A method and apparatus for optimizing leaf-breaking air separation process parameters based on GA-BP neural network

Abstract

This invention discloses a method and apparatus for optimizing leaf trimming and air separation process parameters based on a GA-BP neural network. The method includes selecting the leaf trimming and air separation process parameters to be optimized, acquiring sample data, and normalizing the sample data; constructing a BP neural network and randomly generating a first preset number of prediction arrays; predicting the predicted values ​​of leaf structure indicators, calculating the prediction difference between the predicted values ​​and the actual values ​​of the leaf structure indicators, and setting a GA genetic algorithm fitness function in the BP neural network; selecting an orthogonal array based on the number of leaf trimming and air separation process parameters for orthogonal experiments, performing range analysis on the orthogonal experiment results, and determining a first optimal combination of process parameters; and calculating the predicted process parameter values ​​based on the assigned values ​​in the BP neural network to obtain a second optimal combination of process parameters. This method greatly reduces the amount of experimentation, allowing the BP neural network to make targeted predictions and obtain the optimal combination of process parameters without requiring extensive experimentation.

Description

Technical Field

[0001] This application relates to the field of tobacco processing technology, and more specifically, to a method and apparatus for optimizing leaf threshing and air separation process parameters based on a GA-BP neural network. Background Technology

[0002] Tillage and re-drying is a sorting and preparatory processing process that transforms tobacco leaves from agricultural products into industrial raw materials. After tillage and re-drying, some tobacco stems no longer undergo aging and occupy aging warehouses, resulting in more uniform tobacco leaf quality. At the same time, dust, debris, fragments, short and broken stems can be processed on-site at the re-drying plant, eliminating the need to transport them to the cigarette factory and thus reducing transportation volume.

[0003] The leaf threshing and air separation process is a key step in the leaf threshing and re-drying production line. Its effectiveness directly determines the quality of the leaf structure, thus affecting the effective utilization of tobacco raw materials and the quality of cigarette processing. Leaf threshing refers to the process of tearing tobacco leaves from the stems using a leaf threshing machine. The resulting material is a mixture of tobacco leaves, stems, and leaves with stems. Then, an air separation process is used, where an air separator uses the different floating speeds of the various materials in a uniform upward airflow to separate the tobacco leaves from the mixture. Through multiple threshing and air separation processes, tobacco leaves with a reasonable structure can be obtained.

[0004] In the leaf threshing and air-sorting process, the process parameters of threshing and air-sorting have a close impact on the structure and quality of tobacco leaves. To obtain the optimal process parameters for threshing and air-sorting, orthogonal experiments are commonly designed, utilizing methods such as intuitive analysis, analysis of variance, SSR test, and factor contribution analysis to derive the optimal combination of threshing and air-sorting parameters. However, for the threshing and re-drying process, which involves a large number of combinations and complex relationships, such analytical methods require extensive experiments and repeated parameter adjustments to obtain a parameter combination with guiding significance. Therefore, existing analytical methods lead to excessively large experimental ranges, excessive computational loads, and the obtained optimal process parameters are somewhat arbitrary, resulting in poor actual optimization effects. Summary of the Invention

[0005] To address the aforementioned issues, this application provides a method and apparatus for optimizing leaf-splitting process parameters based on a GA-BP neural network.

[0006] In a first aspect, embodiments of this application provide a method for optimizing leaf-splitting process parameters based on a GA-BP neural network, the method comprising:

[0007] Select the leaf-beating and air-separation process parameters to be optimized, obtain sample data, and normalize the sample data. The sample data includes historical corresponding data between each leaf-beating and air-separation process parameter and the blade structure index. The blade structure index includes the large and medium leaf rate, the fragmentation rate, and the stem content in the leaf. The leaf-beating and air-separation process parameters include at least one leaf-beating machine speed and at least one air-separation fan speed.

[0008] A BP neural network is constructed using the leaf splitting process parameters as input variables and the blade structure index as output variables, and a first preset number of prediction arrays are randomly generated, the prediction arrays including initial weights and thresholds;

[0009] Based on the BP neural network, the predicted values ​​of the leaf structure indicators corresponding to the predicted data are predicted. The predicted difference between the predicted values ​​of the leaf structure indicators and the actual values ​​of the leaf structure indicators corresponding to the normalized sample data is calculated. A GA genetic algorithm fitness function is set in the BP neural network. The fitness value of the GA genetic algorithm fitness function is inversely proportional to the magnitude of the prediction difference.

[0010] Based on the fitness function of the GA genetic algorithm, target individuals that meet the preset fitness criteria are determined, and the target initial weights and target thresholds of the target individuals are assigned to the BP neural network.

[0011] Based on the number of the leaf-splitting process parameters, an orthogonal array is selected for orthogonal experimentation. Range analysis is performed on the orthogonal experiment results to determine the first optimal combination of process parameters.

[0012] Select a predicted process parameter value based on the range of process parameter values ​​corresponding to the first optimal process parameter combination, calculate the predicted process parameter value based on the assigned BP neural network, obtain a second optimal process parameter combination, and optimize the leaf-beating and air-separating process parameters of the leaf-beating machine and the air-separating fan based on the second optimal process parameter combination.

[0013] Secondly, embodiments of this application provide a device for optimizing leaf-beating and air-dividing process parameters based on a GA-BP neural network, the device comprising:

[0014] The selection module is used to select the leaf-beating and air-separation process parameters to be optimized, obtain sample data, and normalize the sample data. The sample data includes historical corresponding data between each leaf-beating and air-separation process parameter and the blade structure index. The blade structure index includes the large and medium leaf rate, the fragment rate, and the stem content in the leaf. The leaf-beating and air-separation process parameters include at least one leaf-beating machine speed and at least one air-separation fan speed.

