A method and system for intelligent optimization of grain processing process parameters based on a genetic algorithm

Abstract

The application discloses a kind of based on genetic algorithm's grain processing process parameter intelligent optimization method and system, it is related to process parameter optimization technical field.The based on genetic algorithm's grain processing process parameter intelligent optimization method, by collecting process parameters and grain attribute, constructs grinding state vector;And collect historical grinding state data and construct process and granularity mapping model, and individual coding is carried out in combination with grinding state vector, generates individual gene set, is initialized by knowledge guidance and constructs process initial population, and the fitness score is obtained by multi-layer evaluation, and inverse mapping reconstruction is formed offspring population, then by self-adapting directional variation update obtains new generation population;Whether new generation population meets termination condition is judged, based on its result, extract process optimization parameter set, the present batch of grain is carried out process intelligent optimization processing by process optimization parameter set, to be able to accurately adapt the processing characteristics of grain, to stabilize control grinding granularity in this way, and improve processing quality.

Description

Technical Field

[0001] This invention relates to the field of process parameter optimization technology, specifically to an intelligent optimization method and system for grain processing process parameters based on genetic algorithms. Background Technology

[0002] The quality of controlling grain processing parameters directly affects the production efficiency, product quality, and energy consumption of grain processing. Grain processing involves multiple adjustable process parameters, which are also affected by differences in grain properties, resulting in complex nonlinear relationships between these parameters. Genetic algorithms have the advantages of global optimization and adaptability to complex nonlinear systems. Under multiple parameters and constraints, they can accurately discover the optimal combination of process parameters that takes into account various production objectives. They can flexibly adapt to the optimization needs of different grain varieties and different production conditions, providing core algorithmic support for the intelligent optimization of grain processing parameters.

[0003] The limitations of existing technologies include at least the following problems: existing technologies rely heavily on manual experience for extensive adjustments. This traditional optimization lacks precise optimization capabilities. Grain processing involves multiple adjustable process parameters, which are affected by differences in grain properties. There are complex nonlinear relationships between these parameters, making it difficult to integrate historical grinding data to construct a mapping relationship between process and particle size. Furthermore, there is no precise iterative optimization and disaster recovery mechanism, which makes it difficult to find the optimal combination of process parameters that balances grinding effect and energy consumption. This can easily lead to a mismatch between process parameters and the actual working conditions of the current batch of grain, resulting in the final grinding effect failing to meet expectations. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for intelligent optimization of grain processing parameters based on genetic algorithms. This solves the problem that existing technologies are unable to accurately optimize multiple parameters in grain processing, resulting in unsatisfactory grinding effects.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solution: a method for intelligent optimization of grain processing parameters based on genetic algorithm, comprising the following steps: collecting the process parameter set and grain attribute set of the current batch of grain, and constructing a grinding state vector;

[0006] Historical grinding state data is collected to construct a process-particle size mapping model with grinding state vector as input and particle size distribution vector as output. Based on the grinding state vector and combined with the process-particle size mapping model, individual encoding is performed to generate individual gene sets, which are then initialized using a knowledge-guided strategy to construct an initial process population. The initial process population is used as the current population, and multi-level gene information evaluation is performed on the individual gene sets to obtain the fitness scores of the corresponding individuals. Parent individuals are selected based on the fitness scores, and arithmetic crossover and inverse mapping reconstruction are performed to form a offspring population. Adaptive directional mutation and update processing is then performed to obtain a new generation population. It is determined whether the new generation population meets the preset termination conditions. If it does, a process optimization parameter set is selected based on the new generation population; otherwise, a catastrophic iteration process is performed. Based on the process optimization parameter set, intelligent process optimization processing is performed on the current batch of grain.

[0007] Further, the specific steps for constructing the process-granularity mapping model are as follows: Based on historical grinding state data, a training sample set is selected and divided into a grinding training set and a grinding validation set; a neural network is constructed, and the grinding training set is input into the network, passed through the input layer to the hidden layer, where deep features are extracted by the hidden layer through a nonlinear activation function, and then the output layer generates a predicted granularity distribution vector; the mean squared error is used as the loss function to calculate the error between the predicted granularity distribution vector and the true granularity distribution vector; the backpropagation algorithm is used to propagate the error back layer by layer, and the Adam optimizer is used to update the parameters of each layer; and the model is validated on the grinding validation set. When the validation result meets the preset termination condition, the process-granularity mapping model is obtained.

[0008] Furthermore, the specific steps for generating individual gene sets are as follows: the process parameter set is used as the process parameters to be optimized and encoded as process parameter genes; based on the preset grinding action model, the grinding state vector is subjected to force feature deconstruction processing to generate grinding force feature genes; the process parameter genes are input into the process and particle size mapping model to predict the particle size distribution vector and use it as a particle size gene; the process parameter genes, grinding force feature genes, and particle size genes are fused and grouped to form an individual gene set.

[0009] Furthermore, the specific steps for constructing the initial process population are as follows: within the preset feasible domain of process parameters, a first preset number of random process parameter gene sets are randomly generated; and a second preset number of historical process parameter gene sets are selected from the preset historical experience gene library; the random process parameter gene sets and the historical process parameter gene sets are converted into corresponding individual gene sets respectively, and after merging and verification by process safety constraints, they are combined into the initial process population.

[0010] Further, the specific steps to obtain the fitness score for each individual are as follows: read the particle size distribution vector corresponding to the particle size gene and predict the flour yield of the corresponding individual; read the grinding force characteristic gene and the process parameter gene and predict the power consumption per ton of material and the bran integrity of the corresponding individual; read the particle size gene and the grinding force characteristic gene and predict the ash content of the corresponding individual; perform fusion constraint processing on the flour yield, power consumption per ton of material, bran integrity, and ash content to obtain the fitness score of the corresponding individual.

[0011] Furthermore, the specific steps for forming the offspring population are as follows: Based on fitness scores, select parent individuals from the current population and extract grinding force characteristic genes from the individual gene sets; perform comprehensive processing on the grinding force characteristic genes to generate offspring grinding force characteristic genes; using the offspring grinding force characteristic genes as targets, solve for the corresponding process parameter genes through inverse mapping; input the solved process parameter genes into the process and particle size mapping model to predict the offspring particle size genes; combine the process parameter genes, offspring grinding force characteristic genes, and offspring particle size genes to form the offspring individual gene set, until a preset number of offspring individuals are generated, forming the offspring population.

[0012] Furthermore, the specific steps for solving the corresponding process parameter genes through inverse mapping are as follows: input the offspring grinding force feature genes into the preset inverse mapping model, and establish a target optimization function with the optimization objective of minimizing the difference between the offspring grinding force features and the forward grinding force features corresponding to the model output process parameters; within the preset feasible domain of process parameters, use an iterative optimization algorithm to solve the target optimization function to obtain candidate process parameter gene combinations; perform double constraint verification on the candidate process parameter gene combinations, and select the candidate process parameter genes that pass the verification and have the highest matching degree with the target grinding force features as the final solved process parameter genes.

