A big data-based intelligent management system and method for a whole process of soybean products

Abstract

The present application relates to the field of bean product processing and intelligent management, aims to solve the problems of traditional bean product management fragmentation, quality control lag, and rigid production capacity scheduling, and provides a full-process collaborative management system and method. The present application includes big data acquisition, data preprocessing, intelligent quality prediction, dynamic production capacity scheduling, full-process monitoring, warehouse logistics management, and terminal sales management modules; through the acquisition of full-process multi-source data, after preprocessing, the improved IA-LSTM algorithm is used to predict the quality of key nodes and intervene, the improved IMPSO algorithm is used to generate a dynamic production capacity scheduling scheme, and the warehouse logistics and terminal sales are linked to realize full-process collaboration. The present application improves the quality stability and production capacity utilization rate of bean products, reduces production cost and supply-demand deviation, and is suitable for various full-process intelligent management scenarios of bean products.

Description

Technical Field

[0001] This invention relates to the field of soybean product processing and big data intelligent management, specifically to a big data-based intelligent management system and method for the entire process of soybean products. Background Technology

[0002] Soy products, as a traditional staple food, involve a complex production process encompassing multiple stages, including raw material selection, soaking, grinding, boiling, shaping, sterilization, storage, and transportation. This intricate process is highly interconnected, and deviations in any stage can impact product quality, production efficiency, and costs. Current management models in the soy product industry generally suffer from numerous pain points: First, the entire process management is fragmented, with data from raw material procurement, production and processing, warehousing and logistics being isolated and lacking unified big data integration and analysis, making it impossible to achieve coordinated linkage between various links. Second, quality control is lagging behind, with a "post-event inspection" model being used extensively, making it impossible to predict potential quality problems in advance, which easily leads to the batch scrapping of unqualified products and waste of resources. Third, production capacity scheduling is rigid, with a fixed production capacity plan being used extensively, which cannot be adjusted in real time according to the dynamic changes in raw material supply, market demand and equipment operating conditions, resulting in overcapacity or insufficient supply. Fourth, the core modules of the existing management system mostly use conventional statistics and simple decision-making algorithms, with low intelligence and poor adaptability, and cannot adapt to the dynamic complexity of soybean product processing.

[0003] In existing technologies, soybean product management systems mostly focus on a single link (such as production and processing monitoring), failing to achieve full-process coverage, and lacking innovative improvements to core modules, thus failing to address the aforementioned pain points. Therefore, developing an intelligent management system and method based on big data technology, designing improved core algorithms, and achieving full-process collaboration, quality prediction, and dynamic capacity scheduling has become an urgent need for the transformation and upgrading of the soybean product industry. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent management system and method for the entire process of soybean products based on big data, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a big data-based intelligent management system for the entire process of soybean products, comprising a big data acquisition module, a data preprocessing module, an intelligent quality prediction module, a dynamic capacity scheduling module, a full-process monitoring module, a warehousing and logistics management module, and a terminal sales management module; the big data acquisition module is deployed at each stage of the entire process, namely, raw material procurement points, preprocessing workshops, production workshops, warehouses, logistics vehicles, and terminal stores, collecting multi-source big data including but not limited to raw material parameters, processing parameters, equipment parameters, environmental parameters, warehousing parameters, logistics parameters, and sales parameters, providing data input for subsequent modules; The data preprocessing module performs noise reduction, missing value imputation, normalization, and feature extraction on the collected multi-source big data to form a standardized big data set, providing reliable standardized data input for intelligent quality prediction module, dynamic capacity scheduling module, etc. The full-process monitoring module, based on IoT technology, receives real-time data from the big data acquisition module and the data preprocessing module, connects to the intelligent quality prediction module and the dynamic capacity scheduling module and terminal equipment at each stage, monitors the operating status of each stage of the entire process in real time, identifies abnormal data and issues graded warnings, and links relevant modules and equipment to achieve rapid handling and closed-loop management of anomalies, ensuring that quality prediction intervention instructions and capacity scheduling plans are implemented and avoiding resource waste and capacity loss caused by the expansion of anomalies. The intelligent quality prediction module, based on the standardized feature parameters provided by the data preprocessing module, predicts the product quality in advance at four key stages of soybean product raw material screening, soaking, boiling, and molding, achieving "pre-judgment and in-process intervention". It outputs quality prediction results and intervention instructions, which are fed back to the full-process monitoring module and dynamic production capacity scheduling module, solving the pain point of lagging traditional quality control, improving product quality stability, and reducing waste of unqualified products. The dynamic capacity scheduling module uses the improved multi-objective particle swarm optimization algorithm IMPSO to integrate the full-process data such as raw material supply, quality prediction, and market demand provided by the big data acquisition module, intelligent quality prediction module, and terminal sales management module. Combined with preset constraints, it calculates the optimal capacity scheduling scheme and dynamically adjusts it to achieve the minimization of production costs, the maximization of capacity utilization, and the minimization of supply and demand deviation. This solves the pain point of rigid traditional capacity scheduling and supports the invention goal of optimizing system capacity and minimizing costs. The warehousing and logistics management module receives warehousing and logistics parameters from the big data collection module and scheduling schemes from the dynamic capacity scheduling module. It is responsible for the warehousing zoning management, inventory early warning, and loss control of raw materials and finished products of soybean products, as well as logistics route optimization, transportation status management, and distribution coordination. It also links with the terminal sales management module to achieve warehousing-logistics-sales coordination, ensuring timely supply of raw materials and efficient delivery of finished products, and supporting the smooth execution of the capacity scheduling scheme. The terminal sales management module receives terminal sales data from the big data collection module, integrates sales information from all terminals, and automatically generates standardized sales ledgers, enabling automated management of sales ledgers and traceability of sales data. It analyzes sales trends and market demand preferences, extracts core demand data, provides accurate market demand support for the dynamic capacity scheduling module, and links with the warehousing and logistics management module and the full-process monitoring module to achieve synergy between sales, production, and logistics, facilitating precise matching of supply and demand.

[0006] Preferably, the multi-source big data collected in the big data acquisition module specifically includes the following: Raw material parameters: Real-time collection of data including but not limited to the moisture content of soybeans. Impurity content Protein content The data includes: origin, quantity received, arrival time, and storage period. Among these, moisture content, impurity content, and protein content directly correspond to the characteristic parameters of the raw material screening node in the intelligent quality prediction module. These parameters are collected, recorded in real time, and labeled with the raw material batch to ensure traceability. Processing parameters are also included. Real-time collection of immersion temperature Soaking time Boiling temperature Boiling time pH value of slurry Molding pressure Molding time Molding environment temperature Sterilization temperature The parameters correspond one-to-one with the 16 characteristic parameters of the intelligent quality prediction module, and the acquisition frequency is synchronized with the process (e.g., the boiling temperature is acquired once every 10 seconds). Equipment parameters: Real-time collection of operating speed, energy consumption, running time, and fault codes of production equipment (grinding mill, pulping machine, molding machine, etc.) at each stage, providing equipment operating condition data for the dynamic capacity scheduling module and abnormal early warning data for the full-process monitoring module; Environmental parameters: Synchronously collect temperature, humidity, and air quality parameters from various workshops (pre-processing, production), warehouses, and logistics vehicles. Among them, workshop environmental parameters are related to the stability of processing parameters, and warehouse and logistics vehicle temperature and humidity parameters are related to product storage and transportation quality, providing support for the intelligent quality prediction module and the warehouse logistics management module. Warehouse parameters: Real-time collection of inventory quantities, locations, inbound times, outbound times, and losses of raw materials and finished products, providing data for inventory early warning and zone management in the warehouse logistics management module, and providing input for raw material supply capacity and finished product inventory data in the dynamic capacity scheduling module; Logistics parameters: Real-time collection of data including but not limited to the location of logistics vehicles, transportation speed, temperature and humidity of the vehicle compartment, transportation time, and loading and unloading records, providing data for route optimization in the warehouse logistics management module and supporting finished product distribution scheduling in the dynamic capacity scheduling module; Sales parameters: Real-time collection of finished product sales, sales price, sales time, product batches, and demand feedback from terminal stores, providing data for sales trend analysis in the terminal sales management module, and providing input for market demand forecasting and capacity adjustment in the dynamic capacity scheduling module.

