A quality management and control method and system based on a soybean meal processing online monitoring system

Abstract

The application relates to the field of artificial intelligence and soybean meal processing technology, and provides a quality control method and system based on a soybean meal processing online monitoring system. The method first collects soybean raw material basic data, planting and production environment climate data, production process data and soybean meal quality index data corresponding to each production batch, carries out data cleaning and standardization processing, and constructs a batch-to-corresponding standardized data set. Correlation analysis is carried out through the Pearson correlation coefficient combined with the grey correlation analysis method, and core characteristic variables affecting the quality of soybean meal are screened. An improved random forest and improved BP neural network fusion prediction model is constructed, the improved random forest is additionally provided with a feature weight correction structure, the BP neural network is configured with a dynamic adjustable learning rate, and iteration training and parameter optimization are completed with the core characteristics as input and the soybean meal quality index as output. The trained model is solidified to the online monitoring system, the current batch of soybean meal quality is predicted in real time, the production process parameters are corrected in a closed loop combined with the target quality, the predicted quality is stably within the preset error range, and intelligent and accurate quality control of soybean meal processing is realized.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and soybean meal processing technology, and in particular to a quality control method and system based on an online monitoring system for soybean meal processing. Background Technology

[0002] Existing online soybean meal monitoring equipment can collect and detect conventional finished product indicators such as moisture, crude protein, and crude fat in real time, replacing the manual sampling laboratory testing mode. This effectively improves the efficiency of soybean meal quality testing and avoids the problems of lag, large sampling errors, and low testing efficiency associated with manual testing, providing basic quality assurance for industrial production. However, existing online soybean meal monitoring equipment only has conventional functions such as moisture and crude protein content, resulting in a limited application scope, insufficient proactive control, difficulty in preventing quality risks in advance, and inability to achieve precise control of the processing flow. Therefore, this problem needs to be solved. Summary of the Invention

[0003] To address the aforementioned technical problems, the purpose of this application is to provide a quality control method and system based on an online monitoring system for soybean meal processing. This aims to solve the problems that existing online monitoring equipment for soybean meal only has conventional functions such as moisture and crude protein content, has a shallow application dimension, lacks proactive control, makes it difficult to avoid quality risks in advance, and cannot achieve precise control of the processing flow.

[0004] In a first aspect, embodiments of this application provide a quality control method based on an online monitoring system for soybean meal processing, the method comprising:

[0005] Collect historical production process data, basic data on soybean raw materials, planting climate data, production environment climate data, and soybean meal quality index data for qualified soybean meal. Clean and process the collected data to form a standardized dataset that corresponds one-to-one with each production batch.

[0006] The standardized dataset was subjected to feature correlation analysis using Pearson correlation coefficient combined with grey relational analysis to screen out core feature variables that are strongly correlated with soybean meal quality indicators.

[0007] A prediction model based on the fusion of an improved random forest and an improved BP neural network is constructed. The data of the core feature variables are used as input and the soybean meal quality index data are used as output. The model is iteratively trained and the hyperparameters are tuned to obtain a soybean meal quality prediction model. The output of the improved random forest is connected to the input of the BP neural network. The improved random forest is configured with a feature weight correction structure, and the BP neural network is configured with a dynamically adjustable learning rate.

[0008] The trained soybean meal quality prediction model is integrated into the online monitoring system for soybean meal processing, and real-time access to the basic data of soybean raw materials, planting climate data, production environment climate data and initial process data of the current production batch is used to predict the quality of the current batch of soybean meal.

[0009] The initial process parameters are adjusted based on the predicted soybean meal quality and the target soybean meal quality. Through model iterative prediction and closed-loop correction of process parameters, the re-predicted soybean meal quality and the target soybean meal quality are kept within a preset error range.

[0010] Furthermore, the improved random forest includes an original feature calculation structure and a feature weight correction structure. The original feature calculation structure consists of decision trees configured in parallel with multiple trees, used to output the original weight data of each feature. The feature weight correction structure is used to pre-store fixed correction coefficients for each feature and correct the original weight data of each feature before outputting it. The adjustable learning rate matching structure has built-in loss threshold levels and learning rate parameters bound to each level. The input end of the adjustable learning rate matching structure is connected to the single iteration loss data of the BP network, and the output end of the adjustable learning rate matching structure is connected to the weight update port of the BP network.

[0011] Furthermore, the step of adjusting the initial process parameters based on the predicted soybean meal quality and the target soybean meal quality, and using model iterative prediction and closed-loop correction of process parameters to ensure that the re-predicted soybean meal quality and the target soybean meal quality are within a preset error range includes:

[0012] Calculate the difference between the predicted soybean meal quality and the target soybean meal quality, and determine whether the difference is within the preset error range;

[0013] If the initial process parameters are within the preset error range, then the initial process parameters will be used as the current optimal process parameters for soybean meal processing.