[0013] Further, the specific steps to obtain the new generation population are as follows: read the fitness score of each offspring individual in the offspring population, and perform initial screening of the population to retain high-quality offspring individuals; calculate the fitness dispersion of the offspring population, dynamically adjust the mutation probability, and extract the grinding force characteristic genes of high-quality offspring individuals to select mutation patterns and obtain the corresponding mutation process parameter genes; input the mutation process parameter genes into the process and particle size mapping model to update the corresponding mutation particle size genes to form a new generation of individuals; merge the unmutated high-quality offspring individuals with the new generation of individuals to form a new generation population.

[0014] Furthermore, the specific steps of the catastrophic iteration process are as follows: record the optimal fitness of the current population during continuous iteration; if the optimal fitness does not improve within a preset number of generations, trigger a catastrophic operation; retain the individual with the highest fitness score in the current population and use its individual gene set as the elite individual of the post-catastrophic population; analyze the corresponding grinding force characteristic genes based on the grinding action model to combine them into a new individual gene set; merge the elite individual with the newly generated individual gene set, and after verification by process safety constraints, form the post-catastrophic population, which is returned as the new current population to continue iteration.

[0015] A smart optimization system for grain processing parameters based on genetic algorithms includes: a grinding state vector construction unit, used to collect the process parameter set and grain attribute set of the current batch of grain and construct a grinding state vector; a mapping model construction unit, used to collect historical grinding state data and construct a process-particle size mapping model with the grinding state vector as input and the particle size distribution vector as output; an encoding and initialization unit, used to perform individual encoding processing based on the grinding state vector and combined with the process-particle size mapping model to generate an individual gene set, and to initialize it using a knowledge-guided strategy to construct an initial process population; and a fitness evaluation unit, used to evaluate the initial process population. The current population performs multi-level genetic information evaluation on individual gene sets to obtain the fitness scores of the corresponding individuals. The offspring generation unit selects parent individuals based on fitness scores and performs arithmetic crossover and inverse mapping reconstruction to form an offspring population. It then performs adaptive directional mutation and update processing to obtain a new generation population. The convergence judgment and catastrophe handling unit determines whether the new generation population meets the preset termination conditions. If it does, it selects the process optimization parameter set based on the new generation population. If it does not meet the conditions, it performs catastrophe iteration processing. The optimization execution unit performs intelligent process optimization processing on the current batch of grains based on the process optimization parameter set.

[0016] The present invention has the following beneficial effects:

[0017] (1) The intelligent optimization method for grain processing parameters based on genetic algorithm collects the process parameters and attribute data of the current batch of grains, constructs an accurate grinding state vector, and constructs a process and particle size mapping model by combining historical data. It can accurately capture the correlation between process parameters and particle size distribution. At the same time, by deconstructing the force features of the grinding state vector, the corresponding gene set is generated, so as to achieve accurate matching between process parameters and actual working conditions of grains. This makes the process parameter settings more adaptable, improves the stability of grinding effect, ensures that grains can obtain suitable processing technology, reduces processing loss caused by improper parameters, and improves processing quality.

[0018] (2) The intelligent optimization method for grain processing parameters based on genetic algorithm adopts a knowledge-guided population initialization strategy, which combines historical high-quality genes and randomly generated genes to construct the initial population. This not only relies on historical experience to ensure the rationality of the initial parameters, but also avoids the trap of local optima through random population. At the same time, an iteration mechanism and a catastrophe handling process are set up. When the optimal fitness stagnates during the iteration process, the population vitality is reactivated through catastrophe operation to ensure that better process parameters can be continuously explored. In addition, by dynamically adjusting the mutation probability and directional mutation mode, the optimization process is made more accurate, effectively solving the problem of being prone to local optima and difficult to continuously improve, and ensuring the stability of the long-term processing process.

[0019] (3) The intelligent optimization system for grain processing parameters based on genetic algorithm can accurately collect the process parameters and attributes of each batch of grain through the grinding state vector construction unit, and construct a vector reflecting the grinding state. The mapping model construction unit combines historical data to build a process and particle size mapping model, which can accurately predict the particle size distribution under different processes and avoid trial and error costs. The encoding and initialization unit generates the initial population through a knowledge-guided strategy, which not only ensures the accuracy of the optimization direction, but also quickly matches the actual processing needs. The fitness evaluation unit can screen the best individuals, and the offspring generation unit continuously iterates and optimizes through crossover and mutation, eventually converging to the preset high-quality parameters, so as to accurately adapt to the processing characteristics of grain, thereby stably controlling the grinding particle size and improving the processing quality.

[0020] Of course, any product implementing this invention does not necessarily need to achieve all of the above advantages at the same time. Attached Figure Description

[0021] Figure 1 This is a flowchart of a method for intelligent optimization of grain processing parameters based on genetic algorithms, according to the present invention.

[0022] Figure 2 This is a flowchart illustrating the specific steps involved in forming a progeny population in a method for intelligent optimization of grain processing parameters based on genetic algorithms, as described in this invention.

[0023] Figure 3 This is a block diagram of a smart optimization system for grain processing parameters based on a genetic algorithm, according to the present invention. Detailed Implementation

[0024] Please see Figure 1, an embodiment of the present invention provides a technical solution: an intelligent optimization method for grain processing process parameters based on genetic algorithm, including the following steps: Collect the process parameter set and grain attribute set of the current batch of grains, and construct the grinding state vector of the current batch of grains. That is, through the high-precision displacement sensor, speed sensor, and flow sensor supporting the flour mill, collect the roll gap, roll speed ratio, and feeding flow rate to form the process parameter set; and detect the moisture content and grain hardness attribute data of the current batch of grains through an online near-infrared analyzer to form the grain attribute set; perform linear dimensionless standardization processing on the above process parameters and grain attribute data to eliminate the differences in dimension and numerical magnitude, and fuse and construct a one-dimensional standardized grinding state vector, with the vector dimension set to 5 dimensions (corresponding to roll gap, roll speed ratio, feeding flow rate, grain moisture, and grain hardness respectively);

[0025] And collect historical grinding state data, and construct a process and particle size mapping model with the grinding state vector as the input and the particle size distribution vector as the output; based on the grinding state vector, perform individual coding processing in combination with the process and particle size mapping model to generate an individual gene set, and use the knowledge guidance strategy for initialization to construct the initial process population; use the initial process population as the current population, perform multi-layer gene information evaluation processing on the individual gene sets of all individuals in the current population to obtain the fitness scores of the corresponding individuals; select parental individuals based on the fitness scores, and perform arithmetic crossover and inverse mapping reconstruction processing to form an offspring population, and perform adaptive directional mutation and update processing to obtain a new generation population;