[0007] Preferably, the specific implementation steps of the intelligent quality prediction module are as follows: A1. Key prediction nodes and feature parameter selection: Based on the characteristics of the entire soybean product processing process, four key nodes affecting quality were identified: raw material screening, soaking, boiling, and forming. Each node corresponds to several core feature parameters, which serve as input features for quality prediction, as detailed below: Raw material screening node: Characteristic parameters include soybean moisture content Impurity content Protein content Origin compatibility (range of values) (obtained by training with historical data) Immersion node: Characteristic parameters include immersion temperature Soaking time , water hardness The mass ratio of soybeans to water ; Boiling point: Characteristic parameters include boiling temperature Boiling time pH value of slurry Stirring speed ; Molding node: Characteristic parameters include molding pressure Molding time Molding environment temperature Sterilization temperature ; After filtering, the input feature vector is obtained. There are a total of 16 feature parameters, with standardized feature values ​​provided by the data preprocessing module (mapped to...). (interval); Step A2: Construction of the improved IA-LSTM algorithm: Traditional LSTM networks pay equal attention to different feature parameters, failing to highlight the impact of key features (such as boiling temperature and molding pressure) on the quality of soybean products, resulting in low prediction accuracy. This invention improves upon LSTM networks by introducing an attention mechanism, focusing on feature parameters with high weights affecting quality, and constructing an IA-LSTM algorithm. Step A3, Algorithm Training and Accuracy Verification: Training sample preparation: Collect big data on the entire process of soy products and corresponding quality testing data from the past 3-5 years, and select 10,000 valid samples, of which 8,000 are used as the training set and 2,000 are used as the test set; each sample contains standardized values ​​of 16 feature parameters and corresponding actual quality grade values. ; Algorithm Training: The training set is input into the IA-LSTM algorithm, and the network parameters are adjusted using the Adam optimization algorithm (learning rate 0.001, number of iterations 100, number of hidden layer nodes 64). The loss function is minimized. Complete the algorithm training to obtain a well-trained quality prediction model; Accuracy verification: Input the test set into the trained quality prediction model, calculate the prediction accuracy, and if the prediction error... If the sample proportion is ≥95%, the quality prediction model is deemed qualified; if it is not qualified, adjust the attention weight calculation parameters and LSTM network structure, and retrain until the accuracy requirements are met. Step A4: Real-time quality prediction and anomaly intervention: Real-time prediction: The real-time standardized feature vector output by the data preprocessing module. Input the trained IA-LSTM model and output the quality prediction level values ​​of the four key nodes respectively. At the same time, output the prediction confidence level. (range of values) , (The prediction result is valid). Quality rating and intervention: A preset quality rating threshold, i.e., excellent: ; Qualified: 80≤ Warning: Unqualified: If the prediction is a warning, a warning signal is sent to the full-process monitoring module, prompting staff to adjust the process parameters of the corresponding node (e.g., if the boiling temperature is too low, the temperature is increased); if the prediction is a non-conformity, a shutdown command is immediately triggered to prevent non-conformity products from flowing into the next stage, and feedback is also sent to the dynamic capacity scheduling module to adjust the subsequent capacity plan.

[0008] Preferably, the specific construction method of the IA-LSTM algorithm in step A2 is as follows: Attention mechanism design: Calculate the attention weight for each feature parameter. , The weight calculation formula is: ,in For the first The importance scores of each feature parameter are calculated by the fully connected layer. ; The weight matrix is ​​(16×1). The bias term (1×1) is obtained by training with historical quality data; Ensure that the weight allocation is reasonable; LSTM network improvement: Incorporating attention weights With the corresponding feature parameters Multiply to obtain the weighted eigenvector. The key features are then fed into the LSTM network to replace the original feature input of the traditional LSTM, thereby enhancing the influence of key features on the prediction results. Loss function optimization: An improved mean squared error loss function is adopted, introducing a quality deviation penalty term to avoid batch quality problems caused by excessive prediction deviation. The loss function calculation formula is as follows: in, The value of the loss function; This represents the number of training samples; For the first Actual quality grade value of each sample (range) (The higher the value, the better the quality). This represents the predicted quality level value of the IA-LSTM algorithm. This is the penalty coefficient (value 0.05, which can be adjusted according to actual production) used to penalize samples with large prediction deviations.

[0009] Preferably, the dynamic capacity scheduling module is implemented using the following steps: B1. Determination of scheduling objectives and constraints: Multi-objective scheduling objectives: Three core scheduling objectives are constructed: minimizing production costs, maximizing capacity utilization, and minimizing supply-demand deviation. A weighted summation method is used to construct the multi-objective optimization function, the formula of which is: 1. The parameters are explained as follows: The smaller the overall scheduling objective function value, the better the scheduling effect. The production cost per unit time (yuan / hour) includes raw material costs, equipment energy costs, and labor costs. Due to capacity utilization deviation, The smaller the value, the higher the capacity utilization rate; Due to supply and demand discrepancies, The smaller the value, the higher the degree of supply and demand matching; , , Let the target weight coefficient satisfy... Adjustments will be made based on production priorities; regular production: , , Peak demand season: , , ; Constraints: Based on the actual production of soy products, four types of constraints are set to ensure the feasibility of the scheduling plan: Raw material constraints: Actual production capacity ≤ raw material supply capacity, i.e. ,in Actual production capacity (tons / hour). Raw material supply capacity (tons / hour) is provided by the big data acquisition module; Equipment constraints: Actual capacity ≤ rated capacity of equipment, and equipment operating time ≤ maximum continuous operating time of equipment. , Hour); Quality constraint: Actual production capacity ≤ quality-predicted qualified production capacity, i.e. ,in The qualified production capacity predicted by the intelligent quality prediction module; Time constraints: The scheduling cycle matches the production cycle (scheduled once a day, with each scheduling cycle lasting 24 hours) to ensure that capacity scheduling is synchronized with actual production; Step B2, Construction of the improved IMPSO algorithm: Traditional particle swarm optimization (PSO) algorithms are prone to premature convergence and local optima, making them unsuitable for the multi-constraint and multi-objective requirements of soybean product production capacity scheduling. This invention improves the PSO algorithm by introducing dynamic inertia weights and adaptive learning factors to construct the IMPSO algorithm, as detailed below: Dynamic inertia weight design: Inertia weight The weights are dynamically adjusted according to the number of iterations. In the early stages, the weights are increased to speed up the global search, and in the later stages, the weights are decreased to improve the accuracy of the local search. The formula is as follows: ,in , which is the maximum inertia weight; , where the minimum inertia weight; This represents the current iteration number. That is, the maximum number of iterations; Adaptive learning factor design: learning factors That is, individual cognitive factors and That is, group social factors. and The formula is as follows: (Adaptive adjustment based on the number of iterations) ; ,in, , , , Early stage big, Smaller, enhancing individual search capabilities, later Small, Larger size enhances the group's convergence ability; Particle encoding and fitness function: Real number encoding is used, with each particle corresponding to a set of capacity scheduling schemes (including capacity allocation values ​​for each production stage); the fitness function is the reciprocal of the optimization objective function, combined with constraint penalty terms, as shown in the formula: ,in, This represents the particle fitness value; the larger the value, the better the scheduling scheme. This is the penalty coefficient (value 10). To determine the degree of constraint violation, when all constraints are satisfied... When a certain type of constraint is violated This is the cumulative value corresponding to the degree of violation; Step B3: Algorithm Iteration and Optimal Scheduling Scheme Selection: Parameter initialization: Set IMPSO algorithm parameters and population size. Maximum number of iterations Particle velocity range Production capacity scheduling range (tons / hour, adjusted according to the rated capacity of the equipment); randomly initialize the particle population, each particle corresponds to a set of capacity scheduling schemes, and filter the initial population according to the constraints, removing particles that violate the constraints; Population Iterative Update: Calculate the fitness value of each particle. Record the individual optimal position of each particle. (The scheduling scheme with the highest fitness) and the optimal position of the population (The scheduling scheme with the highest fitness across the entire population); based on dynamic inertia weights and adaptive learning factor , Update the particle's velocity and position using the following formula: , ,in For the first The particle velocity and position in the next iteration; , Random number (range of values) Repeat the iterations until the maximum number of iterations is reached. ; Optimal solution selection and verification: After the iteration, output the optimal position of the population. The corresponding capacity scheduling scheme is verified to see if it meets all constraints. If it does, it is determined to be the optimal capacity scheduling scheme. If it does not meet the constraints, the algorithm parameters are adjusted and the process is repeated until the optimal scheme that meets the constraints is found. Step B4: Issuance and Dynamic Adjustment of Scheduling Plan: Solution distribution: The optimal capacity scheduling plan, including capacity allocation, equipment uptime, and raw material consumption plan for each production stage, is distributed to the full-process monitoring module, production workshop terminals, and raw material procurement module to guide each stage to execute the scheduling plan (such as adjusting the production speed of each production line and optimizing the raw material procurement quantity). Dynamic adjustment: Real-time updates from the big data acquisition module, intelligent quality prediction module, and terminal sales management module. If situations such as insufficient raw material supply, sudden changes in market demand, or a decrease in qualified production capacity occur, the IMPSO algorithm is immediately restarted to quickly iterate and calculate a new optimal scheduling scheme, thereby achieving real-time dynamic adjustment of production capacity and ensuring that production capacity is always matched with supply and demand, quality, and equipment operating conditions.

