Bean product production workshop intelligent management method and system

By setting control points, collecting and preprocessing data, constructing characteristics and fusion models in the intelligent management method of soy products production workshop, real-time monitoring and prediction of soy products production process is achieved, and data redundancy and The problem of low production efficiency improves product quality and production efficiency.

CN119089289APending Publication Date: 2024-12-06BO LUO LUO FU SHAN RUN XIN FOOD CO LTD

Patent Information

Application Number
CN202411189362.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-28
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

In the production process of soy products, due to the many process steps and complex data, monitoring data is redundant and unable to effectively support manager scheduling, resulting in low production efficiency and difficult to guarantee product quality.

Method used

Using the intelligent management method of soy products production workshop, a virtual model is established for online monitoring and early warning by setting control points, collecting limit value data, data preprocessing, feature structure, data fusion and target object prediction, and improving the regulation efficiency of the production process.

Benefits of technology

Real-time monitoring and prediction of the soy product production process is achieved, production efficiency is improved, product quality is ensured, and abnormal situations are quickly detected through data-driven decision-making.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119089289A_ABST
    Figure CN119089289A_ABST
Patent Text Reader

Abstract

The invention relates to an intelligent management method and system for a bean product production workshop, and belongs to the technical field of bean product processing. Comprising the steps of setting production control points and selecting limit value data according to process requirements of the control points; collecting monitoring data according to the limit value data, and establishing a time sequence data table; data cleaning is completed through the similarity constraint relation of the multi-source heterogeneous data; constructing new features of original input; performing feature expansion extraction processing on a data cleaning result, and completing fusion of multi-source heterogeneous data by constructing a data fusion model; establishing a prediction model to complete prediction of the target object; and constructing a bean product production workshop management system, and collecting all data of a physical space through a virtual model for monitoring and early warning. According to the invention, the limit value data is selected for acquisition and visual display according to the process requirements, and the monitored target object is subjected to online monitoring and structure prediction, so that the production process regulation and control are completed, the production efficiency is improved, and the product quality is guaranteed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of bean product processing, and in particular relates to an intelligent management method and system for a bean product production workshop. Background Art

[0002] Soy products can be divided into traditional soy products, fermented soy products and other soy products. The production of traditional soy products basically goes through processes such as cleaning, soaking, grinding, filtering, boiling and forming. The products include soy milk, tofu, dried tofu and yuba.

[0003] In the actual production process, it is necessary to adjust the appropriate concentration according to the type of soy products to meet the corresponding soy product production needs. Therefore, monitoring of control targets such as soy milk concentration is a key factor in ensuring the quality of soy products. However, in the current production management, due to the large number of process steps, the collected monitoring data is complex and redundant. A large amount of redundant data and complex nonlinear separable space lead to high energy consumption, which cannot provide good data support for managers' scheduling and control. At the same time, the control target objects in the production and processing process are not predicted and monitored online, and the process control quantity cannot be adjusted in time, so the quality of the product cannot be guaranteed in advance, resulting in low production efficiency. Summary of the invention

[0004] In order to solve the above-mentioned problems existing in the prior art, the present invention provides an intelligent management method and system for a soy product production workshop, which selects limit data for collection and visualization according to process requirements, and performs online monitoring and structural prediction on the monitored target objects, thereby completing production process regulation, improving production efficiency and ensuring product quality.

[0005] The purpose of the present invention can be achieved by the following technical solutions: A first aspect of the present disclosure provides an intelligent management method for a bean product production workshop, comprising the following steps: Set control points: Set control points in the production process according to the production requirements of soy products, and select limit data according to the process requirements of the control points; Collect monitoring data: Set up data monitoring equipment at corresponding control points according to the type of limit data, collect monitoring data in real time through the monitoring equipment, and establish a time series data table; Data preprocessing: By setting similarity constraints on multi-source heterogeneous data in the time series data table, the internal structure of the data is associated to complete data cleaning; Feature construction: Perform feature construction on the cleaning results of the time series data table to construct new features of the original input; Data fusion: Perform feature expansion and extraction processing on the results of data cleaning, obtain the optimal classification hyperplane of the features through the SVM algorithm, build a data fusion model, and complete the fusion of multi-source heterogeneous data; Target object prediction: Integrate the constructed features with the target variable to form the final data set, establish a prediction model for the target object for training, and use the trained prediction model to complete the prediction of the target object by inputting the collected feature quantities; Display and alarm: Build a soy product production workshop management system, construct a virtual model of the soy product production workshop, collect all data of the physical space through the virtual model, predict the target object and then visualize it, set the threshold of the target object, and conduct monitoring and early warning.

[0006] Furthermore, the data preprocessing comprises the following steps: Set the time series data table to be a, the total amount of data to be P, and establish a data table set A={a1, a2, …, a P}, the data attribute set is φ ={e1, e2, …, e g}, calculate the similarity between multi-source heterogeneous data, the formula is as follows: ; In the formula, l ij Represented as multi-source heterogeneous data i With data j The similarity between φ is the sample attribute, For data i The attribute characteristics of For data j Attribute characteristics, S is the data dimension; The correlation coefficient between the data is set to , formulate the parameter threshold λ of the data and construct the constraints of the cleaning model: ; The BP neural network is used to build a data cleaning model, set the input layer data nodes and hidden layer data nodes, and build a node weight matrix.

