blade spacing, toothed blade length, tail blade spacing, and tail blade length are encoded in binary code, while the vibration frequency, amplitude, and feed rate are encoded in Gray code.

[0078] B2: Calculate the fitness value of each individual based on the fitness function, wherein the fitness function is:

[0079]

[0080] y z The third screening efficiency is obtained by inputting the target screening performance prediction model for the z-th individual. Let z be the screening efficiency corresponding to the initial design parameters, where z = 1, 2, 3, ..., N.

[0081] B3: Selection operation, based on the tournament algorithm, selects individuals with high fitness from the current population as parents to generate the next generation.

[0082] B4: Crossover operation. For binary-encoded parameter sets, a single-point crossover method is used; for Gray code-encoded parameter sets, a multi-point crossover method is used. This encoding method can better adapt to the characteristics of different parameters and improve the search efficiency of the genetic algorithm.

[0083] B5: Mutation operation. Select the 10% of individuals with the highest fitness from the population for local search. For each selected individual, generate several neighborhood solutions in its vicinity. Search for a better solution among the neighborhood solutions. If a better solution is found, replace the current solution and put the individual after the local search back into the population.

[0084] The neighborhood solution is generated in the following manner:

[0085]

[0086] Where, x new_p For the newly generated neighborhood solution, x old For the current individual, This indicates rounding up, where q is the current iteration number and Q is the second preset iteration number.

[0087] B6: Generate a new population, merge the parent and offspring individuals, select a preset number of individuals as the new generation population based on the fitness value, determine whether the second preset number of iterations has been reached, and if so, use the best individual in the current population as the optimized running parameters.

[0088] Preferably, the second preset number of iterations is 1000.

[0089] In genetic operations, the crossover operation employs different methods depending on the characteristics of different parameter groups. For binary-encoded parameter groups, a single-point crossover method is used; for Gray code-encoded parameter groups, a multi-point crossover method is used. The mutation operation introduces a local search mechanism. During mutation, each parameter undergoes a small-scale random perturbation, and a local search algorithm is used to find better parameter values. After multiple rounds of genetic operations, the optimal parameter combination is obtained. This optimal parameter combination is then applied to production.

[0090] This step confirms whether the initial design parameters meet the expected goals. If they do, the initial design parameters are directly used as the operating parameters for the microporous sieve under test. If they do not meet the expectations, adjustments are made. The adjustments in this invention are based on the initial design parameters, reducing the workload of designers and improving debugging efficiency, allowing for faster and better identification of design parameters that meet the expected goals. For example, the expected goal is a screening efficiency of 96%.

[0091] Understandably, the implementation of initialization, crossover, mutation, and other operations in genetic algorithms are common knowledge in the field, and this invention does not impose any specific limitations on them.

[0092] S3: During the operation of the microporous airflow sieve under test, for each moment, the following operations are performed: the feed rate and vibration frequency in the operating parameters at the current moment are obtained; the feed rate, the vibration frequency, and other parameters in the operating parameters are input into the target screening performance prediction model to obtain the second predicted screening efficiency corresponding to that moment; and the actual screening efficiency at that moment is calculated based on the current actual screening result. If the difference between the second predicted screening efficiency and the actual screening efficiency is greater than a preset threshold, the control parameters of the feed rate controller and the vibration frequency controller are optimized based on the second predicted screening efficiency and the actual screening efficiency so that the microporous airflow sieve under test can achieve the second predicted screening efficiency.

[0093] The optimization of the control parameters of the feed speed controller and vibration frequency controller based on the second predicted screening efficiency and the actual screening efficiency includes: if the difference is greater than a preset critical value, the vibration frequency is adjusted preferentially based on the improved Grey Wolf algorithm, while the feed speed remains unchanged; if the difference is less than the preset critical value, the feed speed is adjusted preferentially based on the improved Grey Wolf algorithm, while the vibration frequency remains unchanged; wherein, the preset critical value is greater than a preset threshold. It can be understood that when the difference is greater than the preset critical value, the vibration frequency is adjusted preferentially because the vibration frequency has a more direct impact on screening efficiency; if the difference is less than the preset critical value, the feed speed is adjusted preferentially to maintain the continuity of the process.

[0094] For example, the preset threshold is 0.5 and the preset critical value is 1.2.

[0095] The improved Grey Wolf algorithm includes:

[0096] A1: Initialize the wolf pack, set the population size and the first preset iteration count, and define the size of each individual in the population {K}. p K i K d} represents a possible solution, and initial values ​​are set for the convergence factor a, coefficient vector A, and coefficient vector C, where K p K i Kd These are the control parameters for the feed rate controller or the vibration frequency controller.