[0010] Preferably, a big data-based intelligent management method for the entire process of soybean products is applied, and the specific steps are as follows: Step 1: Full-process big data collection. Through the big data collection module, collect multi-source big data from each link of raw material procurement, pre-processing, processing and production, warehousing and logistics, and terminal sales, and transmit it to the data pre-processing module; Step 2: Data preprocessing. The data preprocessing module performs noise reduction, missing value imputation, normalization, and feature extraction on the collected big data to form a standardized big data dataset, which is then transmitted to the big data acquisition module, intelligent quality prediction module, dynamic capacity scheduling module, full-process monitoring module, warehousing and logistics management module, and terminal sales management module, respectively. Step 3: Core module operation. The intelligent quality prediction module uses the IA-LSTM algorithm to predict the quality of each key node and outputs the prediction results and intervention instructions. The dynamic capacity scheduling module uses the IMPSO algorithm to calculate the optimal capacity scheduling plan by combining the quality prediction results, market demand, raw material supply and other data. Step 4: Collaborative execution across the entire process. The intelligent quality prediction module, dynamic capacity scheduling module, full-process monitoring module, warehousing and logistics management module, and terminal sales management module execute full-process management according to quality intervention instructions and capacity scheduling plans, namely raw material procurement, production and processing, warehousing and logistics, and terminal sales. The full-process monitoring module monitors the execution status in real time. Step 5: Dynamic iterative optimization. Execution data from each stage is collected in real time and fed back to the data preprocessing module. After reprocessing, the data is input into the intelligent quality prediction module and the dynamic capacity scheduling module. The parameters of the IA-LSTM and IMPSO algorithms are adjusted to optimize the accuracy of quality prediction and the capacity scheduling scheme, thereby achieving continuous iterative upgrades of the entire process management.

[0011] Compared with the prior art, the beneficial effects of the present invention are: This invention designs improved IA-LSTM and IMPSO algorithms to address the pain points of traditional quality control lagging and rigid production capacity scheduling, respectively. The algorithm improvements are highly targeted and adaptable to the dynamic complexity of the entire process management of soy products. They are significantly different from the core algorithms of existing conventional management systems and have obvious creativity. End-to-end collaborative management, breaking down data silos: This invention integrates big data from the entire process of soy products to achieve collaborative linkage between raw material procurement, production and processing, warehousing and logistics, and terminal sales, avoiding fragmented management and improving the efficiency of end-to-end management; High quality controllability and reduced resource waste: This invention uses the IA-LSTM algorithm to achieve pre-judgment and in-process intervention of quality, which greatly reduces the batch scrapping of unqualified products, reduces waste of raw materials and energy, and improves product quality stability. High capacity adaptability and reduced production costs: This invention achieves dynamic capacity scheduling through the IMPSO algorithm, matches raw material supply with market demand, improves capacity utilization, reduces overcapacity or undersupply, and lowers production costs. Highly practical and widely adaptable: The intelligent quality prediction module and dynamic production capacity scheduling module rely on improved software algorithms, eliminating the need for large-scale modifications to existing production equipment. The big data acquisition module, data preprocessing module, intelligent quality prediction module, warehousing and logistics management module, and terminal sales management module adopt mature technologies, resulting in low system construction costs. It is compatible with various soy products such as tofu, dried tofu, and bean curd sticks, and can be applied to soy product production enterprises of different sizes, making it highly valuable for promotion. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the system structure of the present invention; Figure 2 This is a schematic diagram of the workflow of the intelligent quality prediction module of the present invention; Figure 3 This is a schematic diagram of the workflow of the dynamic capacity scheduling module of the present invention; Figure 4 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Example 1 Please see Figure 1-3This invention provides a technical solution: a big data-based intelligent management system for the entire process of soybean products, comprising a big data acquisition module, a data preprocessing module, an intelligent quality prediction module, a dynamic capacity scheduling module, a full-process monitoring module, a warehousing and logistics management module, and a terminal sales management module. The big data acquisition module is deployed at each stage of the entire process, namely, raw material procurement points, preprocessing workshops, production workshops, warehouses, logistics vehicles, and terminal stores, collecting multi-source big data including but not limited to raw material parameters, processing parameters, equipment parameters, environmental parameters, warehousing parameters, logistics parameters, and sales parameters, providing data input for subsequent modules. It should be noted that the big data collection module's specific working steps closely align with the entire process of soybean product raw material procurement, pretreatment, processing, warehousing and logistics, and terminal sales, as detailed below: Step C1: Precise Deployment and Equipment Debugging of Data Collection Points. Based on the technological characteristics and data requirements of each stage of the soybean product production process, and adhering to the principles of "full coverage, no omissions, and high accuracy," data collection equipment is deployed at designated locations to ensure that the collection scope covers six core stages: raw material procurement points, pre-processing workshops, production workshops, warehouses, logistics vehicles, and terminal stores. Simultaneously, equipment debugging is completed to ensure the normal operation of the data collection equipment. Raw material procurement point: Deploy portable data acquisition terminals, moisture analyzers, and impurity detectors, and assign dedicated personnel to operate them to collect various parameters when raw materials such as soybeans enter the site, adapting to the characteristic parameter collection requirements of the raw material screening node in the intelligent quality prediction module; Pre-processing workshop: Sensors and data acquisition devices are deployed at soybean screening, soaking, and washing stations to collect parameters such as temperature and time during the soaking process, which correspond to the characteristic parameter requirements of the soaking node in the intelligent quality prediction module; Production workshop: Temperature sensors, pressure sensors, pH meters, speed sensors, etc. are deployed at key work stations such as grinding, boiling, molding, and sterilization to accurately collect various process parameters during the processing. This matches the characteristic parameter collection requirements of the intelligent quality prediction module for boiling and molding nodes. At the same time, equipment operation sensors are deployed to collect equipment operating parameters. Warehouse: Deploy temperature and humidity sensors, inventory counting terminals, and material identification equipment to collect storage parameters of raw materials and finished products, providing inventory data support for the warehouse logistics management module and the dynamic capacity scheduling module; Logistics vehicles: Equipped with GPS positioning modules, temperature and humidity sensors, and cargo weight sensors, these vehicles collect various parameters in real time during the logistics and transportation process to ensure the transportation quality of soy products (especially fresh soy products) and provide logistics data for the warehousing and logistics management module and the dynamic capacity scheduling module. Terminal stores: Deploy sales data collection terminals and POS linkage modules to collect various parameters of finished product sales, providing market demand data for terminal sales management modules and dynamic production capacity scheduling modules.