[0026] And judge whether the new generation population meets the preset termination conditions. If it meets, select the process optimization parameter set based on the new generation population (specifically: the preset termination conditions, condition one is that the improvement amplitude of the optimal fitness of the population for 20 consecutive generations is less than 1%, condition two is that the total number of algorithm iterations reaches 50 generations, and the iteration is terminated if any condition is met; after the iteration is terminated, select the individual with the highest fitness score in the new generation population, extract its process parameter genes, and decode and restore them to the actual process parameters of the roll gap, roll speed ratio, and feeding flow rate that can be directly executed, which is the final process optimization parameter set). If it does not meet, perform catastrophe iteration processing;

[0027] Based on the process optimization parameter set, perform intelligent optimization processing on the current batch of grains. Specifically: for the process optimization parameter set (decoded as roll gap, roll speed ratio, and feeding flow rate), adopt a stepped and stable control strategy, and gradually adjust each process parameter according to the preset small step size. For example: the roll gap is adjusted by 0.02 mm per step, the roll speed ratio is adjusted by 0.1 per step, and the feeding flow rate is adjusted by 0.5 t / h per step. The adjustment interval for each step can be set to 30 seconds until the values of the parameters corresponding to the process optimization parameter set are met.

[0028] Specifically, the historical grinding status data includes the milling gap, speed ratio, feed flow rate, grain moisture, grain hardness, and the corresponding actual particle size distribution vector recorded during the historical production process. The specific steps for constructing the process-particle size mapping model are as follows:

[0029] Based on historical grinding state data, a training sample set was selected and divided into a grinding training set and a grinding validation set. Specifically, the collected historical grinding state data was cleaned and screened in multiple stages. The first step was to remove invalid samples with missing parameters, abnormal data, or distorted sampling. The second step was to remove abnormal samples with drastic fluctuations in operating conditions and indicators that deviated significantly from the normal production range. The third step was to screen balanced samples covering different moisture contents (e.g., 12%-16%) and different hardness grades (e.g., 65-85 hardness values). Finally, a valid steady-state sample group was selected.

[0030] The grinding training set (used for neural network model parameter learning and weight iteration) and the grinding validation set (used for model generalization accuracy testing) were randomly divided into two layers in a 7:3 ratio to ensure that both types of samples fully cover the working conditions of different grain properties and different combinations of process parameters.

[0031] A neural network is constructed, comprising an input layer, hidden layers, and an output layer. The grinding training set is input into the network, passed through the input layer to the hidden layer, where deep features are extracted using a nonlinear activation function. The output layer then generates a predicted particle size distribution vector. Specifically, a four-layer backpropagation neural network is built. The input layer has 5 nodes, consistent with the dimension of the grinding state vector, corresponding to the milling gap, speed ratio, feed flow rate, grain moisture, and grain hardness, respectively. Two hidden layers are set, with the first hidden layer having 20 nodes and the second having 12 nodes. Both hidden layers use the ReLU nonlinear activation function to fully exploit the implicit nonlinear correlation features between process parameters, grain properties, and particle size distribution. The output layer has 4 nodes, corresponding to the proportions of four core particle size ranges: coarse flour, medium flour, fine flour, and bran. The output layer uses the Sigmoid activation function to ensure that the output result is a proportion value between 0 and 1, directly forming the predicted particle size distribution vector.

[0032] Using the mean squared error (MSE) as the loss function, the error between the predicted granularity distribution vector and the true granularity distribution vector is calculated. Specifically, the model loss function is constructed using the MSE to quantify the deviation between the predicted and true granularity distribution vectors, i.e.: ,in, The total number of training samples, The numbers represent the particle size distribution intervals (1-4, corresponding to coarse flour, medium flour, fine flour, and bran, respectively). For the first The first sample The actual granularity percentage of each interval The model predicts the first The first sample The smaller the percentage of each interval granularity and the smaller the value of the loss function, the higher the model's prediction accuracy.

[0033] The backpropagation algorithm is used to propagate the error back layer by layer. The Adam optimizer is used to update the parameters of each layer. Specifically, the total error calculated by the loss function is propagated back from the output layer to the second hidden layer, the first hidden layer, and the input layer. The connection weights and bias coefficients of the neurons in each layer are dynamically adjusted according to the error gradient (the gradient is calculated using the chain rule). The Adam adaptive moment estimation optimizer is used to update the parameters. For example, the initial learning rate can be set to 0.001 and the weight decay coefficient can be set to 0.0001 to adaptively adjust the learning rate. The training is continued iteratively until the loss function converges and stabilizes, such as when the fluctuation is less than 0.0001.

[0034] The model is then validated on a grinding validation set. When the validation results meet the preset termination conditions, the process-particle size mapping model is obtained. Specifically, the independent grinding validation set is input into the trained neural network model, and the goodness of fit between the predicted particle size distribution and the actual particle size distribution is calculated. The preset model validation termination condition is that the goodness of fit of the validation set is ≥96%, and the maximum prediction error of a single sample is ≤2.5%. After the termination conditions are met for three consecutive validations, the model training is stopped, the optimal model weights and bias parameters are saved, and the process-particle size mapping model is obtained.

[0035] In this implementation plan, samples are screened through multiple steps to remove invalid and abnormal data, and samples that closely match actual production conditions are selected to ensure the quality of samples used for model training and validation. The constructed neural network can accurately capture the correlation between process parameters, grain properties, and particle size distribution. Through a reasonable network structure and activation function, deep correlation features are extracted to accurately output particle size distribution prediction results. The prediction deviation is quantified by the loss function, and the network parameters are continuously adjusted through backpropagation and optimizers to improve the model's prediction accuracy. After a rigorous validation process, it is ensured that the model can stably output reliable results, avoiding blind adjustments and ensuring stable processing effects. At the same time, the model can adapt to different production conditions, improving its adaptability in practical applications.

[0036] Specifically, the generation of individual gene sets is as follows: The process parameter set is taken as the process parameters to be optimized and encoded as process parameter genes. Specifically, the rolling gap, grinding roll speed ratio, and feed flow rate are taken as the parameters to be optimized by the genetic algorithm and encoded as real numbers according to the actual safe operating range of the grinding mill equipment. For example, the rolling gap encoding range is 0.2-1.2mm, the speed ratio encoding range is 1.5-2.8, and the feed flow rate encoding range is 5-15t / h. The encoding precision is set to two decimal places, and the encoding length is set to 32 bits (each set of parameters corresponds to 10-11 bits of encoding), forming a fixed-length process parameter gene.