[0014] If the error exceeds the preset error range, the difference is used as the gradient loss value, and the partial derivatives of each process parameter are calculated to obtain the gradient change direction of each process parameter.

[0015] Based on the gradient change direction of each process parameter, combined with the preset adjustment step size, the initial process parameters are corrected in the reverse direction of the gradient, and each parameter is constrained within the upper and lower limits of each process safety range. If the limit is exceeded, the boundary value is truncated to obtain the compliant and corrected process parameters.

[0016] The corrected process parameters are input into the soybean meal quality prediction model to predict the soybean meal quality, and the process is repeated iteratively.

[0017] Furthermore, the step of using Pearson correlation coefficient combined with grey relational analysis to perform feature correlation analysis on the standardized dataset and screening out core feature variables strongly correlated with soybean meal quality indicators includes:

[0018] Select target features from the standardized dataset; wherein the target features are those that satisfy the applicable conditions of the Pearson correlation coefficient.

[0019] Calculate the Pearson correlation coefficient between each target feature and the soybean meal quality index, and define features with a Pearson correlation coefficient greater than the first preset threshold as strongly linearly correlated features;

[0020] Grey relational analysis was used to calculate the grey relational degree between each feature and the quality index of soybean meal. Features with a grey relational degree greater than the second preset threshold were classified as nonlinear strongly correlated features.

[0021] The linearly strongly correlated features and the nonlinearly strongly correlated features are used as core feature variables.

[0022] Furthermore, the production process data includes soybean pretreatment temperature, softening time, rolling thickness, leaching temperature, leaching pressure, solvent usage, drying temperature, cooling time, and production line operating speed.

[0023] Furthermore, the basic data of the soybean raw materials includes the soybean origin, purchase batch, moisture content of the raw materials upon entry into storage, crude fat content, crude protein content, impurity content, storage duration, and storage temperature and humidity.

[0024] Furthermore, the climate data for the planting environment includes rainfall during the planting season, sunshine duration, average temperature, and temperature fluctuations.

[0025] Furthermore, the workshop temperature, humidity, and ambient air pressure during the production and processing.

[0026] Furthermore, the soybean meal quality index data includes moisture content, crude protein content, crude fat content, crude fiber content, crude ash content, urease activity, and residual oil content.

[0027] Secondly, embodiments of this application provide an online monitoring and control system for soybean meal processing, the system comprising:

[0028] The data acquisition module is used to collect historical production process data of qualified soybean meal, basic data of soybean raw materials, planting climate data, production environment climate data, and soybean meal quality index data, and to clean and process the collected data to form a standardized dataset that corresponds one-to-one with each production batch.

[0029] The correlation analysis module is used to perform feature correlation analysis on the standardized dataset using Pearson correlation coefficient combined with grey relational analysis to screen out core feature variables that are strongly correlated with soybean meal quality indicators.

[0030] A construction module is used to build a prediction model based on the fusion of an improved random forest and an improved BP neural network. The model takes the data of the core feature variables as input and the soybean meal quality index data as output. The model is iteratively trained and the hyperparameters are tuned to obtain the soybean meal quality prediction model. The output of the improved random forest is connected to the input of the BP neural network. The improved random forest is configured with a feature weight correction structure, and the BP neural network is configured with a dynamically adjustable learning rate.

[0031] The solidification module is used to solidify the trained soybean meal quality prediction model into the online monitoring system for soybean meal processing, and to access the basic data of soybean raw materials, planting climate data, production environment climate data and initial process data of the current production batch in real time to predict the quality of the current batch of soybean meal.

[0032] The adjustment module is used to adjust the initial process parameters based on the predicted soybean meal quality and the target soybean meal quality. Through model iterative prediction and closed-loop correction of process parameters, the re-predicted soybean meal quality and the target soybean meal quality are kept within a preset error range.

[0033] The beneficial effects of this invention are:

[0034] (1) This invention collects production process data, soybean raw material basic data, planting climate data and production environment climate data, covering various sources of influence on soybean meal quality and avoiding prediction deviations caused by the lack of key influencing factors.

[0035] (2) The Pearson correlation coefficient is only applicable to linear, normally distributed, and stationary data without significant fluctuations. It cannot adapt to nonlinear correlations caused by soybean production location, climate, batch fluctuations, etc. The grey relational analysis method is highly adaptable to nonlinear, small-sample, highly volatile, and unknown-distributed industrial data, but it cannot quantify the significance of linear correlations. This step uses the Pearson correlation coefficient to evaluate the degree of linear correlation of features and the grey relational degree to mine nonlinear correlation patterns. It can retain production parameters with prominent linear effects and retain key variables with nonlinear changes. It only removes invalid interference data with no obvious correlation in either linear or nonlinear dimensions, reduces invalid data interference, and improves the accuracy and computational efficiency of soybean meal quality prediction.