[0015] After the equipment is deployed, unified debugging is carried out to calibrate the sensor accuracy (e.g., temperature sensor error ≤ ±0.5℃, pressure sensor error ≤ ±0.1MPa), and test the communication stability between the data acquisition terminal and the module host to ensure that the acquisition equipment can acquire data accurately and stably, and avoid errors in subsequent module calculations due to equipment deviation.

[0016] Step C2: Multi-category parameter hierarchical collection and standardized recording: Following the principles of "categorized collection, precise matching, and real-time recording," we conducted tiered data collection for seven major parameter categories (raw material parameters, processing parameters, equipment parameters, environmental parameters, warehousing parameters, logistics parameters, and sales parameters), taking into account the process rhythm of each stage. This ensured that the collected data accurately matched the needs of subsequent modules. The specific collection content and requirements are as follows: Raw material parameters: Real-time collection of data including but not limited to the moisture content of soybeans. Impurity content Protein content The data includes: origin, quantity received, arrival time, and storage period. Among these, moisture content, impurity content, and protein content directly correspond to the characteristic parameters of the raw material screening node in the intelligent quality prediction module. These parameters are collected, recorded in real time, and labeled with the raw material batch to ensure traceability. Processing parameters are also included. Real-time collection of immersion temperature Soaking time Boiling temperature Boiling time pH value of slurry Molding pressure Molding time Molding environment temperature Sterilization temperature The parameters correspond one-to-one with the 16 characteristic parameters of the intelligent quality prediction module, and the acquisition frequency is synchronized with the process (e.g., the boiling temperature is acquired once every 10 seconds). Equipment parameters: Real-time collection of operating speed, energy consumption, running time, and fault codes of production equipment (grinding mill, pulping machine, molding machine, etc.) at each stage, providing equipment operating condition data for the dynamic capacity scheduling module and abnormal early warning data for the full-process monitoring module; Environmental parameters: Synchronously collect temperature, humidity, and air quality parameters from various workshops (pre-processing, production), warehouses, and logistics vehicles. Among them, workshop environmental parameters are related to the stability of processing parameters, and warehouse and logistics vehicle temperature and humidity parameters are related to product storage and transportation quality, providing support for the intelligent quality prediction module and the warehouse logistics management module. Warehouse parameters: Real-time collection of inventory quantities, locations, inbound times, outbound times, and losses of raw materials and finished products, providing data for inventory early warning and zone management in the warehouse logistics management module, and providing input for raw material supply capacity and finished product inventory data in the dynamic capacity scheduling module; Logistics parameters: Real-time collection of data including but not limited to the location of logistics vehicles, transportation speed, temperature and humidity of the vehicle compartment, transportation time, and loading and unloading records, providing data for route optimization in the warehouse logistics management module and supporting finished product distribution scheduling in the dynamic capacity scheduling module; Sales parameters: Real-time collection of finished product sales, sales price, sales time, product batches, and demand feedback from terminal stores, providing data for sales trend analysis in the terminal sales management module, and providing input for market demand forecasting and capacity adjustment in the dynamic capacity scheduling module.

[0017] When collecting all parameters, the collection time, collection point, collection personnel (or equipment number), and product batch should be uniformly labeled, and a standardized data format should be used to record them to avoid data confusion and ensure that the subsequent data preprocessing module can efficiently carry out noise reduction, normalization and other tasks.

[0018] Step C3: Real-time data transmission and temporary storage: A two-way data transmission link is established, consisting of "terminal acquisition - wireless transmission - host aggregation," to ensure that the acquired data can be transmitted to the designated node in real time and securely. A temporary storage mechanism is also set up to prevent data loss. Real-time transmission: Each data acquisition device (sensor, data acquisition terminal) transmits the raw data collected to the host of the big data acquisition module in real time via Internet of Things (IoT) technology. The transmission latency is controlled within 100ms, ensuring that the dynamic capacity scheduling module can obtain real-time raw material supply and market demand data, and the intelligent quality prediction module can obtain process parameters of each key node in real time. Temporary storage: Dual temporary storage units are set in the acquisition devices and the host of the module. When the network is interrupted or the transmission is abnormal, the collected data is automatically temporarily stored (storage time ≥72 hours). After the network is restored, it is automatically synchronized to the host of the module to avoid data loss due to network failure. At the same time, the stored data is encrypted to prevent data leakage and tampering.

[0019] Step C4: Data Integrity Verification and Anomaly Handling Real-time integrity verification is performed on the raw data transmitted to the big data acquisition module host to promptly handle issues such as missing or abnormal data, ensuring that the data transmitted to the data preprocessing module is authentic and valid, and providing reliable support for subsequent module operations. Integrity Verification: The big data acquisition module host compares the category and quantity of acquired parameters with preset standards in real time. If data is missing in a certain stage (e.g., no boiling temperature was acquired), data is duplicated, or the data format is abnormal, an early warning signal is immediately triggered and fed back to the full-process monitoring module, while relevant personnel are notified. Anomaly Handling: For missing data, if it is due to a failure of the acquisition equipment, the backup acquisition equipment is immediately switched to collect the corresponding data. If it is due to human error, the operator is notified to re-enter the data in time. For abnormal data (e.g., the acquired boiling temperature exceeds the normal range of 80-100℃), the abnormal data is marked and the cause of the abnormality is indicated (e.g., sensor failure, process abnormality), and it is synchronously transmitted to the full-process monitoring module. The monitoring module will link with relevant workstations to investigate the abnormality and store the abnormal data separately for subsequent traceability analysis. Verification Pass: After verification, if the data is complete and without abnormalities, it is marked as "qualified raw data" and transmitted in batches to the data preprocessing module at a preset frequency (one batch every 5 minutes) for subsequent noise reduction, normalization, and other processing stages.

[0020] Step C5: Dynamic Adaptation of Acquisition Parameters and Equipment Maintenance In conjunction with the iterative optimization needs of the intelligent quality prediction module and the dynamic production capacity scheduling module, as well as the adjustments to the soybean product production process, the data acquisition parameters and frequency are dynamically adapted. Simultaneously, the data acquisition equipment is regularly maintained to ensure the long-term stable operation of the modules: Parameter Adaptation: When the intelligent quality prediction module adjusts its characteristic parameters (e.g., adding a new quality-influencing parameter) or the dynamic production capacity scheduling module adds new data requirements (e.g., adding refined equipment energy consumption parameters), the parameter acquisition list and acquisition frequency of the data acquisition module are adjusted synchronously. Equipment Maintenance: The data acquisition equipment is inspected daily, accuracy is calibrated weekly, and comprehensive maintenance is performed monthly. Faulty sensors and acquisition terminals are replaced promptly, and dust and debris are cleaned from the equipment to ensure the accuracy and stability of the data acquisition equipment and prevent data deviations caused by equipment aging from affecting the calculation accuracy of subsequent modules.

[0021] This module works in close accordance with the invention's entire process management logic. The collected data directly determines the effect of subsequent data preprocessing, which in turn affects the prediction accuracy of the intelligent quality prediction module and the scheduling rationality of the dynamic capacity scheduling module. It is the fundamental support for the stable operation of the entire management system.

[0022] The data preprocessing module performs noise reduction, missing value imputation, normalization, and feature extraction on the collected multi-source big data to form a standardized big data set, providing reliable standardized data input for the intelligent quality prediction module, dynamic capacity scheduling module, etc. The working steps of the data preprocessing module are explained below: Step D1: Data reception and classification. Receive qualified raw data transmitted in batches from the big data acquisition module, classify and organize it into seven categories such as raw material parameters, processing parameters, and equipment parameters, and associate it with the corresponding collection points, time and product batch to ensure that the data accurately corresponds to the needs of subsequent modules. Step D2: Data denoising. Wavelet threshold denoising algorithm is used to remove abnormal noise data caused by sensor fluctuations in processing parameters, environmental parameters, etc. (such as instantaneous abnormal boiling temperature, temperature and humidity data), and retain valid data. Step D3: Missing value imputation. For a small number of missing data, linear interpolation is used to imput the missing data by combining historical data of the same category and batch, so as to avoid the impact of missing data on the calculation accuracy of the core algorithm (such as imputing missing key parameters such as soybean moisture content and forming pressure). Step D4: Data normalization, mapping various parameters to a unified standard. The range eliminates dimensional differences (such as standardizing parameters with different dimensions such as temperature, output, and energy consumption) and adapts to the input requirements of the IA-LSTM algorithm of the intelligent quality prediction module and the IMPSO algorithm of the dynamic capacity scheduling module. Step D5: Feature extraction and dataset output. Focus on extracting 16 core feature parameters required by the intelligent quality prediction module, while retaining key features such as raw material supply, equipment condition, and market demand required by the dynamic capacity scheduling module. Integrate them into a standardized big data set, and transmit them to each core and auxiliary module to complete the preprocessing process.