[0037] Based on the preset grinding action model, the grinding state vector is subjected to force feature deconstruction processing to generate grinding force feature genes. Specifically, the grinding action model is a grain grinding mechanical coupling model, which is a dedicated model constructed based on the discrete element method (DEM) and Hertz contact theory. The model includes a grain particle discrete element model, a grinding roller contact mechanical model, and a force breakage criterion module.

[0038] The specific expression of the model is: (in, This refers to the normal contact force between the grinding roller and the grain. The contact stiffness coefficient (with a value range of 1.2 × 10⁻⁶) 5 -1.8×10 5 N / m), The contact deformation amount (the deformation amount when the grain comes into contact with the grinding roller, which is obtained by simulating the grain grinding process through a model and tracking the contact state of the particles). It is a non-linear exponent (with a value of 1.5-2.0). The damping coefficient (values ​​range from 0.8 to 1.2 N·s / m) The contact deformation rate (the speed at which grains deform upon contact, obtained directly by monitoring the dynamic process of particle contact deformation in real time through the model);

[0039] Furthermore, before model construction, a large number of wheat grinding tests need to be conducted, and parameter calibration needs to be completed in combination with the actual structural parameters of the grinding roller (grind roller diameter 500-600mm, roller surface roughness 0.8-1.6μm, grinding tooth angle 30°-45°).

[0040] Force feature deconstruction and extraction process: The standardized grinding state vector (including the roller gap, roller speed ratio, feed flow rate, grain moisture, and grain hardness) is used as the model input parameters. Among them, the roller gap and roller speed ratio determine the contact gap and relative motion state of the rollers, the feed flow rate affects the distribution density of grain particles between the rollers, and the grain moisture and hardness determine the compressive strength and shear strength of the grain particles. All of the above parameters are imported through the model interface, and the fixed structure parameters of the rollers are entered simultaneously.

[0041] The grinding mechanics coupling model is activated to simulate the force process of grain particles after entering the gap between the grinding rollers. The motion trajectory and force changes of individual grain particles are tracked by the discrete element method. The contact force between the grinding roller and the grain, and between grain particles, is calculated by combining Hertz contact theory. Then, the breakage state of the grain particles is judged by the force breakage criterion (breakage criterion: when the contact force is greater than the compressive strength threshold of the grain particle, the particle is judged to be broken). The dynamic mechanical data (such as contact deformation amount and contact deformation rate) of the grinding process are output in real time.

[0042] From the dynamic mechanical data output by the model, the following mechanical features are extracted: grinding pressure per unit area (calculated by dividing the total contact force of the grinding rollers by the contact area, in MPa, reflecting the compressive strength of the grinding rollers on the grain), grinding roller linear velocity difference (calculated by the grinding roller speed and diameter, in m / s, reflecting the shear strength of the grinding rollers on the grain), and grinding force uniformity (calculated by the ratio of the standard deviation to the average value of the force on a single grain particle, dimensionless, reflecting the uniformity of the force on the grain between the grinding rollers).

[0043] The above force characteristics are subjected to min-max standardization to eliminate dimensional differences. The standardized values ​​are controlled between 0 and 1. Then, they are encoded according to a fixed coding rule (using 32-bit real number coding, with each mechanical characteristic corresponding to 10-11 bits of coding) to finally form a grinding force characteristic gene. This gene can accurately characterize the actual stress and breakage state of grains during the grinding process.

[0044] The process parameter gene is input into the process-particle size mapping model to predict the particle size distribution vector, which is then used as the particle size gene. Specifically, the process parameter gene is decoded and restored to the actual process parameters (decoding and encoding are inverses), and then substituted into the trained process-particle size mapping model. The model quickly predicts the grinding particle size distribution result corresponding to the set of process parameters through forward propagation calculation. The predicted proportion values ​​of the four particle size intervals are encoded according to the 32-bit real number encoding rule to form the particle size gene, which represents the final grinding effect of the corresponding process scheme.

[0045] The process parameter gene, grinding force characteristic gene, and particle size gene are fused and grouped to form an individual gene set. Specifically, the process parameter gene, grinding force characteristic gene, and particle size gene are sequentially spliced ​​and fused to form a complete gene set for a single individual. Each gene set uniquely corresponds to a complete grinding process scheme.

[0046] The specific steps for constructing the initial process population are as follows: Within the preset feasible domain of process parameters, a first preset number of random process parameter gene sets are randomly generated. Specifically, the preset feasible domain of process parameters can be the safe operating range of the grinding mill, i.e., the rolling gap is 0.2-1.2mm, the speed ratio is 1.5-2.8, and the feed flow rate is 5-15t / h. The first preset number is set to 60. The rand function can be used to uniformly generate 60 sets of compliant process parameter combinations within the above feasible domain, which are then encoded as random process parameter gene sets.

[0047] Then, a second preset number of historical process parameter gene sets are selected from the preset historical experience gene library. Specifically, the historical experience gene library is pre-constructed, which stores more than 200 sets of high-quality and compliant process parameters from wheat milling production in the past two years, with a flour yield of ≥78%, power consumption per ton of material ≤65kWh / t, and ash content ≤0.65%. All of these parameters have been verified by actual production (each set of parameters corresponds to more than 3 hours of stable production data without abnormal operating conditions). The second preset number is set to 40. Based on the moisture and hardness attributes of the current batch of grain, the cosine similarity algorithm (similarity ≥0.85 is considered suitable) is used to select 40 sets of the most suitable high-quality historical process parameters from the library and encode them as historical process parameter gene sets.

[0048] The random process parameter gene set and the historical process parameter gene set are converted into corresponding individual gene sets, merged, and verified by process safety constraints to form an initial process population containing several individuals. Specifically, the random process parameter gene set and the historical process parameter gene set are processed according to the above steps: the grinding action model is called to generate grinding force feature genes, and the process and particle size mapping model is called to generate particle size genes. After completion, a complete individual gene set is formed. After merging the two types of individuals, 100 initial individuals are obtained. Process boundary safety verification is performed on each one (whether the parameters are within the feasible region). Abnormal individuals with parameters exceeding the standard are removed. If the preset size is not met after removal, a new random process parameter gene set is generated and converted into an individual gene set to supplement it. Finally, an initial process population of 100 individuals is formed, which meets the stable iteration requirements of the genetic algorithm.

[0049] The specific steps to obtain the fitness score of each individual are as follows: Read the particle size distribution vector corresponding to the particle size gene in the individual gene set of each individual in the initial population of the process, and predict the flour yield of the corresponding individual. Specifically, extract the particle size gene, decode the encoded standardized value to restore the actual particle size distribution vector, which includes the specific proportion of four core particle size intervals: coarse flour (particle size 150-300μm), medium flour (75-150μm), extra fine flour (40-75μm), and bran (>300μm).