[0007] Furthermore, the data preprocessing process further includes: Based on the constructed node weight matrix combined with the RBM energy function, the node energy function of multi-source heterogeneous data is obtained: ; In the formula, is the data node energy function, is the input layer node data, is the hidden layer node data, For node data The bias coefficient of For node data The bias coefficient of is the data weight, c and d represent the number of nodes of the input layer data and the number of nodes of the hidden layer data respectively; According to the constructed node energy function, the data distribution probability of the model input layer nodes and hidden layer nodes is tested and the data distribution probability is calculated: ; In the formula, is the data distribution probability, ε is the distribution factor, is the normalization coefficient of the data energy function; Assume that the hidden layer nodes When it is 1, the constraint conditions for obtaining the data distribution probability are: ; In the formula, is the constraint condition of data distribution probability, is a binary Sigmoid function, is the data state coefficient; After activating the hidden layer nodes of the neural network model, input Back propagation is performed on the model to complete the model training; based on the node energy anomalies and discrete distribution data captured in the training results, removal processing is performed to complete the cleaning of multi-source heterogeneous data.

[0008] Furthermore, the characteristic structure comprises the following steps: Calculate the index data, including water absorption rate, slag discharge rate and stability. The calculation formula of water absorption rate is: ; in, P 1 is the water absorption rate, m t For soaking t The quality of whole soybeans or broken soybeans after a period of time, m It is the mass of whole soybeans or broken soybeans before soaking; The calculation formula of slag discharge rate is: ; in, P 2 is the water absorption rate, m 1 is the mass of dregs after pulp-dregs separation; m 0 is the mass of raw dry beans before refining; The calculation formula of soy milk stability is: ; In the formula, P 3 is the stability of soy milk, M 1 and M 2 are the solid contents in the upper and lower layers of soy milk after it has been stationary for a certain period of time.

[0009] Furthermore, the characteristic structure also includes: Different feature construction methods are used to transform and combine the original data to characterize the relationship between different parameters: The total soaking efficiency and the total refining efficiency are calculated by the combined characteristic construction method; The total soaking efficiency is the product of soaking time and soaking temperature; The total refining efficiency is the product of refining time and refining water-bean ratio; The ratio of soaking time to grinding time and the ratio of soaking bean water ratio to grinding bean water ratio are calculated by ratio feature construction method: The soybean varieties and filtering methods are converted into numerical values ​​through the category coding construction method; The interactive features of pulping and boiling, as well as soaking and filtering are calculated by the interactive feature construction method: The interaction characteristic between refining and boiling is the product of refining time and boiling time; The soaking and filtering interaction feature is the product of soaking time and filtering mode encoding; The square of the pulping temperature and the square of the soaking pH value are obtained by the polynomial feature construction method.

[0010] Furthermore, the data fusion comprises the following steps: Construct the data feature vector: Based on the time series data table of data cleaning, a multi-source heterogeneous data set is established, and the data is standardized using the mean standardization method; Extract multi-source heterogeneous data based on data standardization processing results n In the m The feature vector for attributes: ; In the formula, is the feature vector of multi-source heterogeneous data, Multi-source heterogeneous data sets The standardized results, T Record periods for data attributes; The result of feature construction is added to the extracted feature vector in the form of feature columns to realize the construction of data feature vector; Multi-source heterogeneous data fusion: The SVM algorithm is used to accelerate the convergence speed through the kernel function, control the dimension of the space, map the nonlinear vector space of data features into the linear separable space, obtain the optimal classification hyperplane, and build a data fusion model for multi-source heterogeneous data; Get the optimal hyperplane for the data u : ; In the formula, is the weighting coefficient, is the input vector, y is the feature vector set, is the output vector, K (﹒) is the kernel function, is the bias term; Based on the obtained optimal super-classification surface, the multi-source heterogeneous data is divided into multiple groups of data, and the generalization parameter is set as , establish a fusion model of multi-source heterogeneous data of SVM; Based on the determined optimal solution of the data, establish the decision function of the data , complete the model solution and perform data fusion: ; In the formula, is the data mean set, The optimal solution for data.

[0011] Furthermore, the expression of the fusion model is: ; ; In the formula, For the fusion model, , For data i , j The optimal solution of , is the optimal solution mean, As constraints, The amount of data set.

[0012] Furthermore, the target object prediction includes the following steps: According to the time series data table, the constructed data features are integrated with the results obtained by the soymilk concentration meter to build a data set, and the data set is divided into a training set and a test set; Establish a neural network prediction model, set the number of neurons in the input layer according to the number of features in the data set, and set the number of neurons in the output layer according to the prediction target object; Use the training set to train the constructed neural network model. During the training process, adjust the weights and biases to minimize the loss function. Use the test set to evaluate the performance of the model and check the generalization ability of the model. After the model training and evaluation are completed, the trained model is used to predict the target object in the soy product processing results.