[0015] The construction module is used to construct a BP neural network with the leaf splitting process parameters as input variables and the blade structure index as output variables, and randomly generate a first preset number of prediction arrays, wherein the prediction arrays include initial weights and thresholds;

[0016] The calculation module is used to predict the predicted value of the leaf structure index corresponding to the predicted data based on the BP neural network, calculate the prediction difference between the predicted value of the leaf structure index and the actual value of the leaf structure index corresponding to the normalized sample data, and set the fitness function of the GA genetic algorithm in the BP neural network. The fitness value of the GA genetic algorithm fitness function is inversely proportional to the magnitude of the prediction difference.

[0017] The determination module is used to determine target individuals that meet the preset fitness criteria based on the fitness function of the GA genetic algorithm, and to assign the target initial weights and target thresholds of the target individuals to the BP neural network;

[0018] The orthogonal module is used to select an orthogonal table to conduct orthogonal experiments based on the number of leaf-splitting process parameters, perform range analysis on the orthogonal experiment results, and determine the first optimal combination of process parameters.

[0019] The optimization module is used to select a predicted process parameter value according to the range of process parameter values ​​corresponding to the first optimal process parameter combination, calculate the predicted process parameter value based on the assigned BP neural network, obtain a second optimal process parameter combination, and optimize the leaf-beating and air-separating process parameters of the leaf-beating machine and the air-separating fan based on the second optimal process parameter combination.

[0020] Thirdly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method provided as in the first aspect or any possible implementation of the first aspect.

[0021] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method provided as in the first aspect or any possible implementation thereof.

[0022] The beneficial effects of this invention are as follows: 1. By designing orthogonal experiments with multiple factors and using range analysis, the primary and secondary factors affecting the experimental results can be effectively determined, thereby determining the optimal combination of process parameters. A small range of process parameter values ​​are selected near the optimal combination of process parameters, and BP neural network prediction is performed. This greatly reduces the amount of experimentation and allows the BP neural network to make targeted predictions, thereby helping to solve the technical problem in the prior art that a large number of experiments are required to obtain the optimal combination of process parameters, which is inefficient.

[0023] 2. After being trained on existing sample data, the BP neural network can predict the blade structure corresponding to various combinations of process parameters, thereby reducing the amount of testing, lowering testing costs, and improving testing efficiency.

[0024] 3. By using a genetic algorithm to optimize the initial weights and thresholds of the BP neural network, the parameter configuration of the BP neural network can be optimized, thereby reducing test errors, improving the stability and accuracy of predictions, and helping to improve the efficiency of obtaining the optimal combination of process parameters.

[0025] 4. Using the prediction results of the trained BP neural network as the fitness value of the genetic algorithm, the genetic algorithm is used to perform optimization calculations again to find the global optimal solution within the selected small range of process parameters, so as to obtain the optimal combination of process parameters, which can effectively improve the prediction accuracy of the model. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 A flowchart illustrating a method for optimizing leaf-beating air separation process parameters based on a GA-BP neural network, provided in an embodiment of this application;

[0028] Figure 2 A schematic diagram of a leaf-beating air separation process parameter optimization device based on a GA-BP neural network provided in this application embodiment;

[0029] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0030] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0031] In the following description, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. The following description provides multiple embodiments of this application, which can be substituted or combined with each other. Therefore, this application can also be considered to include all possible combinations of the same and / or different embodiments described. Thus, if one embodiment includes features A, B, and C, and another embodiment includes features B and D, then this application should also be considered to include embodiments containing one or more other possible combinations of A, B, C, and D, even if such embodiments are not explicitly described in the following text.

[0032] The following description provides examples and does not limit the scope, applicability, or examples set forth in the claims. Changes may be made to the function and arrangement of the described elements without departing from the scope of this application. Various processes or components may be appropriately omitted, substituted, or added to the examples. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Furthermore, features described with respect to some examples may be combined into other examples.

[0033] See Figure 1 , Figure 1 This is a flowchart illustrating a method for optimizing leaf-beating air separation process parameters based on a GA-BP neural network, as provided in an embodiment of this application. In this embodiment, the method includes:

[0034] S101. Select the leaf-beating and air-separation process parameters to be optimized, obtain sample data, and normalize the sample data. The sample data includes historical corresponding data between each leaf-beating and air-separation process parameter and the blade structure index. The blade structure index includes the large and medium leaf rate, the fragmentation rate, and the stem content in the leaf. The leaf-beating and air-separation process parameters include at least one leaf-beating machine speed and at least one air-separation fan speed.

[0035] The entity executing this application may be a cloud server.

[0036] In one embodiment of this application, the percentage of large and medium-sized leaves, the fragmentation rate, and the stem content in the leaves, which can basically reflect the leaf-splitting effect, are selected as evaluation indicators for the blade structure produced by leaf-splitting. At least one leaf-splitting machine speed and at least one air-classifying fan speed are selected as leaf-splitting process parameters. Furthermore, the cloud server will obtain a certain amount of sample data from a historical database. This sample data includes historical correspondence data between each leaf-splitting process parameter and the blade structure indicators. Based on this historical correspondence data, it is possible to determine the corresponding blade structure indicators for each specific value of each leaf-splitting process parameter, which will then serve as the data basis for training judgment in subsequent training processes. After obtaining the sample data, considering that the magnitude of the sample input can easily cause numerical problems, and to avoid unequal numerical problems caused by polarization of input and weights, the sample data will be normalized for ease of subsequent calculations. Through normalization, the numerical value of the sample data is stabilized within the [0, 1] interval. Similarly, after the model is subsequently built, the output result will also be inversely normalized before outputting the result.