[0050] Subsequently, referring to the particle size requirements for special refined flour and medium flour in the "Wheat Flour" standard, it was clarified that the particle size of special refined flour should be controlled between 40-75μm, and the particle size of medium flour should be controlled between 75-150μm. The total proportion of particles in these two ranges in the particle size distribution vector was used as the basic finished product proportion.

[0051] The screening efficiency and material loss coefficient in actual production are accurately calculated. The screening efficiency is determined based on the aperture of the screens used in the current grinding mill (3 levels of screens: 40μm, 75μm, and 150μm) and the screen rotation speed (300r / min). The screening efficiency is 98%, and the loss coefficient is calibrated to 0.96 based on historical production data (i.e., the material loss rate is 4%). The calculation formula is: Powder yield = (proportion of high-quality powder + proportion of medium-quality powder) × screening efficiency × loss coefficient. This formula can be used to accurately predict the finished powder yield of a given product.

[0052] The grinding force characteristic genes and process parameter genes of each individual in the initial population of the process are read, and the power consumption per ton of material and bran integrity of the corresponding individuals are predicted. Specifically:

[0053] First, the grinding force characteristic genes and process parameter genes of the current individual are extracted simultaneously, and the actual grinding mechanical parameters (grinding pressure per unit area, grinding roller linear speed difference, grinding force uniformity) and process control parameters (roller gap, grinding roller speed ratio, feed flow rate) are decoded respectively.

[0054] For the prediction of power consumption per ton of material, the rated power of the grinding mill (preset to 110kW) and the actual load rate (determined by the grinding roller speed ratio, feed flow rate, and grinding pressure per unit area, load rate = actual grinding power / rated power, the actual grinding power is calculated by the grinding pressure per unit area and the grinding roller speed in the grinding force characteristic gene, i.e.: actual grinding power = ), and the power characteristic curve of the grinding mill (obtained in advance through equipment factory testing and on-site debugging, characterizing the correlation between load rate and unit energy consumption), first calculate the actual power consumption per unit time ( = Then, the power consumption per ton of material is converted into the power consumption per unit of material, i.e.: power consumption per ton of material = actual power consumption / feed amount per unit time × 1000, the unit is kWh / t. The lower the power consumption per ton of material, the better the score of this indicator.

[0055] For the prediction of bran integrity, the proportion of bran particles (>300μm) in the particle size distribution vector is used as the basis, and the grinding force uniformity in the grinding force characteristic gene is used for correction. The higher the force uniformity, the less bran is broken and the higher the integrity. That is: bran integrity = bran particle proportion × (1 + grinding force uniformity × 0.3), where the grinding force uniformity is standardized to a value of 0-1. The higher the bran integrity value, the better the score of this indicator.

[0056] The particle size gene and grinding force characteristic gene of each individual in the initial population of the process are read, and the ash value of the corresponding individual is predicted. Specifically, the particle size gene and grinding force characteristic gene of the individual are extracted, and the proportion of bran debris impurities (particle size 20-40μm, which are fine impurities after bran is broken and cannot be separated by a sieve, which will lead to an increase in ash content) in the particle size distribution is decoded, as well as the grinding pressure per unit area in the grinding force characteristic gene (the higher the pressure, the more severe the bran is broken, the more debris impurities there are, and the higher the ash content).

[0057] Based on the above parameters, the prediction formula is: Ash content = 0.35 + (proportion of bran debris and impurities × 0.8) + (grinding pressure per unit area × 0.0004), where 0.35 is the basic ash content of wheat grains (unit: %), the proportion of bran debris and impurities is the standardized value (0-1), and the grinding pressure per unit area is in MPa;

[0058] The flour yield, power consumption per ton of material, bran integrity, and ash content were subjected to fusion constraint processing to obtain the fitness score of the corresponding individual, as follows:

[0059] The above indicators are standardized using the min-max method, and the values ​​of each indicator are uniformly converted into standardized scores in the range of 0-1. Among them, the power consumption per ton of material and the ash content are negative indicators (the larger the value, the lower the standardized score, and the above results need to be subtracted from 1 when calculating).

[0060] The standardized results are weighted with their corresponding preset weights (such as flour yield weight 0.35, power consumption per ton weight 0.3, bran integrity weight 0.2, and ash content weight 0.15) to calculate the individual basic fitness score.

[0061] If the process parameters corresponding to an individual exceed the preset feasible range (roll gap < 0.2mm or > 1.2mm, roller speed ratio < 1.5 or > 2.8, feed flow rate < 5t / h or > 15t / h), or any indicator exceeds the reasonable value range (powder yield < 65% or > 85%, power consumption per ton of material > 75kWh / t, bran integrity < 70%, ash content > 0.85%), a penalty of 10-20 points will be applied according to the degree of exceedance (slight exceedance: within 5% of the reasonable range, deduct 10 points; severe exceedance: 5% or more of the reasonable range, deduct 20 points). The base score and penalty points will be combined and then normalized to map the final score to the 0-100 range to obtain the final fitness score of the individual.

[0062] This implementation plan clarifies the coding methods for process parameters, grinding force characteristics, and particle size distribution, ensuring that each gene group accurately corresponds to the actual processing conditions. By combining historical high-quality parameters with randomly generated parameters, a reasonable initial population is constructed, avoiding blind initialization. At the same time, the fitness of individuals is comprehensively evaluated through multi-dimensional indicators, taking into account processing effect, energy consumption, and product quality, effectively improving the accuracy of process parameter optimization. This ensures that the entire optimization process conforms to actual production needs, thereby reducing errors caused by human intervention.

[0063] Specifically, such as Figure 2 As shown, the specific steps for forming the offspring population are as follows: Based on the fitness score, select two parent individuals from the current population and extract the grinding force characteristic genes from the gene sets of each individual. Specifically, the tournament selection method is used to randomly select a number of individuals (e.g., 5) from the current population, select the two individuals with the highest fitness scores as parent individuals, and extract the grinding force characteristic genes from the two parent individuals respectively.

[0064] The two grinding force trait genes are combined to generate offspring grinding force trait genes. Specifically, an arithmetic cross operation is performed on the two parent grinding force trait genes to fuse the superior traits of the parents and generate offspring mechanical genes that combine the advantages of both. ,in, The weighting coefficients are random values ​​between 0 and 1 (generated randomly using the rand function for each crossover, with values ​​ranging from 0.1 to 0.9). These are two parental grinding force characteristic gene vectors (corresponding to the standardized values ​​of grinding pressure per unit area, grinding roller linear velocity difference, and grinding force uniformity, respectively). This represents the gene vector for offspring grinding force characteristics.