[0013] Furthermore, the control points include soaking, refining, boiling and tapping; The limit data of the soaking time include pH value, temperature, bean-water ratio and soaking time of water; The limit data of the pulping include the soybean water addition ratio and the slag discharge rate; The limit data of pulp cooking include pulp cooking temperature and pulp cooking time; The limit data of soy milk point include soy milk concentration, temperature and pH value.

[0014] A second aspect of the present disclosure provides an intelligent management system for a bean product production workshop, which is applied to an intelligent management method for a bean product production workshop as described above, and includes a physical workshop module, a virtual workshop module, a data-driven module, a visual interaction module, and a workshop scheduling module; The physical workshop module includes workshop control point production entity elements, which are used to receive and execute production tasks issued by the workshop scheduling module and control instructions defined by the virtual workshop module; The virtual workshop module is used to digitally mirror the physical workshop, model the physical elements of the workshop, and complete the behavior simulation of the physical elements and the definition of rules and instructions; The data-driven module is used to process the data collected and fed back by the physical workshop module under the drive of the twin data, including data preprocessing, feature construction, data fusion and target object prediction, and visualize the final processing results and make scheduling decisions; The visualization interaction module is used to register the physical workshop and the virtual workshop in real time, obtain the workshop twin data in real time through the data transmission protocol, and visualize the workshop twin data; The workshop scheduling module is used to judge the processing results of the data driving module and generate production tasks and workshop intelligent scheduling management decisions.

[0015] The beneficial effects of the present invention are: The present invention first selects limit data according to process requirements, then sets data monitoring collection points according to the limit data type, visualizes the entity elements of the collection points and provides real-time data feedback, thereby realizing rapid detection of abnormal situations and providing support for subsequent data-driven decision-making; then, the real-time data is predictively processed through data driving, the collected time series data table is cleaned, the similarity constraint relationship of multi-source heterogeneous data is set, and the association of the internal structure of the data is realized; at the same time, the cleaned data construction features are subjected to data fusion, thereby improving the information density of the feature space. Finally, model training is performed based on the constructed data set to complete the prediction of the target object, provide data guidance for workshop scheduling management decisions, and realize the early prediction of the target object results, thereby performing process control, improving production efficiency and ensuring product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0017] Figure 1 A schematic diagram of the steps of an intelligent management method for a bean product production workshop provided in Embodiment 1 of the present invention; Figure 2 A schematic diagram of the structure of an intelligent management system for a bean product production workshop provided in the second embodiment of the present invention; Figure 3 A schematic diagram of the steps of the characteristic construction provided in the third embodiment of the present invention; Figure 4 A schematic diagram of the steps of target object prediction provided in Embodiment 4 of the present invention. DETAILED DESCRIPTION

[0018] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0019] Embodiment 1 The present application embodiment provides a method for intelligent management of a bean product production workshop. Figure 1 As shown, the following steps are included: S1. Set control points: Set control points in the production process according to the production requirements of soy products, and select limit data according to the process requirements of the control points. The control points include soaking, grinding, boiling and slurrying.

[0020] Select limit data according to the control points. For example, when soaking, the pH value, temperature, bean-water ratio and soaking time of water are used as limit data; when grinding, the ratio of beans and water added and the slag discharge rate are used as limit data; when boiling, the boiling temperature and time are used as limit data; when pulping, the soy milk concentration, temperature and pH value are used as limit data.

[0021] It is understandable that during the tofu processing, the soaking effect of dry beans has an important influence on the extraction rate of soybean protein and the quality of tofu. Before grinding, the soybeans should be soaked in water to soften the cotyledons, reduce the hardness, and break the tissues, cells and protein membranes, so that nutrients such as protein and lipids are easier to extract from the cells. The degree of water absorption by soybeans determines the dissolution rate of other nutrients such as protein and carbohydrates during grinding, which in turn affects the final tofu gel structure. At the same time, soaking makes the soybean fiber absorb water and swell, and the toughness is enhanced. After grinding and breaking, it still retains larger fragments, reduces the amount of fine fiber particles formed, and ensures that it is easier to separate and remove when the pulp is separated. Factors such as soybean variety, water quality of soaking water, water temperature of soaking water, soaking time, and bean-water ratio affect the soaking process parameters. Therefore, the pH value of water, temperature bean-water ratio and soaking time are used as limit data in the soaking process.