[0036] (3) By improving the random forest and configuring the feature weight correction structure, and improving the BP neural network and configuring the adjustable learning rate, the adaptability of the model to the fluctuation of soybean raw materials in different batches and the prediction accuracy of soybean meal quality can be greatly improved. Specifically, this application combines the pain points of soybean meal processing and production. The original random forest only relies on sample data to objectively calculate feature weights. Due to the uncertain interference of planting climate fluctuation, workshop environment disturbance, and large batch quality differences in soybean meal production, the core feature weights of key processes such as leaching temperature, rolling thickness, and solvent usage are easily underestimated, while the feature weights of random environmental noise and invalid disturbance are inflated, resulting in the model weight allocation deviating from the actual process importance. In this regard, this invention retains the advantage of random forest multi-decision tree parallel mining of data objective original weights, and adds a dedicated feature weight correction structure. Through the fixed correction coefficient preset by process experience, the original weights of each core feature are specifically corrected, accurately suppressing the interference of invalid noise in production, strengthening the core process and raw material feature weights that truly dominate the quality indicators such as crude protein content and moisture content of soybean meal, and making the model weight allocation fit the real production mechanism of soybean meal. Meanwhile, addressing the industry challenges of significant batch-to-batch variability in soybean meal raw materials, large fluctuations in operating conditions, poor iterative adaptability of traditional fixed-learning-rate BP neural networks, susceptibility to local optima, and limited prediction accuracy, this invention configures an adjustable learning rate structure for the BP neural network. This allows for adaptive matching of the optimal learning rate and updating of network weights based on the magnitude of the soybean meal quality prediction error loss in a single iteration, effectively avoiding the rigidity and poor convergence of fixed-learning-rate iterations. By improving the collaborative optimization of random forest weight correction and dynamic learning-rate BP networks, the invention addresses the weakness of traditional models in adapting to the diverse production conditions of soybean meal, enhancing the model's generalization adaptability to different soybean raw material batches and operating conditions, and significantly improving the online prediction accuracy and stability of key quality indicators for finished soybean meal products.

[0037] (4) This application relies on Pearson correlation coefficient and grey relational analysis to screen strongly correlated core features from both linear and nonlinear dimensions, eliminate redundant interference data to optimize the input data source, and then build an improved random forest with a feature weight correction structure added to the back end, combined with an improved BP neural network constructed with a built-in dynamically adjustable learning rate. After the improved random forest completes the feature weight screening, it outputs to the BP network for further computation. After sample iterative training and parameter optimization, the model is deployed in the online monitoring system to achieve real-time quality prediction. Then, the predicted soybean meal quality is compared with the target quality. When the error exceeds the standard, the gradient descent algorithm is used to iteratively optimize the process parameters in a closed loop. When the error reaches the standard, the optimal process is locked for production. The whole process forms a full-link control from data preprocessing, feature selection, fusion modeling to online prediction and process closed-loop control. It overcomes the problems of single model fitting defects and weight distortion caused by accidental data interference in the original algorithm, and the problem of fixed learning rate easily getting trapped in local optima. It can also dynamically correct process parameters in reverse according to the target quality of the finished product, effectively reduce the fluctuation range of soybean meal finished product quality, reduce raw material and energy consumption, and improve the finished product qualification rate and production control intelligence level of soybean meal processing. Attached Figure Description

[0038] To more clearly illustrate the technical solution of this application, the 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 from these drawings without creative effort.

[0039] Figure 1 This is a flowchart illustrating the quality control method based on an online monitoring system for soybean meal processing provided in this application embodiment;

[0040] Figure 2 This is a schematic diagram of the structure of the online monitoring and control system for soybean meal processing provided in the embodiments of this application. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0042] Those skilled in the art will understand that, unless explicitly stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in the specification of this application means the presence of features, integers, steps, operations, elements, modules, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, modules, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any modules and all combinations of one or more associated listed items.

[0043] Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0044] Please see Figure 1 This application provides a quality control method based on an online monitoring system for soybean meal processing, the method comprising:

[0045] S1. Collect historical production process data, basic data of soybean raw materials, planting climate data, production environment climate data, and soybean meal quality index data of qualified soybean meal, and clean and process the collected data to form a standardized dataset that corresponds one-to-one with each production batch.

[0046] S2. Using Pearson correlation coefficient combined with grey relational analysis, feature correlation analysis was performed on the standardized dataset to screen out the core feature variables that are strongly correlated with soybean meal quality indicators.