[0037] The normalized mathematical model is:

[0038]

[0039] Among them, X K and Y K These are the original data and the normalized data, respectively; X MAX and X MIN These represent the maximum and minimum values ​​of the original data, respectively; Y MAX and Y MIN These are the endpoint values ​​of the interval containing the normalized data.

[0040] In one possible implementation, selecting the leaf-splitting process parameters to be optimized includes:

[0041] The relationship weights between each process parameter and blade structure index in the leaf-beating and air-separation process are obtained from the preset process flow database. Based on the weight sorting from large to small, a second preset number of leaf-beating machine process parameters and a third preset number of air-separation fan process parameters are selected as the leaf-beating and air-separation process parameters to be optimized.

[0042] In one embodiment of this application, since there are many blade trimming air separation process parameters affecting blade structure indicators, the staff will first conduct orthogonal experiments to analyze the relationship between process parameters and blade structure indicators from historical blade trimming air separation data, constructing a process flow database to determine the weight ranking of the influence of process parameters on blade structure indicators. Then, by screening out process parameters with relatively large weights that significantly affect blade structure, the number of research experiments can be greatly reduced while ensuring the reliability of blade indicators, ensuring the selection of reasonable optimal process parameters. Through experiments, in this example, seven blade trimming air separation process parameters were selected, including the speeds of the first, second, third, fourth, and fifth blade trimmers, the speed of the seventh-grade air separation fan, and the speed of the eighth-grade air separation fan. After the blade trimming air separation process parameters are determined, the corresponding normalized sample data can be seen in Table 1:

[0043]

[0044]

[0045] Table 1 Normalized Sample Data

[0046] S102. Construct a BP neural network with the leaf splitting process parameters as input variables and the blade structure index as output variables, and randomly generate a first preset number of prediction arrays, the prediction arrays including initial weights and thresholds.

[0047] In one embodiment of this application, to select the optimal combination of process parameters and reduce the number of experiments, the cloud server constructs an input layer using leaf splitting process parameters as input variables and an output layer using leaf structure indicators such as the large and medium leaf ratio, fragmentation rate, and stem content in the leaf as output variables. A backpropagation (BP) neural network is then constructed to predict the leaf structure indicators corresponding to different process parameters. The BP neural network is a multi-layer feedforward network trained using the backpropagation algorithm. It continuously adjusts the network's weights and thresholds through gradient descent backpropagation to minimize the sum of squared errors, thereby fitting a nonlinear mapping relationship between the input and output data. Thus, by fitting the nonlinear mapping relationship between process parameters and leaf structure indicators using the BP neural network, accurate prediction of leaf structure indicators corresponding to other process parameters can be achieved, and the optimal combination of process parameters can be selected based on the prediction results.

[0048] After constructing the BP neural network, it is necessary to determine the initial weight coefficients and thresholds for different connections between layers, as well as the activation functions of neurons, to facilitate the formation of a fitting relationship. For example, the activation function for the hidden layers can be the tansig function, the output layer activation function can be the linear function purelin, the training function can be the trainbfg function, the number of iterations maxgen can be 100, the minimum training error can be set to 0.0001, and the learning rate can be 0.01. A newff function is set between the input and hidden layers to construct the neural network and set its parameters, and a trainParam function is set between the hidden and output layers to avoid large errors caused by orders of magnitude. The sigmoid activation function of the hidden layers gives the neural network non-linear fitting capabilities, and the linear activation function of the output layer allows for unrestricted output, thereby improving the data fitting ability of a single hidden layer and ensuring the simplicity and effectiveness of the neural network.

[0049] However, the selection of initial weights and thresholds for a BP neural network directly affects its stability and accuracy. To improve the fault tolerance and reliability of the BP neural network and reduce prediction errors, a genetic algorithm will be used to optimize the BP neural network and select the optimal initial weights and thresholds. Therefore, a first preset number of prediction arrays will be randomly generated, with each array containing initial weights and a threshold. Since the initial weights and thresholds of a BP neural network are generally between 0 and 1, they will typically be randomly generated within the range [0,1].

[0050] In one possible implementation, the BP neural network has P hidden layer nodes, wherein... K is the adjustment parameter, N is the second preset quantity, and M is the third preset quantity.

[0051] In one embodiment of this application, generally 10 hidden layer neurons are set to achieve a relatively low prediction error. In other embodiments, other numbers of hidden layer nodes can be selected, and the number of hidden layers is not specifically limited. After the layers of the BP neural network are determined, the neurons in adjacent layers are fully connected to form a feedforward neural network model with a certain hierarchical structure.

[0052] In one possible implementation, the initial weights and thresholds in the prediction array are represented using any one of binary, Gray code, and floating-point encoding.

[0053] In one embodiment of this application, since genetic algorithms cannot directly process parameters in the problem space, it is necessary to encode the problem to be solved (i.e., initial weights and thresholds) as chromosomes or individuals in the genetic space, thereby achieving the conversion from phenotype to genotype. The specific encoding format can be binary, Gray code, or floating-point encoding. Binary encoding is generally preferred because it resembles the composition of biological chromosomes, is easier to explain using biological genetic theory, and facilitates crossover and mutation in genetic operations. This is used to encode each randomly generated set of initial weights and thresholds.