[0097] A2: Calculate the fitness function value for each wolf, where the fitness function is:

[0098] Fitness = (STEF(t) - SSEF(t)) 2 ,

[0099] Where STEF(t) is the third predicted sieving efficiency at time t, SSEF(t) is the second predicted sieving efficiency at time t, and t is the operating time of the microporous sieve to be tested.

[0100] The third predicted screening efficiency refers to the difference between the control parameter corresponding to the wolf and the second predicted screening efficiency and the actual screening efficiency at time t. This difference is input to the feed rate controller to obtain the feed rate, or to the vibration frequency controller to obtain the vibration frequency. Then, the feed rate or vibration frequency, along with other parameters from the operating parameters at time t, is input into the target screening performance prediction model. The feed rate u1 is:

[0101]

[0102] The vibration frequency u2 is:

[0103]

[0104] The deviation e is:

[0105] e = SSEF(t) - ASEF(t), where ASEF(m) is the actual screening efficiency. These are the control parameters corresponding to the PID controller for the feed rate. These are the control parameters corresponding to the vibration frequency PID controller.

[0106] A3: Sort the wolf pack according to their fitness values ​​and select the top three as the leader wolves.

[0107] A4: In each iteration, the positions of the non-leader wolves are updated using the leader wolf's position, the convergence factor, and the coefficient vectors A and C; whereby the convergence factor a is obtained as follows:

[0108]

[0109] Where a1 and a2 are constants, σ ​​is an adjustment factor, max(Fitness) is the maximum value of the fitness function, and std(Fitness) is the standard deviation of the fitness function values ​​of all wolves in this iteration; the coefficient vector A is obtained as follows:

[0110] A=2a*r1-a+γ*(max(Fitness)-avg(Fitness)),

[0111] Where r1 is a randomly generated vector in the range [0, 1], γ is the learning rate, and avg(Fitness) is the average value of the fitness function. Specifically:

[0112] A w =2a*r 1w -a+γ*(max(Fitness)-avg(Fitness)),

[0113] A w It is the w-th element in the coefficient vector A, r 1w It is the w-th element in vector r1, where w = 1, 2, 3, ..., N, and N usually represents the dimension of the problem. For example, in this invention, w = 1, 2, 3; the coefficient vector C is obtained as follows:

[0114]

[0115] Where d is the current iteration number, D is the first preset iteration number, and β is a constant; it is understandable that substituting the corresponding elements in r1 into the above formula will yield the values ​​of the corresponding elements in the coefficient vector C.

[0116] Preferably, the value of constant a1 is set to 0.5, the value of constant a2 is set to 1.8, the value of adjustment factor σ is 0.8, the learning rate γ is 0.6, and the first preset number of iterations is 800.

[0117] A5: Calculate the fitness function value of the new position and compare it with the original position. If the new position is better than the original position, update the position of the non-leader wolf; otherwise, retain the position of the non-leader wolf.

[0118] A6: Repeat A2-A5 until the first preset number of iterations is reached, and set the leader α wolf at this time as the PID control parameter of the microporous sieve to be tested.

[0119] Understandably, the setting of the initial values ​​of the convergence factor a, coefficient vector A and coefficient vector C in the Grey Wolf Algorithm, as well as the implementation of operations such as position updates, are common knowledge in the field, and this invention does not impose specific limitations on them.

[0120] Compared with existing technologies, this embodiment provides a method for controlling the screening performance of an airflow microporous sieve. Based on the predicted and actual screening efficiency, it controls the outputs of a PID controller for feed speed and a PID controller for vibration frequency. This combines the rapid prediction advantage of the target screening performance prediction model with the real-time adjustment advantage of PID control. It not only achieves accurate prediction of the screening efficiency of the airflow microporous sieve by the target screening performance prediction model, but also utilizes PID control to adjust parameters in real time based on the prediction results, thus providing a highly accurate and efficient screening performance control method. Based on the improved Grey Wolf algorithm, the PID control parameters are optimized, which not only improves the accuracy of obtaining PID control parameters but also optimizes the adaptive adjustment mechanism of the PID controller in the face of complex scene changes, reducing costs and energy consumption and improving efficiency. Furthermore, by optimizing the convergence factor and coefficient vector in the Grey Wolf algorithm, the algorithm can explore the solution space more extensively in the early stages, avoiding premature entrapment in local optima and increasing the chance of finding the global optimum. During the search process, it can more efficiently approach the optimal solution, reducing unnecessary computation and thus accelerating convergence. In addition, the optimized parameters make the algorithm more stable when facing different problems, reducing dependence on initial conditions or problem characteristics and improving robustness. This allows the Grey Wolf algorithm to find the optimal solution faster, more accurately, and more stably. After calculating the screening efficiency based on the current initial design parameters, if the screening efficiency requirements are not met, a genetic algorithm is used to obtain initial design parameters that meet the screening efficiency. Therefore, this invention reduces the difficulty of adjusting the initial design parameters, eliminating the need for professional personnel to adjust them based on specialized knowledge. The provided genetic algorithm alone can quickly obtain initial design parameters that match the current airflow microporous sieve, and this method is applicable to any complex scenario. Furthermore, combining neighborhood optimization with the mutation operation of the genetic algorithm can quickly find high-quality solutions through efficient local search. This not only improves the search efficiency and solution quality of the algorithm, but also enhances the robustness and global search capability of the algorithm, thus exhibiting better performance when solving complex optimization problems.