[0065] Using the grinding force characteristic genes of offspring as the target, the corresponding process parameter genes are solved by inverse mapping;

[0066] The process parameter genes obtained by solving are input into the process and granularity mapping model to predict the offspring granularity genes. Specifically, the process parameter genes obtained by inverse mapping are decoded and input into the trained process and granularity mapping model. The model quickly predicts the corresponding granularity distribution vector through forward propagation. The prediction results are encoded according to the 32-bit real number encoding rule to obtain the offspring granularity genes.

[0067] The process parameter gene, the offspring grinding force characteristic gene, and the offspring particle size gene are combined to form the offspring individual gene set. The above operation is repeated until a preset number of offspring individuals are generated, forming an offspring population, which includes several offspring individuals. Specifically, the steps of parent selection, gene crossover, inverse mapping solution, and particle size prediction are repeated until offspring individuals that meet the preset offspring size (e.g., 100) are generated. The feasibility of the process is verified one by one (whether the parameters are within the feasible region), and abnormal individuals are removed. If the number of individuals after removal is less than 100, the corresponding number of offspring individuals are generated to finally form an offspring population with the same size as the parent population.

[0068] The specific steps for solving the corresponding process parameter genes through inverse mapping are as follows: Input the offspring grinding force feature genes into the preset inverse mapping model. The optimization objective is to minimize the difference between the offspring grinding force features and the forward grinding force features corresponding to the model's output process parameters. Specifically, the objective optimization function is established based on the forward mechanical transmission relationship in grain grinding. This model uses a BP neural network structure, with the input layer being the offspring target grinding force features (3D), the output layer being the process parameters (3D: roller gap, roller speed ratio, feed flow rate), and one hidden layer (10 nodes). The activation function is ReLU. Using the offspring target grinding force features as input, a minimum deviation optimization objective function is constructed: ,in Let be the gene vector of the process parameters to be solved. The first characteristic of the grinding force of the offspring target Item Indicators ( These correspond to the grinding pressure per unit area, the difference in linear speed of the grinding rollers, and the uniformity of grinding force, respectively. For process parameters The corresponding positive grinding force characteristic is the first The optimization objective for each index (calculated using a grinding action model) is to minimize the deviation between the two, while also incorporating constraints from the process feasible region. The requirements are: roll gap 0.2-1.2mm, speed ratio 1.5-2.8, and feed flow rate 5-15t / h.

[0069] Within the preset feasible domain of process parameters, an iterative optimization algorithm is used to solve the objective optimization function to obtain candidate process parameter gene combinations. Specifically, within the feasible domain of rolling gap, speed ratio, and feed flow rate, a gradient descent iterative optimization algorithm is used, with a maximum number of iterations set to 100, a convergence accuracy of 0.001, and a learning rate of 0.005, to quickly solve the objective optimization function and obtain 3-5 sets of candidate process parameter combinations, which are encoded as candidate process parameter gene combinations (the encoding rules are the same as above).

[0070] The candidate process parameter gene combinations are subjected to dual constraint verification. The candidate process parameter gene that passes the verification and has the highest matching degree with the target grinding force characteristics is selected as the final process parameter gene. Specifically, the candidate parameter combinations are subjected to dual verification of process boundary constraints and grinding force safety constraints in sequence. Process boundary constraint verification: check whether the parameters are within the feasible domain; grinding force safety constraint verification: substitute the candidate parameters into the grinding action model, calculate the actual grinding force characteristics, and determine whether there are problems such as excessive force (grinding pressure per unit area > 1.5MPa) or insufficient force (grinding pressure per unit area < 0.5MPa), uneven force (force uniformity < 0.3), etc. Parameters that do not meet the above constraints are eliminated; calculate the matching degree between the remaining candidate parameters and the target grinding force characteristics (using cosine similarity calculation), and select the group with the highest matching degree as the final process parameter gene.

[0071] In this implementation plan, the grinding force characteristics of the offspring are taken as the target. The appropriate process parameter genes are solved by inverse mapping to ensure that the process parameters are accurately matched with the target grinding effect. At the same time, through multi-step verification, non-compliant parameters are eliminated and the process parameters with the best fit are selected to avoid invalid iterations. By relying on the high-quality genes of the parent generation and through reasonable optimization and verification, it is ensured that the offspring population meets the production requirements, effectively guaranteeing the rationality of the process parameters and conforming to the actual production scenario, thereby improving the optimization efficiency and effect.

[0072] Specifically, the steps to obtain the new generation population are as follows: read the fitness score of each offspring individual in the offspring population, and perform a preliminary screening of the population to retain high-quality offspring individuals. Specifically, the fitness score threshold is preset to 60 points. Read the fitness score of each individual in the offspring population one by one, retain offspring individuals with a score ≥ 60 points, and mark them as high-quality offspring individuals.

[0073] The fitness dispersion of the offspring population is calculated, and the mutation probability is dynamically adjusted. Furthermore, the grinding force characteristic genes of high-performing offspring individuals are extracted to select mutation patterns and obtain corresponding mutation process parameter genes. Specifically, the fitness dispersion coefficient of high-performing offspring individuals (i.e., the ratio of the standard deviation to the mean of the fitness of all individuals in the offspring population) is calculated and used as the fitness dispersion. Lower dispersion indicates higher population homogeneity, correspondingly increasing the mutation probability (when the dispersion coefficient < 0.1, the mutation probability is set to 0.15-0.2; when the dispersion coefficient is 0.1-0.2, the mutation probability is set to 0.08-0.14; when the dispersion coefficient > 0...). When the mutation probability is 0.2, the mutation probability is set to 0.05-0.07, and the mutation probability is dynamically controlled within the range of 0.05-0.2. The high-quality offspring grinding force characteristic genes are extracted and compared with the preset optimal grinding force characteristic range (grinding pressure per unit area 0.8-1.2MPa, grinding roller linear speed difference 1.5-2.5m / s, grinding force uniformity 0.6-0.9) to determine the direction of deviation. If the force exceeds the tolerance, the corresponding process parameters are finely adjusted in the reverse direction (e.g., if the pressure is too high, the rolling gap is increased). If the force is insufficient, the process parameters are finely adjusted in the forward direction (e.g., if the pressure is too low, the rolling gap is decreased). The fine adjustment range is ≤10%, and the mutated process parameter genes are obtained.

[0074] The mutated process parameter gene is input into the process and particle size mapping model to update the corresponding mutated particle size gene in order to form a new generation of individuals. Specifically, the mutated process parameter gene is decoded and input into the process and particle size mapping model to re-predict the corresponding particle size distribution vector, update and generate the mutated particle size gene, and combine it with the mutated process parameter gene and the original grinding force characteristic gene to form a complete new generation of individuals.