[0023] The end-to-end monitoring module, based on IoT technology, receives real-time data from the big data acquisition and data preprocessing modules. It connects to the intelligent quality prediction module, dynamic capacity scheduling module, and terminal equipment at each stage, monitoring the operational status of each stage in real time. It identifies abnormal data and issues tiered warnings, coordinating with relevant modules and equipment to achieve rapid anomaly handling and closed-loop management. This ensures the effective implementation of quality prediction intervention commands and capacity scheduling plans, preventing resource waste and capacity loss caused by escalating anomalies. The specific working steps of this module are as follows: Step E1: Monitoring System Construction and Parameter Initialization. Based on the characteristics of the entire soybean product process and the collaborative needs of each module, an IoT monitoring system with "full coverage, hierarchical control, and closed-loop linkage" is built. This involves deploying monitoring nodes, pre-setting parameters, and debugging equipment linkage, laying the foundation for subsequent real-time monitoring. Monitoring Node Deployment: Relying on the data acquisition equipment (sensors, data acquisition terminals, etc.) already deployed in the big data acquisition module, monitoring terminals are simultaneously deployed in six core links to achieve precise correspondence between monitoring nodes and acquisition nodes—procurement monitoring terminals are deployed at the raw material procurement point (linked to raw material testing equipment), workstation monitoring terminals are deployed in the pretreatment workshop (linked to soaking and screening equipment), production line monitoring terminals are deployed in the production workshop (linked to boiling, forming, and sterilization equipment), warehousing monitoring terminals are deployed in the warehouse (linked to temperature and humidity control and inventory equipment), vehicle-mounted monitoring terminals are deployed in logistics vehicles (linked to GPS and temperature and humidity control equipment), and sales monitoring terminals are deployed in terminal stores (linked to cash register and inventory early warning equipment). Monitoring parameter initialization: Preset core monitoring parameters and threshold standards for each stage. The parameters are derived from the standardized data range of the data preprocessing module, the quality thresholds of the intelligent quality prediction module, and the scheduling requirements of the dynamic capacity scheduling module. Define three levels of thresholds: "normal," "early warning," and "abnormal." For example, the normal threshold for boiling temperature in the production workshop is 80-100℃, the early warning threshold is 75-80℃ / 100-105℃, and the abnormal threshold is <75℃ or >105℃. The normal threshold for equipment operating speed is preset according to the scheduling scheme of the dynamic capacity scheduling module, and the normal threshold for storage temperature and humidity is preset according to the storage standards for soybean products. This ensures that the monitoring parameters are accurately matched with the requirements of the intelligent quality prediction module and the dynamic capacity scheduling module. Linkage Equipment Debugging: Complete the linkage debugging of this module with various related modules and terminal devices to ensure that it can receive raw data from the big data acquisition module, standardized data from the data preprocessing module, prediction results and intervention instructions from the intelligent quality prediction module, and scheduling schemes from the dynamic production capacity scheduling module in real time; at the same time, test the linkage response between the monitoring terminal and terminal devices in each link (such as soy boiling machine, temperature and humidity controller, vehicle control equipment) to ensure that instructions are issued, equipment starts and stops, and parameter adjustments are executed quickly with a response delay of ≤500ms.

[0024] Step E2: Real-time monitoring and data synchronization throughout the entire process: Activate the full-domain monitoring mode to collect operational data from each stage in real time, and synchronously link with the data preprocessing module to achieve real-time data synchronization and dynamic updates, comprehensively grasp the operational status of the entire process, and provide support for anomaly identification: Multi-dimensional Real-time Monitoring: Following the principle of "classified monitoring and precise control," multi-dimensional real-time monitoring is conducted across six key stages: Raw material procurement: monitoring the raw material entry inspection process, quantity, and parameter consistency; Pre-processing and production: monitoring process parameters (such as soaking temperature and boiling time) and equipment operating status (speed, energy consumption, fault signals); Warehousing: monitoring inventory quantity, temperature and humidity changes, and raw material and finished product losses; Logistics: monitoring transportation location, vehicle temperature and humidity, and transportation progress; Terminal sales: monitoring sales volume, inventory balance, and product batch consistency. Real-time Data Synchronization: Real-time data collected from each monitoring node is synchronously transmitted to the full-process monitoring module and simultaneously pushed to the data pre-processing module for synchronous processing, ensuring consistency between monitoring data and pre-processed data, intelligent quality prediction module, and dynamic capacity scheduling module's calculation data. Combined with the equipment maintenance plan of the big data acquisition module, routine inspections and monitoring are conducted on the acquisition equipment and monitoring terminals at each stage, providing real-time feedback on equipment malfunctions (such as sensor failures or monitoring terminal offline), and linking with the big data acquisition module to promptly carry out equipment maintenance, avoiding monitoring failures due to acquisition and monitoring equipment malfunctions.

[0025] Step E3: Anomaly Data Identification and Hierarchical Early Warning: Based on preset monitoring parameter thresholds, the system dynamically compares and analyzes real-time monitoring data to quickly identify abnormal data, issues warnings based on severity, and simultaneously coordinates with relevant modules to verify the cause of the anomaly, ensuring early detection and early warning of anomalies. Anomaly Detection: The full-process monitoring module compares the monitored data with preset thresholds in real time. If the monitored data exceeds the normal threshold range (e.g., boiling temperature > 105℃, storage temperature and humidity exceeding the preset range, equipment fault code triggered), it is immediately identified as abnormal data. Simultaneously, it receives quality warnings and non-conforming prediction signals (e.g., predicting a certain node's quality as a warning or non-conforming) from the intelligent quality prediction module, and data verification anomaly signals (e.g., missing data, abnormal data) from the big data acquisition module, and synchronously identifies them as quality-related or data-related anomalies. Tiered Warning System: Anomalies are categorized into three levels based on severity: Level 1 (general anomalies, such as slight data fluctuations, quality prediction warnings), Level 2 (more severe anomalies, such as equipment parameters exceeding limits, minor non-conforming predictions). The system includes three levels of early warning (serious anomalies, such as equipment failure and shutdown, batch quality non-compliance prediction, and severe temperature and humidity exceeding standards in logistics); different warning levels correspond to different warning methods (Level 1 warning pop-up notification, Level 2 warning pop-up + audible and visual alarm, Level 3 warning pop-up + audible and visual alarm + dedicated personnel notification); anomaly verification: after an anomaly warning is triggered, the system immediately links with the data preprocessing module to verify the authenticity of the abnormal data (excluding false warnings caused by data noise and acquisition equipment deviation), and links with the intelligent quality prediction module and big data acquisition module to analyze the cause of the anomaly (such as quality anomalies related to process parameter deviations, equipment anomalies related to equipment operating conditions, and data anomalies related to acquisition processes), forming a preliminary anomaly cause analysis report, which is simultaneously pushed to relevant personnel and core modules.