[0054] S103. Based on the BP neural network, predict the predicted value of the leaf structure index corresponding to the predicted data, calculate the prediction difference between the predicted value of the leaf structure index and the actual value of the leaf structure index corresponding to the normalized sample data, set the fitness function of the GA genetic algorithm in the BP neural network, and the fitness value of the GA genetic algorithm fitness function is inversely proportional to the magnitude of the prediction difference.

[0055] In one embodiment of this application, the cloud server sets a genetic algorithm in the BP neural network. After setting the genetic algorithm, the BP neural network is essentially a GA-BP neural network. After the genetic algorithm is initialized, the cloud server selects a certain proportion of normalized sample data to train the BP neural network and performs predictions. The initial weights and thresholds of the BP neural network are set to the weights and thresholds in the aforementioned randomly generated prediction array. Based on the difference between the predicted and actual values ​​of the blade structure indicators, the error function between the prediction and the sample data can be calculated. This is used to determine the fitness function of the genetic algorithm, ensuring that the fitness value is inversely proportional to the error. This allows for the selection of initial weights and thresholds with smaller prediction errors for iteration, facilitating the acquisition of the initial weights and thresholds of the BP neural network corresponding to the minimum error between the prediction and the actual output after several iterations.

[0056] The fitness function, also known as the evaluation function, assesses the quality of each chromosome (i.e., the initial weights and thresholds). It serves as the basis for selecting high-quality chromosomes, corresponding to excellent leaf structure indicators. When selecting individuals, the genetic algorithm relies on the fitness function's quality assessment of each individual, unaffected by external information. The better the fitness of an individual, the easier it is to select. The fitness value must be inversely proportional to the error, so that a smaller error corresponds to a larger fitness value, making it easier to select. This process ultimately aims to select the optimal initial weights and thresholds, improving the prediction accuracy of the BP network.

[0057] The fitness function can be:

[0058]

[0059] Where T is the number of training samples, a is the actual output of the BP neural network, and t is the expected output of the BP neural network.

[0060] The fitness function can also be:

[0061]

[0062] Where C is The maximum value estimate is given by T, where T is the number of training samples, a is the actual output of the BP neural network, and t is the expected output of the BP neural network.

[0063] For example, in the genetic algorithm settings, the number of iterations (maxgen) is 100, the number of training iterations is 1000, the learning rate is set to 0.01, the minimum training error is set to 0.0001, the initial population size (sizepop) is 70, the crossover probability (pcross) is 0.2, and the mutation probability (pmutation) is 0.8. Taking the aforementioned leaf-splitting process parameters of 7 as an example, the BP neural network structure is a 7-10-3 three-layer network, so the encoding length is S = 7*10 + 10*3 + 7 + 3 = 110. However, considering that a large initial population size will lead to slow network convergence, while a small initial population size will reduce network training accuracy, the initial population size to be optimized is set to 70 to achieve the optimal genetic algorithm effect.

[0064] S104. Based on the fitness function of the GA genetic algorithm, target individuals that meet the preset fitness criteria are determined, and the target initial weights and target thresholds of the target individuals are assigned to the BP neural network.

[0065] In one embodiment of this application, the cloud server will pre-set an expected value as a fitness standard, and the GA genetic algorithm fitness function will iterate the population until a target individual that meets the fitness standard is obtained. The target initial weight and target threshold of the target individual will be assigned to the BP neural network, and subsequent calculations will be performed using the BP neural network assigned to the target individual.

[0066] In one possible implementation, determining the target individuals that meet the preset fitness criteria based on the GA genetic algorithm fitness function includes:

[0067] The first fitness value corresponding to each prediction array is calculated based on the fitness function of the GA genetic algorithm. The first individual with the highest first fitness value is selected to enter the next generation population, and crossover and mutation are performed on the first individual.

[0068] Repeat the steps of calculating the first fitness value corresponding to each prediction array based on the fitness function of the GA genetic algorithm, selecting the first individual with the highest first fitness value to enter the next generation population, and performing crossover and mutation on the first individual until there is a target individual whose first fitness value meets the preset fitness standard.

[0069] In one embodiment of this application, each set of weights and thresholds is compared based on fitness value. The individual with the highest fitness is selected to enter the next generation population. Then, according to the set crossover and mutation probabilities, the adapted individuals undergo adaptive crossover and mutation operations to generate new individuals, which are then passed on to the next generation. Specifically, a roulette wheel selection method is used to select the individual with the highest fitness.

[0070] In one possible implementation, after assigning the initial target weights and target thresholds of the target individual to the BP neural network, the method further includes:

[0071] The normalized sample data is randomly divided into training data and validation data. The BP neural network is trained based on the training data, and the accuracy of the prediction results of the trained BP neural network is verified based on the validation data.

[0072] In one embodiment of this application, the cloud server divides the training data and validation data in a 9:1 ratio to train the assigned BP neural network. This allows the neural network to continuously adjust the weights and thresholds to fit the relationship between process parameters and blade structural indicators. Based on this relationship, it then predicts blade structural indicators for other process parameters. Furthermore, to determine the prediction accuracy of the BP neural network optimized by the genetic algorithm, the error between the predicted values ​​and the actual values ​​in the validation data is compared to assess the prediction precision of the BP neural network.

[0073] S105. Select an orthogonal array to conduct orthogonal experiments based on the number of leaf-splitting process parameters, perform range analysis on the orthogonal experiment results, and determine the first optimal combination of process parameters.