[0022] Refining is the process of soaking soybeans properly, putting them into the hopper of the pulping machine and adding an appropriate amount of water to break the soybean tissue, dissolve nutrients such as protein, and obtain a milky white slurry. In actual production, the soybean grinding degree should be moderate. If it is ground too finely, the fiber fragments will increase. When the pulp is separated from the residue, the small volume of fiber fragments will enter the soy milk together with the protein, affecting the protein gel network structure, resulting in a poor taste and texture of the product. At the same time, too fine fibers can easily cause the sieve holes of the centrifuge or extruder to be blocked, increase the residual protein content in the soybean dregs, affect the filtration effect, and reduce the yield. In addition, the pulping machine reduces the influence of human factors on the concentration of soybean juice by adding water and beans in a certain proportion, and lays a good foundation for the subsequent process of pulping by controlling the concentration of soybean milk. After refining, pulp separation is also required. Its main purpose is to separate and remove the soybean dregs to obtain a sol liquid with soybean protein as the main dispersed substance - soybean milk. In order to facilitate the concentration control and separation control of pulping, the ratio of beans and water added is used as the limit data in the pulping process, and the slag discharge rate is used as the limit data for pulp-slag separation.

[0023] Cooking is to fully denature the soybean protein by heating, which creates necessary conditions for pulping on the one hand, and eliminates anti-nutritional factors and trypsin inhibitors, destroys the activity of lipoxygenase, eliminates beany smell, kills bacteria, and prolongs the shelf life of the product on the other hand. In the cooking process, the cooking temperature, time, and heating method determine the cooking effect, so the above influencing factors are used as the limit data for cooking.

[0024] Soybean milk is a kind of soybean milk that is cooked. Soybean milk is a kind of soybean milk that is cooked. Soybean milk is a kind of soybean milk that is cooked. The soy ...

[0025] S2. Collect monitoring data: According to the type of limit data, data monitoring equipment is set up at the corresponding control point, and monitoring data is collected in real time through the monitoring equipment to establish a time series data table. The monitoring equipment includes quality sensors, temperature sensors, timers, pH testers, and soy milk concentration meters.

[0026] It should be noted that the time series data table includes soybean variety, soaking pH value, soaking temperature, soaking soybean-water ratio, soaking time, grinding soybean-water ratio, grinding time, filtering method, cooking temperature, cooking time and soy milk concentration.

[0027] S3, data preprocessing: by setting the similarity constraint relationship of multi-source heterogeneous data in the time series data table, the internal structure of the data is associated to complete data cleaning; specifically, the following steps are included: Set the time series data table to be a, the total amount of data to be P, and establish a data table set A={a1, a2, …, a P}, the data attribute set is φ ={e1, e2, …, e g}, calculate the similarity between multi-source heterogeneous data, the formula is as follows: ; In the formula, l ij Represented as multi-source heterogeneous data i With data j The similarity between φ is the sample attribute, For data i The attribute characteristics of For data j , S is the data dimension.

[0028] The correlation coefficient between the data is set to , formulate the parameter threshold λ of the data and construct the constraints of the cleaning model: ; Use BP neural network to build data cleaning model, set input layer data nodes and hidden layer data nodes, and build node weight matrix; Based on the constructed node weight matrix combined with the RBM (Unsupervised Boltzmann Machine, Restricted BoltzmannMachine) energy function, the node energy function of multi-source heterogeneous data is obtained: ; In the formula, is the data node energy function, is the input layer node data, is the hidden layer node data, For node data The bias coefficient of For node data The bias coefficient of is the data weight, c and d represent the number of nodes in the input layer data and the number of nodes in the hidden layer data, respectively.

[0029] According to the constructed node energy function, the data distribution probability of the model input layer nodes and hidden layer nodes is tested and the data distribution probability is calculated: ; In the formula, is the data distribution probability, ε is the distribution factor, is the normalization coefficient of the data energy function.

[0030] Assume that the hidden layer nodes When it is 1, the constraint conditions for obtaining the data distribution probability are: ; In the formula, is the constraint condition of data distribution probability, is a binary Sigmoid function, is the data state coefficient.

[0031] After activating the hidden layer nodes of the neural network model, input The model is back-propagated to complete the model training. The node energy anomalies and discrete distribution data captured in the training results are removed to complete the cleaning of multi-source heterogeneous data.

[0032] It is understandable that by setting similarity constraint relationships between multi-source heterogeneous data and realizing the correlation of the internal structure of the data, not only the speed of the later BP neural network internal data cleaning operation can be improved, but also the efficiency of data cleaning work can be improved.

[0033] S4, feature construction: construct features based on the cleaning results of the time series data table, construct new features of the original input, reduce the complexity of data feature extraction, and improve the speed of feature extraction.

[0034] S5, data fusion: perform feature extraction on the results of data cleaning, obtain the optimal classification hyperplane of the features through the SVM algorithm, build a data fusion model, and complete the fusion of multi-source heterogeneous data. Specifically, it includes the following steps: S51. Construct data feature vector: Based on the time series data table of data cleaning, a multi-source heterogeneous data set is established, and the data is standardized using the mean standardization method; Extract multi-source heterogeneous data based on data standardization processing results n In the m The feature vector for attributes: ; In the formula, is the feature vector of multi-source heterogeneous data, Multi-source heterogeneous data sets The standardized results, T Record period for data attributes.

[0035] The result of feature construction is added to the extracted feature vector in the form of feature columns to realize the construction of data feature vector.

[0036] It is understandable that by adding the results of feature construction to the extracted feature vector, the integrity of the data feature vector is improved, the influence of various factors in the process is more comprehensively described, complex relationships are captured, and the prediction accuracy of the subsequent target object is improved.