[0026] Step E4: Multi-module linkage and rapid exception handling: For different types and levels of anomalies, the intelligent quality prediction module, dynamic capacity scheduling module, and terminal equipment at each stage are linked to formulate targeted handling plans, enabling rapid response and closed-loop handling of anomalies, and minimizing the impact of anomalies on the entire process. Handling Quality-Related Anomalies (Linked to Intelligent Quality Prediction Module): Upon receiving quality warnings and non-conformity prediction signals from the intelligent quality prediction module, immediately issue intervention instructions to the corresponding production station terminal equipment—for example, if a quality warning is predicted for the soaking node, instruct the soaking equipment to adjust the soaking temperature and time; if the prediction is non-conformity, immediately instruct the corresponding station to stop to prevent non-conforming products from flowing into the next stage; simultaneously, feed the handling results back to the intelligent quality prediction module, linking it to re-predict quality until the anomaly is eliminated. Handling Capacity and Equipment-Related Anomalies (Linked to Dynamic Capacity Scheduling Module): If anomalies such as equipment failure, insufficient raw material supply, or logistics delays are detected, causing capacity to be unable to be executed according to the scheduling plan, immediately push an anomaly signal to the dynamic capacity scheduling module, linking it to recalculate the optimal scheduling plan; simultaneously, issue equipment start / stop and parameter adjustment instructions. For terminal equipment (e.g., immediate shutdown for repair in case of equipment failure, expedited procurement by the raw material procurement module when needed), production and logistics plans are adjusted according to the new scheduling scheme to ensure stable production capacity; For data and monitoring anomaly handling (linking with big data acquisition and preprocessing modules): if data acquisition or preprocessing anomalies are detected, the big data acquisition module is immediately linked to investigate the acquisition equipment and process, and the data preprocessing module is linked to reprocess the abnormal data; if monitoring terminals are detected to be offline or malfunctioning, staff are immediately notified to carry out equipment repairs to ensure the monitoring system quickly returns to normal; Tiered handling and control: Level 1 warnings are handled by staff based on the warning prompts, and the results are reported after handling; Level 2 and 3 warnings are handled by staff based on the preliminary anomaly cause analysis report, linking with various modules for rapid handling, with real-time monitoring of the handling process to ensure the effective implementation of handling measures.

[0027] Step E5: Feedback and Closed-Loop Management of Handling Results After an anomaly handling is completed, the handling results are collected in real time and fed back to all related modules to update monitoring data, forming a closed loop for anomaly handling. At the same time, the anomaly handling process is recorded to provide data support for subsequent optimization of the entire process. Results Feedback: After staff complete the anomaly handling, they enter the handling results (e.g., process parameters adjusted, equipment maintenance completed, quality prediction returned to normal) into the monitoring terminal. The full-process monitoring module receives the handling results in real time and simultaneously pushes them to the intelligent quality prediction module, dynamic capacity scheduling module, and data preprocessing module, linking each module to update relevant data (e.g., the intelligent quality prediction module re-predicts, the dynamic capacity scheduling module confirms the execution of the scheduling plan, and the data preprocessing module updates standardized data). Closed-Loop Verification: The module host continuously tracks the real-time monitoring data after the handling. If the monitoring data returns to the normal threshold range, the intelligent quality prediction module and the dynamic capacity scheduling module will verify the data. The scheduling module reports that the anomaly has been eliminated, indicating that the anomaly handling is complete and a closed loop has been formed. If the monitoring data is still abnormal, the warning is triggered again, and relevant modules are linked to handle the anomaly again until it is completely eliminated. Record retention and optimization: The warning time, anomaly type, anomaly cause, handling measures, and handling results of each anomaly are recorded to form an anomaly handling record. This record is simultaneously pushed to the big data acquisition module and the data preprocessing module, serving as important data support for subsequent monitoring parameter optimization, intelligent quality prediction module, dynamic capacity scheduling module algorithm iteration (IA-LSTM, IMPSO algorithm), and process adjustment, continuously improving the accuracy of the entire process monitoring and the efficiency of anomaly handling.

[0028] Step E6: Dynamic optimization and adaptation of the monitoring system: In response to the iterative needs of end-to-end management, the system links various related modules to dynamically optimize monitoring parameters, monitoring nodes, and linkage mechanisms. This ensures that the monitoring module is always compatible with the intelligent quality prediction module, the dynamic capacity scheduling module's calculations, end-to-end process adjustments, and changes in market demand. Parameter optimization: Based on the algorithm iteration requirements of the intelligent quality prediction module and the dynamic capacity scheduling module, as well as the anomaly handling archives and full-process operation data, regularly optimize the monitoring parameter thresholds (such as adjusting the warning range of process parameters and equipment operation thresholds) to improve the accuracy of anomaly identification and reduce false alarms; Node adaptation: If the big data acquisition module adds new collection points, or the intelligent quality prediction module and the dynamic capacity scheduling module add new data requirements, simultaneously optimize the deployment of monitoring nodes and add corresponding monitoring terminals to ensure that the monitoring range is consistent with the collection range and the requirements of the intelligent quality prediction module and the dynamic capacity scheduling module; Linkage mechanism optimization: Based on anomaly handling experience, optimize the linkage response mechanism with core modules and terminal equipment, shorten the linkage response delay, improve the hierarchical early warning and handling process, and improve the efficiency of anomaly handling and the level of closed-loop management.

[0029] This module works in accordance with the invention's closed-loop management logic throughout the entire process. It receives real-time data from the big data collection and data preprocessing modules at the top and links with the intelligent quality prediction module, dynamic capacity scheduling module, and terminal equipment at the bottom. This effectively ensures that the intervention commands of the intelligent quality prediction module and the scheduling plans of the dynamic capacity scheduling module are implemented, avoiding resource waste and capacity loss caused by abnormal expansion, and further improving the intelligent and collaborative management level of the entire management system.