[0074] In one embodiment of this application, the BP neural network optimized by the genetic algorithm can already achieve high-precision prediction of leaf structure indicators. To obtain leaf-splitting process parameters that optimize the proportion of large and medium-sized leaves, fragmentation rate, and stem content in the leaves, an orthogonal array will be used to arrange orthogonal experiments to obtain leaf structure indicators under different combinations of process parameters. Taking the aforementioned seven leaf-splitting process parameters as an example, an orthogonal array L18 (7 factors, 3 levels) will be used, with specific data settings shown in Tables 2 and 3.

[0075]

[0076] Table 2 Factor Levels in Orthogonal Experiments

[0077]

[0078]

[0079] Table 3. Orthogonal Experiment Parameter Design Table

[0080] After obtaining the effects of different leaf-splitting process parameters on the leaf ratio, fragmentation rate, and stem content in the leaves, it is necessary to conduct a direct analysis of the influence of different process parameters on the changes in leaf structure indicators. As shown in Table 4, a range analysis was performed on the orthogonal experimental data. K1, K2, and K3 represent the average values ​​of level 1, 2, and 3 indicators in the experiment, respectively. R is the range, and P is the variance.

[0081]

[0082] Table 4. Analysis of Extreme Variation in Film Output Rate

[0083] In range analysis, the optimal process combination for each factor is determined based on the magnitude of the mean values ​​of each indicator. A higher percentage of large and medium-sized leaves, and lower leaf fragmentation and stem content, indicates a more rational leaf structure, thus yielding the first optimal combination of process parameters. Taking Table 4 as an example, based on the relationship between the mean values ​​of the indicators, the optimal process combination for the percentage of large and medium-sized leaves is A2B1C1D2E2F2G1, with the largest range being two rotation speeds and the smallest being five rotation speeds; the optimal process combination for the leaf fragmentation and stem content is A2B1C3D2E2F3G2.

[0084] S106. Select a predicted process parameter value according to the range of process parameter values ​​corresponding to the first optimal process parameter combination, calculate the predicted process parameter value based on the assigned BP neural network, obtain a second optimal process parameter combination, and optimize the leaf-beating and air-separating process parameters of the leaf-beating machine and the air-separating fan based on the second optimal process parameter combination.

[0085] In one embodiment of this application, to obtain the relatively optimal process parameters for the percentage of stems, large and medium-sized leaves, and fragmentation rate in the leaves, the cloud server will also select predicted process parameter values ​​from the range of each parameter in the first optimal process parameter combination obtained from the range analysis of the orthogonal experimental results, i.e., the range of process parameter values. Next, the predicted process parameter values ​​are calculated using the assigned BP neural network to determine the globally optimal second optimal process parameter combination from the calculation results, and the actual process parameters of the leaf trimmer and air classifier are optimized based on the second optimal process parameter combination.

[0086] For example, based on the aforementioned orthogonal experimental range analysis results and in conjunction with Table 2, it can be seen that the first optimal process parameter combination for the large and medium blade ratio, A2B1C1D2E2F2G1, corresponds to the following process parameters: first-beat speed 493 r / min, second-beat speed 471 r / min, third-beat speed 619 r / min, fourth-beat speed 795 r / min, fifth-beat speed 794 r / min, seventh-part air classifier speed 49 Hz, and eighth-part air classifier speed 45 Hz; the first optimal process parameter combination for the fragmentation rate and the stem content in the blade, A2B1C3D2E2F3G2, corresponds to the following process parameters: first-beat speed 493 r / min, second-beat speed 471 r / min, third-beat speed 621 r / min, fourth-beat speed 795 r / min, fifth-beat speed 794 r / min, seventh-part air classifier speed 50 Hz, and eighth-part air classifier speed 46 Hz.

[0087] The specific ranges of the set process parameters are as follows: the speed range for the first blower is 483-493 r / min, the speed range for the second blower is 471-480 r / min, the speed range for the third blower is 620-621 r / min, the speed range for the fourth blower is 792-796 r / min, the speed range for the fifth blower is 794-796 r / min, the speed range for the seventh-grade air classifier is 48-49 Hz, and the speed range for the eighth-grade air classifier is 46-47 Hz.

[0088] In one possible implementation, the step of calculating the predicted process parameter values ​​based on the assigned values ​​of the BP neural network to obtain a second optimal combination of process parameters includes:

[0089] The predicted process parameter values ​​are calculated based on the assigned values ​​of the BP neural network to obtain the prediction results.

[0090] The prediction results are used as the individual fitness values ​​of the GA genetic algorithm. Based on the individual fitness values, the BP neural network is subjected to a preset number of selections, crossovers, and mutations to obtain the global optimal solution of the BP neural network. The global optimal solution is the second optimal combination of process parameters.

[0091] In one embodiment of this application, the cloud server uses the prediction results of the GA-BP neural network as the individual fitness value of the genetic algorithm, and selects individuals with high large and medium leaf ratio, low fragmentation rate and low stem content in leaves for iteration. Then, through multiple selection, crossover and mutation operations, the global optimal solution of the function model to be optimized, i.e. the second optimal combination of process parameters, is found.

[0092] Experiments show that after 1000 iterations, the proportion of large and medium-sized leaves converged to 83.42%, the proportion of stems in leaves converged to 4.4%, and the fragmentation rate converged to 7.5%. The optimal process parameter values ​​obtained by the genetic algorithm are shown in Table 5.