[0037] S52. Multi-source heterogeneous data fusion: The SVM algorithm is used to accelerate the convergence speed through the kernel function, control the dimension of the space, map the nonlinear vector space of data features into the linear separable space, obtain the optimal classification hyperplane, and build a data fusion model for multi-source heterogeneous data; Get the optimal hyperplane for the data u : ; In the formula, is the weighting coefficient, is the input vector, y is the feature vector set, is the output vector, K (﹒) is the kernel function, is the bias term.

[0038] Based on the obtained optimal super-classification surface, the multi-source heterogeneous data is divided into multiple groups of data, and the generalization parameter is set as , establish a fusion model of multi-source heterogeneous data using SVM J : ; ; In the formula, , For data i , j The optimal solution of , is the optimal solution mean, As constraints, The amount of data set.

[0039] It should be noted that the data set size includes the number of monitoring devices and the number of newly constructed features.

[0040] Based on the determined optimal solution of the data, establish the decision function of the data , complete the model solution and perform data fusion: ; In the formula, is the data mean set, The optimal solution for data.

[0041] It is understandable that the multi-source heterogeneous data processed by the above method is classified according to the characteristics of the data, so as to achieve data cleaning processing of the multi-source data characteristics, reduce duplicate records of multi-source heterogeneous data, quickly process missing records, and effectively improve the quality of the data.

[0042] S6. Target object prediction: Integrate the constructed features with the target variables to form the final data set, establish a prediction model for the target object for training, and use the trained prediction model to complete the prediction of the target object by inputting the collected feature quantities.

[0043] S7. Display and alarm: Build a soy product production workshop management system, construct a virtual model of the soy product production workshop, collect all data of the physical space through the virtual model, predict the target object and then visualize it, set the threshold of the target object, and conduct monitoring and early warning.

[0044] In this embodiment, the soymilk concentration is taken as the prediction target object, and a preset range of the soymilk concentration is set according to the soymilk concentration required by the product. When the prediction result exceeds the preset range, an alarm is performed.

[0045] Through the above method, firstly, the limit data is selected according to the process requirements, and then the data monitoring collection points are set according to the limit data type, so as to realize the rapid detection of abnormal situations and provide support for subsequent data-driven decision-making; then the collected time series data table is cleaned, and the similarity constraint relationship of multi-source heterogeneous data is set to realize the association of the internal structure of the data, provide a more comprehensive and in-depth data perspective, and improve the reliability of decision-making; at the same time, the cleaned data is fused through the fusion model of multi-source heterogeneous data, which improves the comprehensiveness and integrity of the data set, and before data fusion, the constructed new features are added to the feature extraction results to improve the information density of the feature space. Finally, the model training is carried out based on the constructed data set to complete the prediction of the target object, and the process monitoring data and target object prediction results of this process are visualized and monitored in real time. The status of each control point in the processing process is fed back, and the target object results are obtained in advance. Preventive process management is realized, and production efficiency and product quality are improved.

[0046] Embodiment 2 Another embodiment of the present application also provides an intelligent management system for a bean product production workshop, such as Figure 2 As shown, it includes a physical workshop module, a virtual workshop module, a data-driven module, a visual interaction module and a workshop scheduling module; The physical workshop module includes workshop control point production entity elements such as processing equipment, management and control personnel and processing raw materials, and is used to receive and execute production tasks issued by the workshop scheduling module and control instructions defined by the virtual workshop module.

[0047] In this embodiment, in order to facilitate the rapid construction of a soy product production workshop management system, modeling is performed based on the control points set during the processing process, data collection and feedback are performed on the entity elements in the control points, and the update and optimization of the virtual workshop are completed.

[0048] The virtual workshop module is used to digitally mirror the physical workshop, model the physical elements of the workshop, and complete the physical element behavior simulation and rule instruction definition.

[0049] It should be noted that the virtual workshop modeling in this embodiment includes workshop control point element modeling, workshop behavior modeling and workshop rule modeling. Element modeling determines the model structure and size according to the information in the equipment manual, uses SolidWorks software for modeling, modularizes and hierarchically models the model according to the limit data, simulates the actual motion state of each production element, and thus realizes the logical control of the model motion. Workshop behavior modeling uses an extended random high-level judgment Petri net to establish a production system model, and sets the library, transition and transition rules in the model according to the processing technology. Workshop rule modeling includes data rules and model rules, wherein the data rule refers to triggering a specific response when the workshop limit data reaches a certain condition, so as to achieve the purpose of attracting user attention or displaying a certain state. The parameters can be directly collected unprocessed data or data processed by the data-driven module (such as predicted concentration, etc.); the model rule is that when the model reaches a specific state, a specific response is triggered, for example, when the person's position is too close to the dangerous area, a danger warning will be issued to improve workshop safety.

[0050] The data-driven module is used to fuse data and predict and monitor control point target objects under the drive of twin data.