[0030] The intelligent quality prediction module, based on standardized feature parameters provided by the data preprocessing module, predicts product quality in advance at four key stages of soybean product raw material selection, soaking, boiling, and molding, achieving "pre-judgment and in-process intervention." It outputs quality prediction results and intervention instructions, feeding them back to the full-process monitoring module and dynamic production capacity scheduling module. This addresses the pain point of lagging traditional quality control, improves product quality stability, and reduces waste from substandard products. The specific implementation steps are as follows: A1. Key prediction nodes and feature parameter selection: Based on the characteristics of the entire soybean product processing process, four key nodes affecting quality were identified: raw material screening, soaking, boiling, and forming. Each node corresponds to several core feature parameters, which serve as input features for quality prediction, as detailed below: Raw material screening node: Characteristic parameters include soybean moisture content Impurity content Protein content Origin compatibility (range of values) (obtained by training with historical data) Immersion node: Characteristic parameters include immersion temperature Soaking time , water hardness The mass ratio of soybeans to water ; Boiling point: Characteristic parameters include boiling temperature Boiling time pH value of slurry Stirring speed ; Molding node: Characteristic parameters include molding pressure Molding time Molding environment temperature Sterilization temperature ; After filtering, the input feature vector is obtained. There are a total of 16 feature parameters, with standardized feature values ​​provided by the data preprocessing module (mapped to...). (interval); Step A2: Construction of the improved IA-LSTM algorithm: Traditional LSTM networks pay equal attention to different feature parameters, failing to highlight the impact of key features (such as boiling temperature and molding pressure) on the quality of soybean products, resulting in low prediction accuracy. This invention improves upon LSTM networks by introducing an attention mechanism, focusing on feature parameters with high weights affecting quality, and constructing an IA-LSTM algorithm. The specific construction method of the IA-LSTM algorithm is as follows: Attention mechanism design: Calculate the attention weight for each feature parameter. , The weight calculation formula is: ,in For the first The importance scores of each feature parameter are calculated by the fully connected layer. ; The weight matrix is ​​(16×1). The bias term (1×1) is obtained by training with historical quality data; Ensure that the weight allocation is reasonable; LSTM network improvement: Incorporating attention weights With the corresponding feature parameters Multiply to obtain the weighted eigenvector. The key features are then fed into the LSTM network to replace the original feature input of the traditional LSTM, thereby enhancing the influence of key features on the prediction results. Loss function optimization: An improved mean squared error loss function is adopted, introducing a quality deviation penalty term to avoid batch quality problems caused by excessive prediction deviation. The loss function calculation formula is as follows: in, The value of the loss function; This represents the number of training samples; For the first Actual quality grade value of each sample (range) (The higher the value, the better the quality). This represents the predicted quality level value of the IA-LSTM algorithm. This is a penalty coefficient (valued at 0.05, which can be adjusted according to actual production) used to penalize samples with large prediction deviations. Step A3, Algorithm Training and Accuracy Verification: Training sample preparation: Collect big data on the entire process of soy products and corresponding quality testing data from the past 3-5 years, and select 10,000 valid samples, of which 8,000 are used as the training set and 2,000 are used as the test set; each sample contains standardized values ​​of 16 feature parameters and corresponding actual quality grade values. ; Algorithm Training: The training set is input into the IA-LSTM algorithm, and the network parameters are adjusted using the Adam optimization algorithm (learning rate 0.001, number of iterations 100, number of hidden layer nodes 64). The loss function is minimized. Complete the algorithm training to obtain a well-trained quality prediction model; Accuracy verification: Input the test set into the trained quality prediction model, calculate the prediction accuracy, and if the prediction error... If the sample proportion is ≥95%, the quality prediction model is deemed qualified; if it is not qualified, adjust the attention weight calculation parameters and LSTM network structure, and retrain until the accuracy requirements are met. Step A4: Real-time quality prediction and anomaly intervention: Real-time prediction: The real-time standardized feature vector output by the data preprocessing module. Input the trained IA-LSTM model and output the quality prediction level values ​​of the four key nodes respectively. At the same time, output the prediction confidence level. (range of values) , (The prediction result is valid). Quality rating and intervention: A preset quality rating threshold, i.e., excellent: ; Qualified: 80≤ Warning: Unqualified: If the prediction is an early warning, a warning signal is sent to the full-process monitoring module, prompting staff to adjust the process parameters at the corresponding node (e.g., increasing the temperature if the boiling temperature is too low). If the prediction is a non-conformity, a shutdown command is immediately triggered to prevent non-conforming products from flowing into the next stage. Simultaneously, feedback is sent to the dynamic capacity scheduling module to adjust subsequent capacity plans. The dynamic capacity scheduling module uses the improved multi-objective particle swarm optimization algorithm (IMPSO), integrating full-process data such as raw material supply, quality prediction, and market demand from the big data acquisition module, intelligent quality prediction module, and terminal sales management module. Combined with preset constraints, it calculates the optimal capacity scheduling scheme and dynamically adjusts it to minimize production costs, maximize capacity utilization, and minimize supply-demand deviation. This addresses the pain points of rigid traditional capacity scheduling and supports the invention's goal of optimizing system capacity and minimizing costs. The specific implementation steps are as follows: B1. Determination of scheduling objectives and constraints: Multi-objective scheduling objectives: Three core scheduling objectives are constructed: minimizing production costs, maximizing capacity utilization, and minimizing supply-demand deviation. A weighted summation method is used to construct the multi-objective optimization function, the formula of which is: 1. The parameters are explained as follows: The smaller the overall scheduling objective function value, the better the scheduling effect. The production cost per unit time (yuan / hour) includes raw material costs, equipment energy costs, and labor costs. Due to capacity utilization deviation, The smaller the value, the higher the capacity utilization rate; Due to supply and demand discrepancies, The smaller the value, the higher the degree of supply and demand matching; , , Let the target weight coefficient satisfy... Adjustments will be made based on production priorities; regular production: , , Peak demand season: , , ; Constraints: Based on the actual production of soy products, four types of constraints are set to ensure the feasibility of the scheduling plan: Raw material constraints: Actual production capacity ≤ raw material supply capacity, i.e. ,in Actual production capacity (tons / hour). Raw material supply capacity (tons / hour) is provided by the big data collection module; Equipment constraints: Actual production capacity ≤ rated production capacity, and equipment operating time ≤ maximum continuous operating time. , Hour); Quality constraint: Actual production capacity ≤ quality-predicted qualified production capacity, i.e. ,in The qualified production capacity predicted by the intelligent quality prediction module; Time constraints: The scheduling cycle matches the production cycle (scheduled once a day, with each scheduling cycle lasting 24 hours) to ensure that capacity scheduling is synchronized with actual production; Step B2, Construction of the improved IMPSO algorithm: Traditional particle swarm optimization (PSO) algorithms are prone to premature convergence and local optima, making them unsuitable for the multi-constraint and multi-objective requirements of soybean product production capacity scheduling. This invention improves the PSO algorithm by introducing dynamic inertia weights and adaptive learning factors to construct the IMPSO algorithm, as detailed below: Dynamic inertia weight design: Inertia weight The weights are dynamically adjusted according to the number of iterations. In the early stages, the weights are increased to speed up the global search, and in the later stages, the weights are decreased to improve the accuracy of the local search. The formula is as follows: ,in , which is the maximum inertia weight; , where the minimum inertia weight; This represents the current iteration number. That is, the maximum number of iterations; Adaptive learning factor design: learning factors That is, individual cognitive factors and That is, group social factors. and The formula is as follows: (Adaptive adjustment based on the number of iterations) ; ,in, , , , Early stage big, Smaller, enhancing individual search capabilities, later Small, Larger size enhances the group's convergence ability; Particle encoding and fitness function: Real number encoding is used, with each particle corresponding to a set of capacity scheduling schemes (including capacity allocation values ​​for each production stage); the fitness function is the reciprocal of the optimization objective function, combined with constraint penalty terms, as shown in the formula: ,in, This represents the particle fitness value; the larger the value, the better the scheduling scheme. This is the penalty coefficient (value 10). To determine the degree of constraint violation, when all constraints are satisfied... When a certain type of constraint is violated This is the cumulative value corresponding to the degree of violation; Step B3: Algorithm Iteration and Optimal Scheduling Scheme Selection: Parameter initialization: Set IMPSO algorithm parameters and population size. Maximum number of iterations Particle velocity range Production capacity scheduling range (tons / hour, adjusted according to the rated capacity of the equipment); randomly initialize the particle population, each particle corresponds to a set of capacity scheduling schemes, and filter the initial population according to the constraints, removing particles that violate the constraints; Population Iterative Update: Calculate the fitness value of each particle. Record the individual optimal position of each particle. (The scheduling scheme with the highest fitness) and the optimal position of the population (The scheduling scheme with the highest fitness across the entire population); based on dynamic inertia weights and adaptive learning factor , Update the particle's velocity and position using the following formula: , ,in For the first The particle velocity and position in the next iteration; , Random number (range of values) Repeat the iterations until the maximum number of iterations is reached. ; Optimal solution selection and verification: After the iteration, output the optimal position of the population. The corresponding capacity scheduling scheme is verified to see if it meets all constraints. If it does, it is determined to be the optimal capacity scheduling scheme. If it does not meet the constraints, the algorithm parameters are adjusted and the process is repeated until the optimal scheme that meets the constraints is found. Step B4: Issuance and Dynamic Adjustment of Scheduling Plan: Solution distribution: The optimal capacity scheduling plan, including capacity allocation, equipment uptime, and raw material consumption plan for each production stage, is distributed to the full-process monitoring module, production workshop terminals, and raw material procurement module to guide each stage to execute the scheduling plan (such as adjusting the production speed of each production line and optimizing the raw material procurement quantity). Dynamic Adjustment: The system receives real-time updates from the big data acquisition module, intelligent quality prediction module, and terminal sales management module. If situations arise such as insufficient raw material supply, sudden changes in market demand, or a decrease in capacity due to quality prediction failures, the IMPSO algorithm is immediately restarted to rapidly iterate and calculate a new optimal scheduling scheme. This enables real-time dynamic adjustment of capacity, ensuring that capacity is always matched with supply and demand, quality, and equipment operating conditions. The terminal sales management module receives terminal sales data from the big data collection module, integrates sales information from all terminals, and automatically generates standardized sales ledgers, achieving automated management of sales ledgers and traceability of sales data. It analyzes sales trends and market demand preferences, extracts core demand data, and provides precise market demand support for the dynamic capacity scheduling module. It also links with the warehousing and logistics management module and the full-process monitoring module to achieve synergy between sales, production, and logistics, facilitating precise matching of supply and demand. The specific working content of the terminal sales management module is briefly described below in conjunction with the invention: Step F1: Sales Data Reception and Standardization. As the receiving terminal for sales data from the big data acquisition module, this step receives, verifies, and standardizes the collected raw sales data to ensure data quality and lay the foundation for subsequent analysis and application. It also works in conjunction with the big data acquisition module to complete the data loop. Data reception: Real-time reception of raw terminal sales data transmitted in batches from the big data acquisition module, covering finished product sales, sales price, sales time, product batches, demand feedback, store inventory balance, return and exchange records, etc. of each terminal store. The reception frequency is consistent with the data push frequency of the big data acquisition module (one batch every 5 minutes) to ensure data real-time performance. Data verification: Referencing the data verification standards of the big data acquisition module, the received sales data is verified for completeness and authenticity. Duplicate data and abnormal data (such as unreasonably high or negative sales) are removed, and missing data (such as missing product batches or sales prices) are supplemented. If major data anomalies are found (such as missing batch data or sales that deviate significantly from the normal range), an early warning signal is immediately triggered and fed back to the full-process monitoring module and the big data acquisition module to jointly investigate and resolve issues with the acquisition terminal or the acquisition process. Standardization and organization: The verified sales data is classified and organized according to a unified standard, and associated with corresponding dimension information—classified by product category (tofu, dried tofu, bean curd sticks, etc.), terminal store, sales period, and product batch, and labeled with data collection time and store number. It is transformed into standardized data consistent with the output format of the data preprocessing module, ensuring compatibility with the data format of the dynamic capacity scheduling module and the warehouse logistics management module, and facilitating data collaboration among multiple modules.