[0093]

[0094] Table 5. Optimization results of the genetic algorithm

[0095] For example, the BP neural network model obtained by this application after assignment and training, i.e., the second optimal combination of process parameters obtained by further genetic optimization of the GA-BP model optimized by the genetic algorithm, was used in an experiment on the leaf-cutting wind-dividing line. The results are shown in Table 6. The prediction results of GA-BP are close to the actual experimental results. The relative errors between the optimized values ​​of large and medium leaf rate, leaf stem content rate, and fragment rate and the predicted values ​​obtained by GA-BP optimization are all less than 5%, and are better than the results obtained by orthogonal experimental range analysis. This indicates that the GA-BP model has accurate predictive ability and can effectively improve the leaf yield. In addition, since orthogonal experimental design has certain errors, it is easy for error fluctuations to occur in the optimal process parameter configuration, resulting in incorrect parameter selection, which has limitations. The GA-BP model can effectively eliminate such errors and has high reliability.

[0096]

[0097] Table 6 Comparison of Film Output Rate Results

[0098] Furthermore, by combining genetic algorithms to optimize the BP neural network and assigning optimal initial weights and thresholds to the network, the stability and accuracy of predictions can be greatly improved. To verify the optimization effect of the genetic algorithm on the BP neural network, this example selects 7 random sets of data to perform prediction analysis on the BP neural network model and the GA-BP neural network model optimized by the genetic algorithm. The results are shown in Table 7. It can be seen that the determination coefficient R2 of the BP neural network is 91%, and the determination coefficient R2 of the GA-BP neural network is 98%. R2 can be used to represent the fitting accuracy of the BP neural network. The values ​​of both are close to 1, indicating that both models have high accuracy and can predict relatively accurately. However, the R2 of the GA-BP network model is 7% higher than that of the unoptimized BP network model. The error percentage of the BP network model is 0.08, and the error percentage of the GA-BP network model is 0.04, with the average error percentage decreasing by 0.04.

[0099]

[0100] Table 7 Prediction Errors of BP and GA-BP

[0101] The prediction results show that the BP neural network still has some errors in prediction. These errors may stem from errors in the experimental data or inherent flaws in the neural network model itself. Therefore, genetic algorithms can improve the fault tolerance and reliability of the BP neural network, reducing prediction errors. After optimization using genetic algorithms, the initial weights and thresholds of the BP neural network are optimized, resulting in more stable network performance and predictions that are closer to the true values.

[0102] The following will be combined with the appendix Figure 2 This paper provides a detailed description of the leaf-beating air separation process parameter optimization device based on a GA-BP neural network provided in the embodiments of this application. It should be noted that the appendix... Figure 2 The leaf-beating air separation process parameter optimization device shown is used to execute the present application. Figure 1 The methods shown in the embodiments are illustrated for ease of explanation, showing only the parts relevant to the embodiments of this application. For specific technical details not disclosed, please refer to this application. Figure 1 The example shown.

[0103] Please see Figure 2 , Figure 2 This is a schematic diagram of a leaf-beating air separation process parameter optimization device based on a GA-BP neural network provided in an embodiment of this application. Figure 2 As shown, the device includes:

[0104] Selection module 201 is used to select the leaf-beating and air-separation process parameters to be optimized, obtain sample data, and normalize the sample data. The sample data includes historical corresponding data between each leaf-beating and air-separation process parameter and the blade structure index. The blade structure index includes the large and medium leaf rate, the fragment rate, and the stem content in the leaf. The leaf-beating and air-separation process parameters include at least one leaf-beating machine speed and at least one air-separation fan speed.

[0105] The construction module 202 is used to construct a BP neural network with the leaf splitting process parameters as input variables and the blade structure index as output variables, and randomly generate a first preset number of prediction arrays, wherein the prediction arrays include initial weights and thresholds;

[0106] The calculation module 203 is used to predict the predicted value of the leaf structure index corresponding to the predicted data based on the BP neural network, calculate the prediction difference between the predicted value of the leaf structure index and the actual value of the leaf structure index corresponding to the normalized sample data, and set the fitness function of the GA genetic algorithm in the BP neural network. The fitness value of the fitness function of the GA genetic algorithm is inversely proportional to the magnitude of the prediction difference.

[0107] The determination module 204 is used to determine the target individuals that meet the preset fitness criteria based on the fitness function of the GA genetic algorithm, and to assign the target initial weights and target thresholds of the target individuals to the BP neural network;

[0108] Orthogonal module 205 is used to select an orthogonal table for orthogonal experiment based on the number of leaf-splitting process parameters, perform range analysis on the orthogonal experiment results, and determine the first optimal combination of process parameters;

[0109] The optimization module 206 is used to select a predicted process parameter value according to the range of process parameter values ​​corresponding to the first optimal process parameter combination, calculate the predicted process parameter value based on the assigned BP neural network, obtain a second optimal process parameter combination, and optimize the leaf-beating and air-separating process parameters of the leaf-beating machine and the air-separating fan based on the second optimal process parameter combination.

[0110] In one possible implementation, the selection module 201 includes:

[0111] The selection unit is used to obtain the relationship weights between each process parameter and blade structure index in the leaf-beating and air-separation process from the preset process flow database. Based on the weight sorting from large to small, it selects a second preset number of leaf-beating machine process parameters and a third preset number of air-separation fan process parameters as the leaf-beating and air-separation process parameters to be optimized.

[0112] In one possible implementation, the BP neural network has P hidden layer nodes, wherein... K is the adjustment parameter, N is the second preset quantity, and M is the third preset quantity.

[0113] In one possible implementation, the determining module 204 includes:

[0114] The first calculation unit is used to calculate the first fitness value corresponding to each of the prediction arrays based on the fitness function of the GA genetic algorithm, select the first individual with the highest first fitness value to enter the next generation population, and perform crossover and mutation on the first individual;

[0115] The repeating unit is used to repeat the steps of calculating the first fitness value corresponding to each prediction array based on the fitness function of the GA genetic algorithm, selecting the first individual with the highest first fitness value to enter the next generation population, and performing crossover and mutation on the first individual until there is a target individual whose first fitness value meets the preset fitness standard.