[0051] In this embodiment, the data collected and fed back by the physical workshop module is processed by the data driving module, including data preprocessing, feature construction, data fusion and target object prediction, and the final processing results are visualized and scheduling decisions are made.

[0052] The visualization interaction module is used to register the physical workshop and the virtual workshop in real time, obtain the workshop twin data in real time through the data transmission protocol, and visualize the workshop twin data.

[0053] It is understandable that for data information, the status dashboard can display the real-time and statistical information of each production resource in detail, combined with intelligent information acquisition strategies to assist users in workshop management; for workshop process information, the three-dimensional virtual scene module is used as the main visualization method to achieve visual control of the workshop manufacturing process from three aspects: logistics, equipment and products; for warning information, it attracts the attention of managers through color, voice and other means to achieve timely management and regulation.

[0054] The workshop scheduling module is used to judge the processing results of the data driving module and generate production tasks and workshop intelligent scheduling management decisions.

[0055] In this embodiment, the multi-source heterogeneous information of the workshop is modeled through the OPC UA model. First, the types, attributes and limit data of the control points involved in the workshop management process are sorted out, and the modeling information is verified according to the relevant specifications in the field of workshop information visualization to form a workshop scheduling information model.

[0056] In this embodiment, the digital twin technology is used to establish an intelligent management system for the soy product production workshop, the physical elements of the control points are visualized and the data is fed back in real time, and the real-time data is predictively processed through the data-driven module to provide data guidance for the workshop scheduling management decisions, thereby improving the response speed of production management and decision support. By predicting potential target results, process control is carried out in advance, production efficiency is improved, and product quality is guaranteed.

[0057] Embodiment 3 Another embodiment of the present application provides a method for intelligent management of a bean product production workshop. Figure 3 As shown, the characteristic structure in step S4 is specifically: S41. Calculate index data, including water absorption rate, slag discharge rate and stability, wherein the calculation formula of water absorption rate is: ; in, P 1 is the water absorption rate, m t For soaking t The quality of whole soybeans or broken soybeans after a period of time, m It refers to the mass of whole soybeans or broken soybeans before soaking.

[0058] The calculation formula of slag discharge rate is: ; in, P 2 is the water absorption rate, m 1 is the mass of dregs after pulp-dregs separation; m 0 is the mass of raw dry beans before refining.

[0059] The calculation formula of soy milk stability is: ; In the formula, P 3 is the stability of soy milk, M 1 and M 2 are the solid contents in the upper and lower layers of soy milk after it has been stationary for a certain period of time.

[0060] It should be noted that P 3The smaller it is, the more serious the stratification of soy milk is, and the worse the stability of soy milk is. After measuring the stability of soy milk at a specific concentration, it can be concluded that as the concentration of soy milk decreases, P3 value gradually decreases, that is, the stability of soy milk gradually decreases.

[0061] S42. Use different feature construction methods to transform and combine the original data to characterize the relationship between different parameters: The total soaking efficiency and the total refining efficiency are calculated by combining the characteristics; The total soaking efficiency is the product of soaking time and soaking temperature; The total refining efficiency is the product of the refining time and the refining water-bean ratio.

[0062] The ratio of soaking time to grinding time and the ratio of soaking bean water ratio to grinding bean water ratio are calculated by the ratio feature: The soaking time to refining time ratio is the ratio of the soaking time to the refining time; The ratio of soaking bean-water ratio to grinding bean-water ratio is the ratio of soaking bean-water ratio to grinding bean-water ratio.

[0063] Convert soybean varieties and filtering methods into numerical values ​​using category codes: Soybean variety code: convert soybean varieties into numerical values, for example: variety A=1, variety B=2; Filter mode encoding: Convert the filter mode to a numerical value, for example: mode 1=1, mode 2=2.

[0064] The interactive features of refining and boiling, and soaking and filtering are calculated by interactive features: The interaction characteristic between refining and boiling is the product of refining time and boiling time; The soaking and filtering interaction feature is the product of the soaking time and the filtering method encoding.

[0065] The square of the cooking temperature and the square of the soaking pH value are obtained through polynomial characteristics.

[0066] Embodiment 4 Another embodiment of the present application provides a method for intelligent management of a bean product production workshop. Figure 4 As shown, the target object prediction in step S6 is specifically: S61. According to the time series data table, the constructed data features are integrated with the results obtained by the soymilk concentration meter to construct a data set, and the data set is divided into a training set and a test set.

[0067] S62. Establish a neural network prediction model, set the number of neurons in the input layer according to the number of features in the data set, and set the number of neurons in the output layer according to the prediction target object.

[0068] In this embodiment, the concentration of soy milk is used as a single neuron output prediction target object, and the concentration of the produced soy milk is predicted in advance through the process quantity of the processing technology, so that the process quantity is adjusted according to the concentration requirement of the processed product type to achieve processing concentration control.

[0069] S63. Use the training set to train the constructed neural network model. During the training process, adjust the weights and biases to minimize the loss function (such as using mean square error). Use the test set to evaluate the performance of the model and check the generalization ability of the model.