[0047] S3. Construct a prediction model based on the fusion of an improved random forest and an improved BP neural network. Use the data of the core feature variables as input and the soybean meal quality index data as output. Iterate and train the model and fine-tune the hyperparameters to obtain the soybean meal quality prediction model. The output of the improved random forest is connected to the input of the BP neural network. The improved random forest is configured with a feature weight correction structure, and the BP neural network is configured with a dynamically adjustable learning rate.

[0048] S4. Solidify the trained soybean meal quality prediction model into the online monitoring system for soybean meal processing, and access the basic data of soybean raw materials, planting climate data, production environment climate data and initial process data of the current production batch in real time to predict the quality of the current batch of soybean meal;

[0049] S5. Adjust the initial process parameters according to the predicted soybean meal quality and the target soybean meal quality. Through model iterative prediction and closed-loop correction of process parameters, make the re-predicted soybean meal quality and the target soybean meal quality within the preset error range.

[0050] In step S1, the production process data includes soybean pretreatment temperature, softening time, rolling thickness, leaching temperature, leaching pressure, solvent usage, drying temperature, cooling time, and production line operating speed. The basic data for soybean raw materials includes soybean origin, purchase batch, raw material moisture content upon storage, raw material crude fat content, raw material crude protein content, impurity content, storage time, and storage temperature and humidity. The climate data for the planting environment includes rainfall during the planting season, sunshine duration, average air temperature, and temperature fluctuations. The workshop temperature, humidity, and ambient air pressure during production and processing are also included. The soybean meal quality index data includes moisture content, crude protein content, crude fat content, crude fiber content, crude ash content, urease activity, and residual oil rate.

[0051] In step S2, it should be noted that the single correlation algorithm in the soybean meal quality prediction model has the following problems: Pearson correlation coefficient is only applicable to linear, normally distributed, and stationary data without significant fluctuations, and cannot adapt to nonlinear correlations caused by soybean origin, climate, batch fluctuations, etc.; grey relational analysis is highly adaptable to nonlinear, small-sample, highly volatile, and unknown-distribution industrial data, but cannot quantify the significance of linear correlations. This step uses Pearson correlation coefficient to evaluate the degree of linear correlation of features and grey relational degree to mine nonlinear correlation patterns. This can retain production parameters with prominent linear effects and retain key variables with nonlinear changes, while only eliminating invalid interference data with no obvious correlation in either linear or nonlinear dimensions, reducing invalid data interference and improving the accuracy and computational efficiency of soybean meal quality prediction.

[0052] In step S3, this application improves the random forest by configuring a feature weight correction structure and improves the BP neural network by configuring an adjustable learning rate. This significantly improves the model's adaptability to fluctuating conditions of different batches of soybean raw materials and the accuracy of predicting the quality of soybean meal. Specifically, this application addresses the pain points in soybean meal processing and production. The original random forest only objectively calculates feature weights based on sample data. Due to uncertain interferences such as fluctuations in planting climate, workshop environmental disturbances, and large differences in the quality of raw material batches, the core feature weights of key processes such as leaching temperature, rolling thickness, and solvent usage are easily underestimated, while the feature weights of random environmental noise and invalid disturbances are inflated, causing the model weight allocation to deviate from the actual importance of the process. To address this, this invention retains the advantage of random forests in parallel mining of data with multiple decision trees and adds a dedicated feature weight correction structure. Through fixed correction coefficients preset by process experience, the original weights of each core feature are specifically corrected. This accurately suppresses invalid noise interference in production and strengthens the core process and raw material feature weights that truly dominate the quality indicators of soybean meal, such as crude protein content and moisture content, so that the model weight allocation fits the actual production mechanism of soybean meal. Meanwhile, addressing the industry challenges of significant batch-to-batch variability in soybean meal raw materials, large fluctuations in operating conditions, poor iterative adaptability of traditional fixed-learning-rate BP neural networks, susceptibility to local optima, and limited prediction accuracy, this invention configures an adjustable learning rate structure for the BP neural network. This allows for adaptive matching of the optimal learning rate and updating of network weights based on the magnitude of the soybean meal quality prediction error loss in a single iteration, effectively avoiding the rigidity and poor convergence of fixed-learning-rate iterations. By improving the collaborative optimization of random forest weight correction and dynamic learning-rate BP networks, the invention addresses the weakness of traditional models in adapting to the diverse production conditions of soybean meal, enhancing the model's generalization adaptability to different soybean raw material batches and operating conditions, and significantly improving the online prediction accuracy and stability of key quality indicators for finished soybean meal products.

[0053] In the embodiments of this application, the target soybean meal quality is at least qualified quality.