[0116] In one possible implementation, the determining module 204 further includes:

[0117] The training unit is used to randomly divide the normalized sample data into training data and validation data, train the BP neural network based on the training data, and verify the accuracy of the prediction results of the trained BP neural network based on the validation data.

[0118] In one possible implementation, the optimization module 206 includes:

[0119] The second calculation unit is used to calculate the predicted process parameter values ​​based on the assigned BP neural network to obtain the prediction results.

[0120] The optimization unit is used to take the prediction result as the individual fitness value of the GA genetic algorithm, and after performing a preset number of selections, crossovers and mutations on the BP neural network based on the individual fitness value, the global optimal solution of the BP neural network is obtained. The global optimal solution is the second optimal combination of process parameters.

[0121] In one possible implementation, the initial weights and thresholds in the prediction array are represented using any one of binary, Gray code, and floating-point encoding.

[0122] Those skilled in the art will clearly understand that the technical solutions of the embodiments of this application can be implemented by means of software and / or hardware. In this specification, "unit" and "module" refer to software and / or hardware capable of independently completing or cooperating with other components to complete a specific function, wherein the hardware may be, for example, a Field-Programmable Gate Array (FPGA), an Integrated Circuit (IC), etc.

[0123] Each processing unit and / or module in the embodiments of this application can be implemented by an analog circuit that implements the functions described in the embodiments of this application, or by software that executes the functions described in the embodiments of this application.

[0124] See Figure 3 It shows a schematic diagram of the structure of an electronic device according to an embodiment of this application, which can be used to implement... Figure 1 The method in the illustrated embodiment. (As shown) Figure 3 As shown, the electronic device 300 may include: at least one central processing unit 301, at least one network interface 304, user interface 303, memory 305, and at least one communication bus 302.

[0125] The communication bus 302 is used to enable communication between these components.

[0126] The user interface 303 may include a display screen and a camera. Optionally, the user interface 303 may also include a standard wired interface and a wireless interface.

[0127] The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0128] The central processing unit 301 may include one or more processing cores. The central processing unit 301 connects to various parts within the electronic device 300 using various interfaces and lines. It executes various functions of the terminal 300 and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 305, and by calling data stored in the memory 305. Optionally, the central processing unit 301 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The central processing unit 301 may integrate one or a combination of several of the following: a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), and a modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display on the screen; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the central processing unit 301 and may be implemented as a separate chip.

[0129] The memory 305 may include random access memory (RAM) or read-only memory. Optionally, the memory 305 may include a non-transitory computer-readable storage medium. The memory 305 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 305 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 305 may also be at least one storage device located remotely from the aforementioned central processing unit 301. Figure 3 As shown, the memory 305, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and program instructions.

[0130] exist Figure 3 In the illustrated electronic device 300, the user interface 303 is mainly used to provide an input interface for the user and to acquire user input data; while the central processing unit 301 can be used to call the GA-BP neural network-based leaf-beating air separation process parameter optimization application stored in the memory 305, and specifically perform the following operations:

[0131] Select the leaf-beating and air-separation process parameters to be optimized, obtain sample data, and normalize the sample data. The sample data includes historical corresponding data between each leaf-beating and air-separation process parameter and the blade structure index. The blade structure index includes the large and medium leaf rate, the fragmentation rate, and the stem content in the leaf. The leaf-beating and air-separation process parameters include at least one leaf-beating machine speed and at least one air-separation fan speed.

[0132] A BP neural network is constructed using the leaf splitting process parameters as input variables and the blade structure index as output variables, and a first preset number of prediction arrays are randomly generated, the prediction arrays including initial weights and thresholds;

[0133] Based on the BP neural network, the predicted values ​​of the leaf structure indicators corresponding to the predicted data are predicted. The predicted difference between the predicted values ​​of the leaf structure indicators and the actual values ​​of the leaf structure indicators corresponding to the normalized sample data is calculated. A GA genetic algorithm fitness function is set in the BP neural network. The fitness value of the GA genetic algorithm fitness function is inversely proportional to the magnitude of the prediction difference.

[0134] Based on the fitness function of the GA genetic algorithm, target individuals that meet the preset fitness criteria are determined, and the target initial weights and target thresholds of the target individuals are assigned to the BP neural network.

[0135] Based on the number of the leaf-splitting process parameters, an orthogonal array is selected for orthogonal experimentation. Range analysis is performed on the orthogonal experiment results to determine the first optimal combination of process parameters.

[0136] Select a predicted process parameter value based on the range of process parameter values ​​corresponding to the first optimal process parameter combination, calculate the predicted process parameter value based on the assigned BP neural network, obtain a second optimal process parameter combination, and optimize the leaf-beating and air-separating process parameters of the leaf-beating machine and the air-separating fan based on the second optimal process parameter combination.