[0075] The high-quality offspring individuals that have not undergone mutation are merged with the new generation individuals to form a new generation population. Specifically, the high-quality offspring individuals that have not participated in mutation are retained and merged with the new generation individuals that have passed the mutation test. If the population size after merging is less than 100, a small number of random high-quality individuals (randomly selected from the historical experience gene bank) are added to make up the population size to 100. After final process safety verification (parameter compliance), a new generation population is formed.

[0076] In this implementation plan, high-quality offspring individuals are selected, and those that meet the scoring criteria are retained to lay the foundation for population optimization. Simultaneously, by calculating the population's fitness dispersion and dynamically adjusting the mutation probability, the plan effectively addresses population homogeneity issues, flexibly activates the population's exploratory capabilities, and avoids optimization getting stuck in local optima. Furthermore, by combining the grinding power characteristics of high-quality individuals with targeted fine-tuning based on the target range, process parameters are precisely corrected to ensure that the mutation direction aligns with optimal grinding requirements. Then, the model predicts and updates the granular genes to generate a complete new generation of individuals, ensuring that each individual has a reliable correlation between process and grinding effect. Finally, unmutated high-quality individuals and mutated qualified individuals are merged to reasonably supplement the population size and complete compliance verification, forming an overall high-quality new generation population. This ensures the continuation of the population's superior genes and continuously improves the efficiency and effectiveness of iterative optimization.

[0077] Specifically, the steps of the catastrophe iteration process are as follows: record the optimal fitness of the current population during the continuous iteration process. If the optimal fitness does not improve within a preset number of generations, a catastrophe operation is triggered. Specifically, the preset number of continuous detection generations is 15 generations, and the optimal fitness value of each generation (the highest fitness score in each generation) is recorded throughout the process. If the improvement of the optimal fitness of the population is less than 0.5% for 15 consecutive generations, the algorithm is determined to be stuck in a local optimum stagnation state, and a catastrophe operation is automatically triggered.

[0078] The individual with the highest fitness score in the current population is retained, and its individual gene set is used as the elite individual in the post-disaster population. Specifically, the individual with the highest fitness score in the current population is selected. This individual corresponds to the optimal process scheme in the current iteration cycle, and its individual gene set is completely retained as the elite individual in the post-disaster population.

[0079] All other individual gene sets were discarded, and a preset number of process parameter genes were randomly generated within the range of process parameter values. Each newly generated set of process parameter genes was then input into the process and granularity mapping model to predict the corresponding granularity gene. Specifically, except for elite individuals, the gene sets of the remaining 99 individuals were discarded. Within the feasible domain of process parameters, 99 sets of process parameter genes were uniformly and randomly generated again using the rand function. Each new set of genes was decoded and input into the process and granularity mapping model to predict the corresponding granularity distribution one by one, and encoded as a granular gene, thus completing the construction of the basic gene of the new individual.

[0080] Based on the grinding action model, the corresponding grinding force characteristic genes are analyzed and combined into new individual gene sets. Specifically, for each newly generated process parameter gene set, the grinding mechanics action model is called, the process parameters and grinding roller fixed structure parameters are substituted, the corresponding grinding force characteristic genes are analyzed and extracted, and fused and grouped with the process parameter genes and particle size genes according to a fixed coding rule (96 bits, 32 bits per gene set) to form 99 new complete individual gene sets.

[0081] The elite individuals are merged with the newly generated individual gene sets, and after process safety constraint verification, a post-catastrophic population is formed and returned as the new current population to continue iteration. Specifically, the retained elite individuals are merged with 99 sets of new individual gene sets to form a post-catastrophic population of 100 individuals. The process safety of each individual (whether the parameters are compliant) is verified one by one. After removing abnormal individuals, if the number is less than 100, the corresponding number of new individuals are generated as the new current population. The fitness evaluation step is then returned to continue iterative optimization until the algorithm termination condition is met.

[0082] In this implementation plan, by recording the optimal fitness of the population, it is possible to promptly determine whether a catastrophe has been triggered, ensuring the continuous advancement of the optimization process. The best individuals in the current population are retained as elite individuals to perpetuate superior genes and prevent the loss of superior process solutions. At the same time, the remaining individuals are discarded, and new individual gene sets are generated and new parameter combinations are injected to break the homogeneity of the population and reactivate the optimization vitality. The newly generated individuals undergo model analysis and gene combination, and then pass security verification to ensure compliance. Finally, the post-catastrophe population continues to iterate, thereby ensuring that the algorithm can continuously discover better process parameters, meet the optimization needs of actual production, and improve the overall optimization effect and efficiency.

[0083] Please see Figure 3 This invention provides a technical solution: an intelligent optimization system for grain processing parameters based on a genetic algorithm, comprising: a grinding state vector construction unit, used to collect the process parameter set and grain attribute set of the current batch of grain, and construct a grinding state vector; a mapping model construction unit, used to collect historical grinding state data, and construct a process-particle size mapping model with the grinding state vector as input and the particle size distribution vector as output; an encoding and initialization unit, used to perform individual encoding processing based on the grinding state vector and combined with the process-particle size mapping model, generate an individual gene set, and initialize it using a knowledge-guided strategy to construct an initial process population; and a fitness evaluation unit. The system is used to: 1) Use the initial population as the current population, perform multi-level genetic information evaluation on the individual gene sets to obtain the fitness score of the corresponding individuals; 2) Use the offspring generation unit to select parent individuals based on the fitness score, perform arithmetic crossover and inverse mapping reconstruction to form the offspring population, and perform adaptive directional mutation and update processing to obtain the new generation population; 3) Use the convergence judgment and catastrophe handling unit to determine whether the new generation population meets the preset termination conditions. If it does, the system selects the process optimization parameter set based on the new generation population; if it does not, the system performs catastrophe iteration processing; 4) Use the optimization execution unit to perform intelligent process optimization processing on the current batch of grain based on the process optimization parameter set.

[0084] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0085] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for intelligent optimization of grain processing parameters based on genetic algorithms, characterized in that, Includes the following steps: Collect the process parameter set and grain attribute set of the current batch of grain, and construct a grinding state vector; It also collects historical grinding state data and constructs a process-particle size mapping model with grinding state vector as input and particle size distribution vector as output; Based on the grinding state vector and combined with the process and particle size mapping model, individual encoding is performed to generate individual gene sets, and a knowledge-guided strategy is used for initialization to construct the initial process population; Using the initial population as the current population, multi-level genetic information evaluation is performed on the individual gene sets to obtain the fitness scores of the corresponding individuals. Parent individuals are selected based on fitness scores, and arithmetic crossover and inverse mapping reconstruction are performed to form offspring populations. Adaptive directional mutation and update are then performed to obtain a new generation population. It also determines whether the new generation of the population meets the preset termination conditions. If it does, it selects the process optimization parameter set based on the new generation of the population. If it does not meet the conditions, it performs a catastrophic iteration process. Based on the set of process optimization parameters, intelligent process optimization is performed on the current batch of grains.