[0054] This invention collects production process data, basic soybean raw material data, planting climate data, and production environment climate data, covering all sources affecting soybean meal quality and avoiding prediction bias caused by the lack of key influencing factors. By leveraging Pearson correlation coefficient and grey relational analysis, strong correlation core features are screened from both linear and nonlinear dimensions to eliminate redundant and interfering data and optimize the input data source. A fusion prediction model is then built, incorporating an improved random forest with a feature weight correction structure and an improved BP neural network with a built-in dynamically adjustable learning rate. After feature weighting and screening using the improved random forest, the model is output to the BP network for further computation. Following iterative training and parameter optimization, the model is deployed in an online monitoring system for real-time quality prediction. The predicted soybean meal quality is then compared with the target quality. When the error exceeds the standard, the gradient descent algorithm is used for closed-loop iterative optimization of process parameters. Once the error reaches the standard, the optimal process is locked for production. The entire process, from data preprocessing, feature optimization, fusion modeling to online prediction and closed-loop process control, forms a full-link management system. This overcomes the limitations of single-model fitting and the problems of weight distortion caused by accidental data interference and the tendency of fixed learning rates to fall into local optima in native algorithms. Furthermore, it can dynamically correct process parameters based on the target quality of the finished product, effectively reducing the fluctuation range of soybean meal quality, minimizing raw material and energy consumption, and improving the finished product qualification rate and the level of intelligent production management in soybean meal processing.

[0055] In one embodiment, after obtaining process parameters that meet the preset error range, the corresponding process parameters are sent to the corresponding execution equipment to process soybeans in order to obtain qualified soybean meal.

[0056] In one embodiment, the training loss function of the prediction model is:

[0057]

[0058] Where L is the total model loss. , These are the measured and predicted values ​​for various soybean meal indicators; Weights are preset for each indicator, where K is the number of soybean meal quality indicators. Here, k and m represent the number of key indicators, and both are index values. It is the allowable error threshold for key indicators. , These are the measured and predicted values ​​for each key indicator. This is a penalty item for exceeding the limit of key indicators, and it only applies to key indicators. In this example where the soybean meal quality indicators include moisture content, crude protein content, crude fat content, crude fiber content, crude ash content, urease activity, and residual oil rate, crude protein content, residual oil rate, and urease activity are key indicators, and their priority is higher than other indicators. The weight of these key indicators is higher than that of other indicators.

[0059] This application's embodiments differentiate weights for various soybean meal quality indicators, strengthen the training priority of key quality indicators, and impose additional penalties on errors exceeding thresholds. This forces the model training process to prioritize convergence to key quality indicators, effectively solving the problems of traditional loss functions failing to distinguish the importance of indicators and insufficient prediction accuracy for core quality control items. This significantly improves the practicality and reliability of the soybean meal quality prediction model. It should be noted that key indicators play a crucial role in product quality grading.

[0060] In one embodiment, the improved random forest includes an original feature calculation structure and a feature weight correction structure. The original feature calculation structure consists of decision trees configured in parallel with multiple trees, used to output the original weight data of each feature. The feature weight correction structure is used to pre-store fixed correction coefficients for each feature and correct the original weight data of each feature before outputting it. The adjustable learning rate matching structure has built-in loss threshold levels and learning rate parameters bound to each level. The input of the adjustable learning rate matching structure is connected to the single iteration loss data of the BP network, and the output of the adjustable learning rate matching structure is connected to the weight update port of the BP network.

[0061] This application improves the random forest model by configuring a feature weight correction structure and modifying the BP neural network with an adjustable learning rate. This significantly enhances the model's adaptability to fluctuating conditions of different batches of soybean raw materials and improves the accuracy of predicting the quality of finished soybean meal. Specifically, this application addresses the pain points in soybean meal processing. The original random forest model only objectively calculates feature weights based on sample data. Due to uncertainties in soybean meal production, such as fluctuations in planting climate, workshop environmental disturbances, and large differences in the quality of raw material batches, the weights of key process features such as leaching temperature, rolling thickness, and solvent usage are easily underestimated, while the weights of random environmental noise and invalid disturbances are inflated, causing the model's weight allocation to deviate from the actual importance of the process. To address this, this invention retains the advantage of random forests in parallel mining of objective original weights using multiple decision trees, and adds a dedicated feature weight correction structure. Through fixed correction coefficients preset by process experience, the original weights of each core feature are specifically corrected. This accurately suppresses invalid noise interference in production and strengthens the weights of core process and raw material features that truly dominate quality indicators such as crude protein content and moisture content in soybean meal, allowing the model's weight allocation to conform to the actual production mechanism of soybean meal. Meanwhile, addressing the industry challenges of significant batch-to-batch variability in soybean meal raw materials, large fluctuations in operating conditions, poor iterative adaptability of traditional fixed-learning-rate BP neural networks, susceptibility to local optima, and limited prediction accuracy, this invention configures a dynamically adjustable learning rate structure with tiered loss thresholds for the BP neural network. This allows for adaptive matching of the optimal learning rate and updating of network weights based on the magnitude of the soybean meal quality prediction error loss in a single iteration, effectively avoiding the rigidity and poor convergence of fixed-learning-rate iterations. By improving the collaborative optimization between random forest weight correction and the dynamic learning-rate BP network, the invention addresses the weakness of traditional models in adapting to varying soybean meal production conditions, enhancing the model's generalization adaptability to different soybean raw material batches and operating conditions, and significantly improving the online prediction accuracy and stability of key quality indicators for finished soybean meal products.