[0137] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method. The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives, as well as magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0138] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0139] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0140] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0141] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0142] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0143] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0144] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0145] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of embodiments of this disclosure upon considering the specification and practicing the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A method for optimizing leaf-beating air separation process parameters based on GA-BP neural network, characterized in that, The method includes: Select the leaf-beating and air-separation process parameters to be optimized, obtain sample data, and normalize the sample data. The sample data includes historical corresponding data between each leaf-beating and air-separation process parameter and the blade structure index. The blade structure index includes the large and medium leaf rate, the fragmentation rate, and the stem content in the leaf. The leaf-beating and air-separation process parameters include at least one leaf-beating machine speed and at least one air-separation fan speed. A BP neural network is constructed using the leaf splitting process parameters as input variables and the blade structure index as output variables, and a first preset number of prediction arrays are randomly generated, the prediction arrays including initial weights and thresholds; Based on the BP neural network, the predicted values ​​of the leaf structure indicators corresponding to the prediction array are predicted. The prediction difference between the predicted values ​​of the leaf structure indicators and the actual values ​​of the leaf structure indicators corresponding to the normalized sample data is calculated. A GA genetic algorithm fitness function is set in the BP neural network. The fitness value of the GA genetic algorithm fitness function is inversely proportional to the magnitude of the prediction difference. Based on the fitness function of the GA genetic algorithm, target individuals that meet the preset fitness criteria are determined, and the target initial weights and target thresholds of the target individuals are assigned to the BP neural network. Based on the number of the leaf-splitting process parameters, an orthogonal array is selected for orthogonal experimentation. Range analysis is performed on the orthogonal experiment results to determine the first optimal combination of process parameters. Select a predicted process parameter value based on the range of process parameter values ​​corresponding to the first optimal process parameter combination, calculate the predicted process parameter value based on the assigned BP neural network, obtain a second optimal process parameter combination, and optimize the leaf-beating and air-separating process parameters of the leaf-beating machine and the air-separating fan based on the second optimal process parameter combination.

2. The method according to claim 1, characterized in that, The selection of the leaf-beating and air-dividing process parameters to be optimized includes: The relationship weights between each process parameter and blade structure index in the leaf-beating and air-separation process are obtained from the preset process flow database. Based on the weight sorting from large to small, a second preset number of leaf-beating machine process parameters and a third preset number of air-separation fan process parameters are selected as the leaf-beating and air-separation process parameters to be optimized.

3. The method according to claim 2, characterized in that, The BP neural network has P hidden layer nodes, where... K is the adjustment parameter, N is the second preset quantity, and M is the third preset quantity.

4. The method according to claim 1, characterized in that, The process of determining target individuals that meet preset fitness criteria based on the GA genetic algorithm fitness function includes: The first fitness value corresponding to each prediction array is calculated based on the fitness function of the GA genetic algorithm. The first individual with the highest first fitness value is selected to enter the next generation population, and crossover and mutation are performed on the first individual. Repeat the steps of calculating the first fitness value corresponding to each prediction array based on the fitness function of the GA genetic algorithm, selecting the first individual with the highest first fitness value to enter the next generation population, and performing crossover and mutation on the first individual until there is a target individual whose first fitness value meets the preset fitness standard.

5. The method according to claim 1, characterized in that, After assigning the initial target weights and target thresholds of the target individual to the BP neural network, the process further includes: The normalized sample data is randomly divided into training data and validation data. The BP neural network is trained based on the training data, and the accuracy of the prediction results of the trained BP neural network is verified based on the validation data.

6. The method according to claim 1, characterized in that, The calculation of the predicted process parameter values ​​based on the assigned values ​​by the BP neural network to obtain the second optimal combination of process parameters includes: The predicted process parameter values ​​are calculated based on the assigned values ​​of the BP neural network to obtain the prediction results. The prediction results are used as the individual fitness values ​​of the GA genetic algorithm. Based on the individual fitness values, the BP neural network is subjected to a preset number of selections, crossovers, and mutations to obtain the global optimal solution of the BP neural network. The global optimal solution is the second optimal combination of process parameters.

7. The method according to claim 1, characterized in that, The initial weights and thresholds in the prediction array are represented using any one of binary, Gray code, and floating-point encoding.

8. A device for optimizing leaf-beating and air-dividing process parameters based on a GA-BP neural network, characterized in that, The device includes: The selection module is used to select the leaf-beating and air-separation process parameters to be optimized, obtain sample data, and normalize the sample data. The sample data includes historical corresponding data between each leaf-beating and air-separation process parameter and the blade structure index. The blade structure index includes the large and medium leaf rate, the fragment rate, and the stem content in the leaf. The leaf-beating and air-separation process parameters include at least one leaf-beating machine speed and at least one air-separation fan speed. The construction module is used to construct a BP neural network with the leaf splitting process parameters as input variables and the blade structure index as output variables, and randomly generate a first preset number of prediction arrays, wherein the prediction arrays include initial weights and thresholds; The calculation module is used to predict the predicted values ​​of the leaf structure indicators corresponding to the prediction array based on the BP neural network, calculate the prediction difference between the predicted values ​​of the leaf structure indicators and the actual values ​​of the leaf structure indicators corresponding to the normalized sample data, and set the fitness function of the GA genetic algorithm in the BP neural network. The fitness value of the fitness function of the GA genetic algorithm is inversely proportional to the magnitude of the prediction difference. The determination module is used to determine target individuals that meet the preset fitness criteria based on the fitness function of the GA genetic algorithm, and to assign the target initial weights and target thresholds of the target individuals to the BP neural network; The orthogonal module is used to select an orthogonal table to conduct orthogonal experiments based on the number of leaf-splitting process parameters, perform range analysis on the orthogonal experiment results, and determine the first optimal combination of process parameters. The optimization module is used to select a predicted process parameter value according to the range of process parameter values ​​corresponding to the first optimal process parameter combination, calculate the predicted process parameter value based on the assigned BP neural network, obtain a second optimal process parameter combination, and optimize the leaf-beating and air-separating process parameters of the leaf-beating machine and the air-separating fan based on the second optimal process parameter combination.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-7.

Images