[0121] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0122] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for controlling the sieving performance of an airflow microporous sieve, characterized in that, include: Obtain the initial design parameters of the airflow microporous sieve to be tested, and input the initial design parameters into the target screening performance prediction model to obtain the first predicted screening efficiency corresponding to the initial design parameters; If the first predicted screening efficiency meets the expected target, the initial design parameters are used as the operating parameters of the airflow microporous sieve to be tested. If it does not meet the expected target, the initial design parameters are optimized using a genetic algorithm, and the optimized design parameters are used as the operating parameters of the airflow microporous sieve to be tested. During the operation of the microporous airflow sieve under test, the following operations are performed at each moment: the feed rate and vibration frequency in the operating parameters at the current moment are obtained; the feed rate, the vibration frequency, and other parameters in the operating parameters are input into the target screening performance prediction model to obtain the second predicted screening efficiency corresponding to that moment; and the actual screening efficiency at that moment is calculated based on the current actual screening result. If the difference between the second predicted screening efficiency and the actual screening efficiency is greater than a preset threshold, the control parameters of the feed rate controller and the vibration frequency controller are optimized based on the second predicted screening efficiency and the actual screening efficiency so that the microporous airflow sieve under test can achieve the second predicted screening efficiency.

2. The method for controlling the sieving performance of an airflow microporous sieve according to claim 1, characterized in that, The target screening performance prediction model is trained based on a historical operating parameter sample dataset. Each data point in the historical operating parameter sample dataset includes particle size distribution ratio, moisture content, screen aperture size, screen aperture shape, screen surface inclination angle, airflow velocity, vibration frequency and amplitude, feed rate, toothed blade spacing, toothed blade length, tail blade spacing, tail blade length, and screening efficiency.

3. The method for controlling the sieving performance of an airflow microporous sieve according to claim 1, characterized in that, The optimization of the control parameters of the feed speed controller and the vibration frequency controller based on the second predicted screening efficiency and the actual screening efficiency includes: if the difference is greater than a preset critical value, the vibration frequency is adjusted based on the improved gray wolf algorithm while the feed speed remains unchanged; if the difference is less than the preset critical value, the feed speed is adjusted based on the improved gray wolf algorithm while the vibration frequency remains unchanged; wherein the preset critical value is greater than a preset threshold.

4. The method for controlling the sieving performance of an airflow microporous sieve according to claim 3, characterized in that, The improved Grey Wolf algorithm includes: A1: Initialize the wolf pack, set the population size and the first preset iteration count, and define the size of each individual in the population {K}. p K i K d } represents a possible solution, and initial values ​​are set for the convergence factor a, coefficient vector A, and coefficient vector C, where K p K i K d These are the control parameters for the feed speed controller or the vibration frequency controller. A2: Calculate the fitness function value for each wolf, where the fitness function is: Fitness=(STEF(t)-SSEF(t)) 2 , Where STEF(t) is the third predicted sieving efficiency at time t, SSEF(t) is the second predicted sieving efficiency at time t, and t is the operating time of the microporous sieve to be tested. The third predicted screening efficiency refers to the difference between the control parameter corresponding to the wolf and the second predicted screening efficiency and the actual screening efficiency at time t, which is input to the feed rate controller to obtain the feed rate or input to the vibration frequency controller to obtain the vibration frequency. Then, the feed rate or vibration frequency and other parameters in the operating parameters at time t are input to the target screening performance prediction model. A3: Sort the wolf pack according to their fitness values ​​and select the top three as the leader wolves; A4: In each iteration, the positions of the non-leader wolves are updated using the leader wolf's position, the convergence factor, and the coefficient vectors A and C; whereby the convergence factor a is obtained as follows: Where a1 and a2 are constants, σ ​​is an adjustment factor, max(Fitness) is the maximum value of the fitness function, and std(Fitness) is the standard deviation of the fitness function values ​​of all wolves in this iteration; the coefficient vector A is obtained as follows: A=2a*r1-a+γ*(max(Fitness)-avg(Fitness)), Where r1 is a randomly generated vector in the range [0, 1], γ is the learning rate, and avg(Fitness) is the average value of the fitness function; the coefficient vector C is obtained as follows: Where d is the current iteration number, D is the first preset iteration number, and β is a constant; A5: Calculate the fitness function value of the new position and compare it with the original position. If the new position is better than the original position, update the position of the non-leader wolf; otherwise, retain the position of the non-leader wolf. A6: Repeat A2-A5 until the first preset number of iterations is reached, and set α_wolf at this time as the PID control parameter of the microporous sieve to be tested.