[0070] S64. After the model training and evaluation are completed, the trained model is used to predict the target object in the soy product processing results.

[0071] The present invention first selects limit data according to process requirements, then sets data monitoring collection points according to the limit data type, visualizes the entity elements of the collection points and provides real-time data feedback, thereby realizing rapid detection of abnormal situations and providing support for subsequent data-driven decision-making; then, the real-time data is predictively processed through data driving, the collected time series data table is cleaned, the similarity constraint relationship of multi-source heterogeneous data is set, and the association of the internal structure of the data is realized; at the same time, the cleaned data construction features are subjected to data fusion, thereby improving the information density of the feature space. Finally, model training is performed based on the constructed data set to complete the prediction of the target object, provide data guidance for workshop scheduling management decisions, and realize the early prediction of the target object results, thereby performing process control, improving production efficiency and ensuring product quality.

[0072] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. An intelligent management method for a bean product production workshop, characterized in that: The following steps are involved: Set control points: Set control points in the production process according to the production requirements of soy products, and select limit data according to the process requirements of the control points; Collect monitoring data: Set up data monitoring equipment at corresponding control points according to the type of limit data, collect monitoring data in real time through the monitoring equipment, and establish a time series data table; Data preprocessing: By setting similarity constraints on multi-source heterogeneous data in the time series data table, the internal structure of the data is associated to complete data cleaning; Feature construction: Perform feature construction on the cleaning results of the time series data table to construct new features of the original input; Data fusion: Perform feature expansion and extraction processing on the results of data cleaning, obtain the optimal classification hyperplane of the features through the SVM algorithm, build a data fusion model, and complete the fusion of multi-source heterogeneous data; Target object prediction: Integrate the constructed features with the target variable to form the final data set, establish a prediction model for the target object for training, and use the trained prediction model to complete the prediction of the target object by inputting the collected feature quantities; Display and alarm: Build a soy product production workshop management system, construct a virtual model of the soy product production workshop, collect all data of the physical space through the virtual model, predict the target object and then visualize it, set the threshold of the target object, and conduct monitoring and early warning.

2. The intelligent management method for a bean product production workshop according to claim 1, characterized in that: The data preprocessing comprises the following steps: Set the time series data table to be a, the total amount of data to be P, and establish a data table set A={a1, a2, …, a P }, the data attribute set is φ ={e1, e2, …, e g }, calculate the similarity between multi-source heterogeneous data, the formula is as follows: ; In the formula, l ij Represented as multi-source heterogeneous data i With data j The similarity between φ is the sample attribute, For data i The attribute characteristics of For data j Attribute characteristics, S is the data dimension; The correlation coefficient between the data is set to , formulate the parameter threshold λ of the data and construct the constraints of the cleaning model: ; The BP neural network is used to build a data cleaning model, set the input layer data nodes and hidden layer data nodes, and build a node weight matrix.

3. The intelligent management method for a bean product production workshop according to claim 2, characterized in that: The data preprocessing process further includes: Based on the constructed node weight matrix combined with the RBM energy function, the node energy function of multi-source heterogeneous data is obtained: ; In the formula, is the data node energy function, is the input layer node data, is the hidden layer node data, For node data The bias coefficient of For node data The bias coefficient of is the data weight, c and d represent the number of nodes of the input layer data and the number of nodes of the hidden layer data respectively; According to the constructed node energy function, the data distribution probability of the model input layer nodes and hidden layer nodes is tested and the data distribution probability is calculated: ; In the formula, is the data distribution probability, ε is the distribution factor, is the normalization coefficient of the data energy function; Assume that the hidden layer nodes When it is 1, the constraint conditions for obtaining the data distribution probability are: ; In the formula, is the constraint condition of data distribution probability, is a binary Sigmoid function, is the data state coefficient; After activating the hidden layer nodes of the neural network model, input Back propagation is performed on the model to complete the model training; the node energy anomalies and discrete distribution data captured in the training results are removed to complete the cleaning of multi-source heterogeneous data.

4. The intelligent management method for a bean product production workshop according to claim 1, characterized in that: The characteristic structure comprises the following steps: Calculate the index data, including water absorption rate, slag discharge rate and stability. The calculation formula of water absorption rate is: ; in, P 1 is the water absorption rate, m t For soaking t The quality of whole soybeans or broken soybeans after a period of time, m It is the mass of whole soybeans or broken soybeans before soaking; The calculation formula of slag discharge rate is: ; in, P 2 is the water absorption rate, m 1 is the mass of dregs after pulp-dregs separation; m 0 is the mass of raw dry beans before refining; The calculation formula of soy milk stability is: ; In the formula, P 3 is the stability of soy milk, M 1 and M 2 are the solid contents in the upper and lower layers of soy milk after it has been stationary for a certain period of time.