[0062] In one embodiment, the step of adjusting the initial process parameters based on the predicted soybean meal quality and the target soybean meal quality, and using model iterative prediction and closed-loop correction of process parameters to ensure that the re-predicted soybean meal quality and the target soybean meal quality are within a preset error range includes:

[0063] Calculate the difference between the predicted soybean meal quality and the target soybean meal quality, and determine whether the difference is within the preset error range;

[0064] If the initial process parameters are within the preset error range, then the initial process parameters will be used as the current optimal process parameters for soybean meal processing.

[0065] If the error exceeds the preset error range, the difference is used as the gradient loss value, and the partial derivatives of each process parameter are calculated to obtain the gradient change direction of each process parameter.

[0066] Based on the gradient change direction of each process parameter, combined with the preset adjustment step size, the initial process parameters are corrected in the reverse direction of the gradient, and each parameter is constrained within the upper and lower limits of each process safety range. If the limit is exceeded, the boundary value is truncated to obtain the compliant and corrected process parameters.

[0067] The corrected process parameters are input into the soybean meal quality prediction model to predict the soybean meal quality, and the process is repeated iteratively.

[0068] In this embodiment, the step of correcting the initial process parameters in the opposite direction of the gradient based on the gradient change direction of each process parameter and in combination with a preset adjustment step size includes: for each process parameter, correcting the corresponding initial process parameter according to the following formula to obtain the corrected process parameter: ;in, These are the corrected process parameters. For initial process parameters, To preset the adjustment step size, The direction of gradient change of process parameters

[0069] This application's embodiments achieve closed-loop optimization of process parameters by comparing the error between the predicted soybean meal quality and the target soybean meal quality. When the error meets the standard, the existing process parameters are directly used for production. When the error exceeds the standard, the gradient descent algorithm is used to construct the loss based on the quality error, calculate the gradient of the process parameters, and correct the parameters in reverse. At the same time, the parameter adjustment range is limited by the process safety threshold to avoid the parameters exceeding the limit and deviating from the actual production conditions. After repeated iterations, the parameters converge to the acceptable error range. Under the premise of ensuring the compliance and feasibility of the process parameters, the optimal process parameters that are adapted to the target quality are quickly found. This realizes the adaptive control of the production process driven by the target product quality, stabilizes the quality of the finished soybean meal, and reduces the cost of production trial and error.

[0070] In one embodiment, the step of using Pearson correlation coefficient combined with grey relational analysis to perform feature correlation analysis on the standardized dataset and screening out core feature variables strongly correlated with soybean meal quality indicators includes:

[0071] Select target features from the standardized dataset; wherein the target features are those that satisfy the applicable conditions of the Pearson correlation coefficient.

[0072] Calculate the Pearson correlation coefficient between each target feature and the soybean meal quality index, and define features with a Pearson correlation coefficient greater than the first preset threshold as strongly linearly correlated features;

[0073] Grey relational analysis was used to calculate the grey relational degree between each feature and the quality index of soybean meal. Features with a grey relational degree greater than the second preset threshold were classified as nonlinear strongly correlated features.

[0074] The linearly strongly correlated features and the nonlinearly strongly correlated features are used as core feature variables.

[0075] In this application embodiment, the Pearson correlation coefficient is mainly used to measure the strength and direction of the linear relationship between two variables. Its applicable conditions mainly include the following aspects: (1) Both variables are continuous data: that is, the values ​​of the two variables can be any real numbers. (2) The data satisfies a normal distribution: the Pearson correlation coefficient requires that the population from which the sample comes should be normally distributed or close to normally distributed. In practical applications, if the sample size is large enough (usually considered to be greater than 30), even if the data deviates slightly from normality, it will not have a significant impact on the results. (3) There is a linear relationship between the variables: that is, the trend of change between the two variables presents a linear correlation, rather than a curve or other nonlinear pattern. (4) The two variables are independent of each other: each observation should be independent and not affected by other observations. (5) There are no obvious outliers: there should be no extreme values ​​in the dataset that deviate significantly from the normal range, because these outliers may have a significant impact on the Pearson correlation coefficient, leading to distorted results.