5. The method for controlling the sieving performance of an airflow microporous sieve according to claim 1, characterized in that, After obtaining the historical operating parameter sample dataset, it is also necessary to perform data cleaning operations on the historical operating parameter sample dataset. The data cleaning operations include: removing outliers using dynamic thresholding, supplementing missing values ​​using a small BPNN auxiliary model, or normalizing the data using a sliding window.

6. The method for controlling the sieving performance of an airflow microporous sieve according to claim 1, characterized in that, The initial design parameters include: particle size distribution ratio, moisture content, sieve aperture size, sieve aperture shape, sieve surface inclination angle, airflow velocity, vibration frequency and amplitude, feed rate, toothed blade spacing, toothed blade length, tail blade spacing, and tail blade length. The optimization of the initial design parameters using a genetic algorithm includes: B1: Initialize the population, set the population size N, and each individual in the population includes: particle size distribution ratio, moisture content, sieve aperture size, sieve aperture shape, sieve surface inclination angle, airflow velocity, vibration frequency and amplitude, feed rate, toothed blade spacing, toothed blade length, tail blade spacing, and tail blade length. The particle size distribution ratio, moisture content, sieve aperture size, sieve aperture shape, sieve surface inclination angle, airflow velocity, toothed blade spacing, toothed blade length, tail blade spacing, and tail blade length are encoded in binary code, while the vibration frequency, amplitude, and feed rate are encoded in Gray code. B2: Calculate the fitness value of each individual based on the fitness function, wherein the fitness function is: y z The third screening efficiency is obtained by inputting the target screening performance prediction model for the z-th individual. Let z be the screening efficiency corresponding to the initial design parameters, where z = 1, 2, 3, ..., N; B3: Selection operation, which selects individuals with high fitness from the current population as parents based on the tournament algorithm to generate the next generation; B4: Interleaving operation. For binary encoded parameter groups, a single-point interleaving method is used; for Gray code encoded parameter groups, a multi-point interleaving method is used. B5: Mutation operation. Select the 10% of individuals with the highest fitness from the population for local search. For each selected individual, generate several neighborhood solutions in its vicinity. Search for a better solution than the current solution in the neighborhood solutions. If a better solution is found, replace the current solution and put the individual after the local search back into the population. B6: Generate a new population, merge the parent and offspring individuals, select a preset number of individuals as the new generation population based on the fitness value, determine whether the iteration stopping condition has been met, and if so, use the best individual in the current population as the optimized running parameters.

7. The method for controlling the sieving performance of an airflow microporous sieve according to claim 6, characterized in that, The neighborhood solution is generated in the following manner: Where, x new_p For the newly generated neighborhood solution, x old For the current individual, This indicates rounding up, where q is the current iteration number and Q is the second preset iteration number.

8. The method for controlling the sieving performance of an airflow microporous sieve according to claim 1, characterized in that, The target screening performance prediction model is a BP neural network model, and a multi-stage training strategy is used to train the BP neural network model; the activation function of the BP neural network model is: f(x)=max(0,x)+αmin(0,x), Where α is an adjustable parameter, 0 < α < 0.1, and the loss function is: L=λ*MSE+(1-λ)*HuberLoss, Where MSE is the mean squared error, HuberLoss is the Huber loss, and 0 < λ < 1.

9. The method for controlling the sieving performance of an airflow microporous sieve according to claim 8, characterized in that, In the warm-up phase of the multi-stage training strategy, the learning rate of each iteration increases by 0.0001 compared to the previous iteration.

10. The method for controlling the sieving performance of an airflow microporous sieve according to claim 1, characterized in that, include: Intermediate data of the operation of the microporous sieve under test is collected. After the sieving process is completed, the target sieving performance prediction model is retrained based on the intermediate data to improve the stability of the target sieving performance prediction model.

Images