5. The intelligent management method for a bean product production workshop according to claim 4, characterized in that: The characteristic structure also includes: Different feature construction methods are used to transform and combine the original data to characterize the relationship between different parameters: The total soaking efficiency and the total refining efficiency are calculated by the combined characteristic construction method; The total soaking efficiency is the product of soaking time and soaking temperature; The total refining efficiency is the product of refining time and refining water-bean ratio; The ratio of soaking time to grinding time and the ratio of soaking bean water ratio to grinding bean water ratio are calculated by ratio feature construction method: The soybean varieties and filtering methods were converted into numerical values ​​through the category coding construction method; The interactive features of pulping and boiling, as well as soaking and filtering are calculated by the interactive feature construction method: The interaction characteristic between refining and boiling is the product of refining time and boiling time; The soaking and filtering interaction feature is the product of soaking time and filtering mode encoding; The square of the pulping temperature and the square of the soaking pH value are obtained by the polynomial feature construction method.

6. The intelligent management method for a bean product production workshop according to claim 1, characterized in that: The data fusion comprises the following steps: Construct the data feature vector: Based on the time series data table of data cleaning, a multi-source heterogeneous data set is established, and the data is standardized using the mean standardization method; Extract multi-source heterogeneous data based on data standardization processing results n In the m The feature vector for attributes: ; In the formula, is the feature vector of multi-source heterogeneous data, Multi-source heterogeneous data sets The standardized results, T Record periods for data attributes; The result of feature construction is added to the extracted feature vector in the form of feature columns to realize the construction of data feature vector; Multi-source heterogeneous data fusion: The SVM algorithm is used to accelerate the convergence speed through the kernel function, control the dimension of the space, map the nonlinear vector space of data features into the linear separable space, obtain the optimal classification hyperplane, and build a data fusion model for multi-source heterogeneous data; Get the optimal hyperplane for the data u : ; In the formula, is the weighting coefficient, is the input vector, y is the feature vector set, is the output vector, K (﹒) is the kernel function, is the bias term; Based on the obtained optimal super-classification surface, the multi-source heterogeneous data is divided into multiple groups of data, and the generalization parameter is set as , establish a fusion model of multi-source heterogeneous data of SVM; Based on the determined optimal solution of the data, establish the decision function of the data , complete the model solution and perform data fusion: ; In the formula, is the data mean set, The optimal solution for data.

7. The intelligent management method for a bean product production workshop according to claim 6, characterized in that: The expression of the fusion model is: ; ; In the formula, For the fusion model, , For data i , j The optimal solution of , is the optimal solution mean, As constraints, The amount of data set.

8. The intelligent management method for a bean product production workshop according to claim 1, characterized in that: The target object prediction comprises the following steps: According to the time series data table, the constructed data features are integrated with the results obtained by the soymilk concentration meter to build a data set, and the data set is divided into a training set and a test set; Establish a neural network prediction model, set the number of neurons in the input layer according to the number of features in the data set, and set the number of neurons in the output layer according to the prediction target object; Use the training set to train the constructed neural network model. During the training process, adjust the weights and biases to minimize the loss function. Use the test set to evaluate the performance of the model and check the generalization ability of the model. After the model training and evaluation are completed, the trained model is used to predict the target object in the soy product processing results.

9. The intelligent management method for a bean product production workshop according to claim 1, characterized in that: The control points include soaking, refining, boiling and tapping; The limit data of the soaking time include pH value, temperature, bean-water ratio and soaking time of water; The limit data of the pulping include the soybean water addition ratio and the slag discharge rate; The limit data of pulp cooking include pulp cooking temperature and pulp cooking time; The limit data of soy milk point include soy milk concentration, temperature and pH value.

10. An intelligent management system for a bean product production workshop, applied to an intelligent management method for a bean product production workshop as claimed in any one of claims 1 to 9, characterized in that: It includes physical workshop module, virtual workshop module, data-driven module, visual interaction module and workshop scheduling module; The physical workshop module includes workshop control point production entity elements, which are used to receive and execute production tasks issued by the workshop scheduling module and control instructions defined by the virtual workshop module; The virtual workshop module is used to digitally mirror the physical workshop, model the physical elements of the workshop, and complete the behavior simulation of the physical elements and the definition of rules and instructions; The data-driven module is used to process the data collected and fed back by the physical workshop module under the drive of the twin data, including data preprocessing, feature construction, data fusion and target object prediction, and visualize the final processing results and make scheduling decisions; The visualization interaction module is used to register the physical workshop and the virtual workshop in real time, obtain the workshop twin data in real time through the data transmission protocol, and visualize the workshop twin data; The workshop scheduling module is used to judge the processing results of the data driving module and generate production tasks and workshop intelligent scheduling management decisions.

Citation Information

Patent Citations

  • Mapreduced short period load prediction method of multinucleated function learning SVM realizing multi-source heterogeneous data fusion

    CN105184424A

  • High-end battery production line intelligent factory design method based on digital twinning

    CN117521494A

  • Screw air compressor fault prediction method based on multi-source data fusion

    CN118378195A

  • System and method for predicting power plant operational parameters utilizing artificial neural network deep learning methodologies

    US20170091615A1

Cited By

  • Data storage method, electronic device, non-volatile storage medium and program product

    CN119179448A

  • Soft bean curd production monitoring system and use method thereof

    CN120044908A