[0076] In this embodiment, the Pearson correlation coefficient is used to assess the degree of linear correlation of features, and the grey relational degree is used to mine nonlinear correlation patterns. A union screening rule is adopted in which any indicator that meets the standard is selected. This can retain production parameters with prominent linear effects and key variables with nonlinear changes. Only invalid interference data with no obvious correlation in linear and nonlinear dimensions are removed, thereby reducing invalid data interference and improving the accuracy and computational efficiency of soybean meal quality prediction.

[0077] Example 2:

[0078] like Figure 2 As shown in the figure, this application provides an online monitoring and control system for soybean meal processing, the system comprising:

[0079] The data acquisition module 1 is used to collect historical production process data of qualified soybean meal, basic data of soybean raw materials, planting climate data, production environment climate data and soybean meal quality index data, and clean and process the collected data to form a standardized dataset corresponding to each production batch.

[0080] Correlation analysis module 2 is used to perform feature correlation analysis on the standardized dataset using Pearson correlation coefficient combined with grey relational analysis to screen out core feature variables that are strongly correlated with soybean meal quality indicators.

[0081] Module 3 is used to construct a prediction model based on the fusion of an improved random forest and an improved BP neural network. It takes the data of the core feature variables as input and the soybean meal quality index data as output, and iteratively trains and optimizes the model to obtain the soybean meal quality prediction model. The output of the improved random forest is connected to the input of the BP neural network. The improved random forest is configured with a feature weight correction structure, and the BP neural network is configured with a dynamically adjustable learning rate.

[0082] The solidification module 4 is used to solidify the trained soybean meal quality prediction model into the soybean meal processing online monitoring system, and to access the basic data of soybean raw materials, planting climate data, production environment climate data and initial process data of the current production batch in real time to predict the quality of the current batch of soybean meal.

[0083] Adjustment module 5 is used to adjust the initial process parameters based on the predicted soybean meal quality and the target soybean meal quality. Through model iterative prediction and closed-loop correction of process parameters, the re-predicted soybean meal quality and the target soybean meal quality are within a preset error range.

[0084] It should be noted that the technical solutions in the embodiments of this specification, if involving the processing of personal information, will all be processed under the premise of having a legal basis (such as obtaining the consent of the personal information subject), and will only be processed within the scope stipulated or agreed. The collection, storage, use, processing, transmission, provision, and presentation of related information all comply with the provisions of relevant laws and regulations, do not infringe on the privacy of others, and do not violate public order and good morals.

[0085] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in this application and in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0086] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0087] The above description is only a preferred embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural changes made based on the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A quality control method based on an online monitoring system for soybean meal processing, characterized in that, The method includes: Collect historical production process data, basic data on soybean raw materials, planting climate data, production environment climate data, and soybean meal quality index data for qualified soybean meal. Clean and process the collected data to form a standardized dataset that corresponds one-to-one with each production batch. The standardized dataset was subjected to feature correlation analysis using Pearson correlation coefficient combined with grey relational analysis to screen out core feature variables that are strongly correlated with soybean meal quality indicators. A prediction model based on the fusion of an improved random forest and an improved BP neural network is constructed. The data of the core feature variables are used as input and the soybean meal quality index data are used as output. The model is iteratively trained and the hyperparameters are tuned to obtain a soybean meal quality prediction model. The output of the improved random forest is connected to the input of the BP neural network. The improved random forest is configured with a feature weight correction structure, and the BP neural network is configured with a dynamically adjustable learning rate. The trained soybean meal quality prediction model is integrated into the online monitoring system for soybean meal processing, and real-time access to the basic data of soybean raw materials, planting climate data, production environment climate data and initial process data of the current production batch is used to predict the quality of the current batch of soybean meal. The initial process parameters are adjusted based on the predicted soybean meal quality and the target soybean meal quality. Through model iterative prediction and closed-loop correction of process parameters, the re-predicted soybean meal quality and the target soybean meal quality are kept within a preset error range.

2. The quality control method based on an online monitoring system for soybean meal processing according to claim 1, characterized in that, The improved random forest includes an original feature calculation structure and a feature weight correction structure. The original feature calculation structure consists of decision trees set up in parallel with multiple trees, used to output the original weight data of each feature. The feature weight correction structure is used to pre-store the fixed correction coefficients of each feature and correct the original weight data of each feature before outputting it. The adjustable learning rate matching structure has built-in loss threshold levels and learning rate parameters bound to each level. The input of the adjustable learning rate matching structure is connected to the single iteration loss data of the BP network, and the output of the adjustable learning rate matching structure is connected to the weight update port of the BP network.