2. The intelligent optimization method for grain processing parameters based on genetic algorithm according to claim 1, characterized in that, The specific steps for constructing the process-granularity mapping model are as follows: Based on historical grinding status data, a training sample set was selected and divided into a grinding training set and a grinding validation set. A neural network is constructed, and the grinding training set is input into the network. The input layer is passed to the hidden layer, the hidden layer extracts deep features through a non-linear activation function, and the output layer generates a prediction granularity distribution vector. The mean squared error is used as the loss function to calculate the error between the predicted granularity distribution vector and the true granularity distribution vector; The backpropagation algorithm is used to propagate the error back layer by layer, and the Adam optimizer is used to update the parameters of each layer. The process is then validated on a grinding validation set. When the validation results meet the preset termination conditions, the process-particle size mapping model is obtained.

3. The intelligent optimization method for grain processing parameters based on genetic algorithm according to claim 1, characterized in that, The specific steps for generating an individual's gene set are as follows: The process parameter set is used as the process parameters to be optimized and encoded as process parameter genes; Based on the preset grinding action model, the grinding state vector is subjected to force feature deconstruction processing to generate grinding force feature genes; The process parameter gene is input into the process and particle size mapping model to predict the particle size distribution vector, which is then used as the particle size gene. The process parameter genes, grinding force characteristic genes, and particle size genes are fused and grouped to form an individual gene set.

4. The intelligent optimization method for grain processing parameters based on genetic algorithm according to claim 3, characterized in that, The specific steps for constructing the initial population for the process are as follows: Within the preset feasible domain of process parameters, a first preset number of random process parameter gene sets are randomly generated; And select a second preset number of historical process parameter gene sets from the preset historical experience gene library; The random process parameter gene set and the historical process parameter gene set are converted into corresponding individual gene sets, merged, and verified by process safety constraints to form the initial process population.

5. The intelligent optimization method for grain processing parameters based on genetic algorithm according to claim 3, characterized in that, The specific steps to obtain the fitness score for each individual are as follows: Read the particle size distribution vector corresponding to the particle size gene and predict the powder yield of the corresponding individual; Read the grinding force characteristic genes and process parameter genes, and predict the power consumption per ton of material and bran integrity of the corresponding individuals; Read the particle size gene and grinding force characteristic gene, and predict the ash value of the corresponding individual; The flour yield, power consumption per ton of material, bran integrity, and ash content were combined and constrained to obtain the fitness scores of the corresponding individuals.

6. The intelligent optimization method for grain processing parameters based on genetic algorithm according to claim 1, characterized in that, The specific steps for forming a progeny population are as follows: Based on fitness scores, parent individuals are selected from the current population, and grinding force characteristic genes are extracted from the individual gene sets. The grinding force characteristic genes were comprehensively processed to generate offspring grinding force characteristic genes. Using the grinding force characteristic genes of offspring as the target, the corresponding process parameter genes are solved by inverse mapping; The obtained process parameter genes are input into the process and particle size mapping model to predict the offspring particle size genes. The process parameter gene, the offspring grinding force characteristic gene, and the offspring particle size gene are combined to form the offspring individual gene set, until a preset number of offspring individuals are generated, forming an offspring population.

7. The intelligent optimization method for grain processing parameters based on genetic algorithm according to claim 6, characterized in that, The specific steps for solving the corresponding process parameter genes through inverse mapping are as follows: The offspring grinding force characteristic genes are input into a preset inverse mapping model. The objective optimization function is established with the goal of minimizing the difference between the offspring grinding force characteristics and the forward grinding force characteristics corresponding to the model output process parameters. Within the preset feasible region of process parameters, the objective optimization function is solved by an iterative optimization algorithm to obtain candidate process parameter gene combinations; The candidate process parameter gene combinations are subjected to dual constraint verification. The candidate process parameter gene that passes the verification and has the highest matching degree with the target grinding force characteristics is selected as the final process parameter gene obtained by solving.

8. The intelligent optimization method for grain processing parameters based on genetic algorithm according to claim 1, characterized in that, The specific steps to obtain the new generation of the population are as follows: Read the fitness score of each offspring individual in the offspring population, and perform initial screening of the population to retain high-quality offspring individuals; The fitness dispersion of the offspring population is calculated, the mutation probability is dynamically adjusted, and the grinding force characteristic genes of high-quality offspring individuals are extracted to select mutation patterns and obtain the corresponding mutation process parameter genes. The mutant process parameter genes are input into the process and granularity mapping model to update the corresponding mutant granularity genes, so as to form a new generation of individuals; Unmutated, high-quality offspring are merged with new-generation individuals to form a new generation population.

9. The intelligent optimization method for grain processing parameters based on genetic algorithm according to claim 1, characterized in that, The specific steps for disaster iteration processing are as follows: Record the optimal fitness of the current population during continuous iterations. If the optimal fitness does not improve within a preset number of generations, trigger a catastrophe operation. The individual with the highest fitness score in the current population is retained, and its gene set is used as the elite individual in the post-cataclysmic population. Based on the grinding action model, the corresponding grinding force characteristic genes were analyzed and combined into a new individual gene set; The elite individuals and the newly generated individual gene sets are merged, and after being verified by process safety constraints, a post-disaster population is formed, which is then returned as the new current population to continue iterating.

10. A smart optimization system for grain processing parameters based on genetic algorithms, employing the smart optimization method for grain processing parameters based on genetic algorithms according to any one of claims 1-9, characterized in that, include: The grinding state vector construction unit is used to collect the process parameter set and grain attribute set of the current batch of grain and construct the grinding state vector. The mapping model construction unit is used to collect historical grinding state data and construct a process-particle size mapping model with the grinding state vector as input and the particle size distribution vector as output. The encoding and initialization unit is used to encode individuals based on the grinding state vector and combined with the process and particle size mapping model, generate individual gene sets, and use a knowledge-guided strategy to initialize and construct the initial population of the process. The fitness assessment unit is used to take the initial population of the process as the current population, perform multi-level genetic information assessment on the individual gene set, and obtain the fitness score of the corresponding individual. The offspring generation unit is used to select parent individuals based on fitness scores, and to perform arithmetic crossover and inverse mapping reconstruction to form an offspring population. Adaptive directional mutation and update are then performed to obtain a new generation population. The convergence judgment and disaster handling unit is used to determine whether the new generation population meets the preset termination conditions. If it does, the process optimization parameter set is selected based on the new generation population. If it does not meet the conditions, disaster iteration processing is performed. The optimized execution unit is used to perform intelligent process optimization on the current batch of grains based on the process optimization parameter set.

Images