3. The quality control method based on an online monitoring system for soybean meal processing according to claim 1, characterized in that, The step of adjusting the initial process parameters based on the predicted soybean meal quality and the target soybean meal quality, and using model iterative prediction and closed-loop correction of process parameters to ensure that the re-predicted soybean meal quality and the target soybean meal quality are within a preset error range includes: Calculate the difference between the predicted soybean meal quality and the target soybean meal quality, and determine whether the difference is within the preset error range; If the initial process parameters are within the preset error range, then the initial process parameters will be used as the current optimal process parameters for soybean meal processing. If the error exceeds the preset error range, the difference is used as the gradient loss value, and the partial derivatives of each process parameter are calculated to obtain the gradient change direction of each process parameter. Based on the gradient change direction of each process parameter, combined with the preset adjustment step size, the initial process parameters are corrected in the reverse direction of the gradient, and each parameter is constrained within the upper and lower limits of each process safety range. If the limit is exceeded, the boundary value is truncated to obtain the compliant and corrected process parameters. The corrected process parameters are input into the soybean meal quality prediction model to predict the soybean meal quality, and the process is repeated iteratively.

4. The quality control method based on the online monitoring system for soybean meal processing according to claim 1, characterized in that, The step of using Pearson correlation coefficient combined with grey relational analysis to perform feature correlation analysis on the standardized dataset and screening out core feature variables strongly correlated with soybean meal quality indicators includes: Select target features from the standardized dataset; wherein the target features are those that satisfy the applicable conditions of the Pearson correlation coefficient. Calculate the Pearson correlation coefficient between each target feature and the soybean meal quality index, and define features with a Pearson correlation coefficient greater than the first preset threshold as strongly linearly correlated features; Grey relational analysis was used to calculate the grey relational degree between each feature and the quality index of soybean meal. Features with a grey relational degree greater than the second preset threshold were classified as nonlinear strongly correlated features. The linearly strongly correlated features and the nonlinearly strongly correlated features are used as core feature variables.

5. The quality control method based on an online monitoring system for soybean meal processing according to claim 1, characterized in that, The production process data includes soybean pretreatment temperature, softening time, rolling thickness, leaching temperature, leaching pressure, solvent usage, drying temperature, cooling time, and production line operating speed.

6. The quality control method based on an online monitoring system for soybean meal processing according to claim 1, characterized in that, The basic data of the soybean raw materials includes the soybean origin, purchase batch, moisture content of the raw materials upon entry into storage, crude fat content of the raw materials, crude protein content of the raw materials, impurity content, storage duration, and storage temperature and humidity.

7. The quality control method based on an online monitoring system for soybean meal processing according to claim 1, characterized in that, The climate data for the planting environment includes rainfall during the planting season, sunshine duration, average temperature, and temperature fluctuations.

8. The quality control method based on the online monitoring system for soybean meal processing according to claim 1, characterized in that, The workshop temperature, humidity, and ambient air pressure during production and processing.

9. The quality control method based on an online monitoring system for soybean meal processing according to claim 1, characterized in that, The soybean meal quality indicators include moisture content, crude protein content, crude fat content, crude fiber content, crude ash content, urease activity, and residual oil content.

10. An online monitoring and control system for soybean meal processing, characterized in that, The system includes: The data acquisition module is used to collect historical production process data of qualified soybean meal, basic data of soybean raw materials, planting climate data, production environment climate data, and soybean meal quality index data, and to clean and process the collected data to form a standardized dataset that corresponds one-to-one with each production batch. The correlation analysis module is used to perform feature correlation analysis on the standardized dataset using Pearson correlation coefficient combined with grey relational analysis to screen out core feature variables that are strongly correlated with soybean meal quality indicators. A construction module is used to build a prediction model based on the fusion of an improved random forest and an improved BP neural network. The model takes the data of the core feature variables as input and the soybean meal quality index data as output. The model is iteratively trained and the hyperparameters are tuned to obtain the soybean meal quality prediction model. The output of the improved random forest is connected to the input of the BP neural network. The improved random forest is configured with a feature weight correction structure, and the BP neural network is configured with a dynamically adjustable learning rate. The solidification module is used to solidify the trained soybean meal quality prediction model into the online monitoring system for soybean meal processing, and to access the basic data of soybean raw materials, planting climate data, production environment climate data and initial process data of the current production batch in real time to predict the quality of the current batch of soybean meal. The adjustment module is used to adjust the initial process parameters based on the predicted soybean meal quality and the target soybean meal quality. Through model iterative prediction and closed-loop correction of process parameters, the re-predicted soybean meal quality and the target soybean meal quality are kept within a preset error range.

Images