[0031] Step F2: Sales data integration and automated ledger generation: Based on standardized sales data, centralized integration of sales data across all terminals and time periods is achieved, automatically generating sales ledgers to replace traditional manual ledgers, improving management efficiency and ensuring the traceability of sales data. Data Integration: A database integrating terminal sales data is established to centrally store standardized sales data from various terminal stores, enabling "queryable data for individual stores and aggregated data for all terminals." The integrated content includes daily / weekly / monthly / quarterly sales summaries, detailed sales data for each product, sales performance for each store, returns and exchanges summaries, and product batch sales tracking, ensuring comprehensive and complete data coverage. Automated Ledger Generation: Based on the integrated sales data, standardized sales ledgers are automatically generated, covering four main categories: detailed sales ledgers, product batch sales ledgers, store performance ledgers, and returns and exchanges ledgers. The ledgers are automatically generated. The system labels data sources and generation times, and supports quick queries and exports by sales time, product category, store number, product batch, and other dimensions, eliminating the need for manual data entry and reducing human error. The system also features ledger updates and retention: sales ledgers are updated in real-time, automatically updating the corresponding ledger content each time a new batch of sales data is received; historical ledgers are encrypted and retained for a period consistent with the traceability period for soy product production batches (no less than 1 year), for subsequent data queries, trend analysis, and anomaly tracing. Simultaneously, core ledger data is pushed to the data preprocessing module as historical data reserves.

[0032] Step F3: Sales Trends and Demand Preference Analysis: Based on the integrated standardized sales data, and using existing trend analysis algorithms, multi-dimensional analysis is conducted to extract core demand information, providing accurate data support for the dynamic capacity scheduling module and meeting the multi-objective requirements of dynamic capacity scheduling. Sales Trend Analysis: Analyze sales trends of various products (tofu, dried tofu, etc.) on a daily, weekly, monthly, and quarterly basis, identify peak sales periods (such as holidays, daily peak dining times) and trough periods, predict sales trends for the next 1-7 days, calculate the average daily sales and fluctuation range of each product, and analyze sales differences among terminal stores to clarify the demand scale of key sales stores; Demand Preference Analysis: Combine product category, sales period, and demand feedback information from sales data to analyze demand preferences in the terminal market—such as product preferences of stores in different regions (a certain region prefers dried tofu, another region prefers tofu), and demand differences at different times (a preference for soy milk during breakfast, a preference for tofu during lunch / dinner), extract core demand characteristics, identify products with rapid sales growth and slow-moving products, and generate a demand preference analysis report; Data Refinement: Refine the results of trend analysis and demand preference analysis into core data that can be directly used by the dynamic capacity scheduling module, including predicted sales of each product, demand priority, regional demand distribution, and demand fluctuation coefficient, ensuring that the data is accurately matched with the supply-demand deviation optimization target and capacity allocation requirements of the dynamic capacity scheduling module.

[0033] Step F4: Demand Data Push and Multi-Module Collaboration: By linking the dynamic capacity scheduling module, warehousing and logistics management module, and end-to-end monitoring module, it enables accurate demand data delivery and multi-module collaboration, supporting closed-loop management throughout the entire process and aligning with the logic of collaborative invention management. Pushing data to the dynamic capacity scheduling module: Based on the dynamic capacity scheduling module's scheduling cycle (once daily), push sales trend analysis results, demand preference data, and predicted sales in real time. Simultaneously, in the event of sudden market demand changes (such as a sudden increase in sales or sluggish sales of a product), immediately push emergency demand warnings, supporting the dynamic capacity scheduling module to adjust weighting coefficients and recalculate the optimal scheduling plan to ensure accurate matching of capacity and market demand, helping to minimize supply-demand discrepancies. Collaborating with the warehousing and logistics management module: Pushing real-time inventory levels and sales data from each terminal store to the warehousing and logistics management module, supporting the module to optimize delivery plans—such as prioritizing delivery to stores with insufficient inventory or high sales volume, adjusting delivery volumes for sluggish products, reducing inventory backlog, and achieving coordinated "sales-warehousing-logistics" operations. Collaborating with the end-to-end monitoring module: Pushing signals such as abnormal sales data and sudden demand changes to the end-to-end monitoring module, triggering warnings and handling anomalies, while simultaneously receiving anomaly handling feedback from the end-to-end monitoring module, updating sales data and analysis results, and ensuring data consistency.

[0034] Step F5: Sales Ledger Management and Anomaly Handling: This system enables full lifecycle management of sales records, handles various anomalies in the sales process, ensures standardized and traceable sales management, and supports the beneficial effects of the invention (reduced labor costs and improved management efficiency). Refined Sales Ledger Management: Automated sales ledgers are meticulously maintained, supporting modifications, additions, queries, and exports. Modification permissions are clearly defined, and modification records are maintained to ensure the ledgers' authenticity and standardization. Simultaneously, it integrates with the financial process, automatically extracting sales amounts and revenue data to support financial accounting and reduce manual processing costs. Sales Anomaly Handling: For sales anomalies (such as returns, near-expiry products, and store inventory backlog / shortage), targeted solutions are developed based on sales data and feedback from various modules. For example, returned products are linked to their corresponding production batches, and feedback is sent to the intelligent quality prediction module to investigate potential quality issues, and to the warehouse logistics management module to arrange for returns and recovery. For near-expiry products, alerts are pushed to terminal stores and the dynamic capacity scheduling module to adjust subsequent capacity allocation. Data Traceability: Based on sales ledgers and integrated sales data, the entire sales process is traceable. Product batches can be used to query corresponding sales stores, sales times, and sales volumes. Sales anomaly records allow tracing the causes, handling processes, and results of anomalies, providing data support for subsequent quality traceability and process optimization.

[0035] Step F6: Dynamic Adaptation and Optimization of Modules In response to the iterative needs of end-to-end invention management, the dynamic capacity scheduling module and big data acquisition module are linked to dynamically optimize the data collection scope, analysis dimensions, and ledger management mode, ensuring that the modules always adapt to the needs of end-to-end management. Data Adaptation: If the dynamic capacity scheduling module adds new demand data dimensions (such as regional demand priority, product specification requirements), the collection scope and analysis dimensions of sales data will be adjusted accordingly, and corresponding data integration and push will be added to ensure the accuracy of demand data support. If the big data collection module adjusts the collection parameters of sales data, the data reception and verification standards of this module will be optimized accordingly to ensure data compatibility. Algorithm Optimization: Combining historical sales data and anomaly handling experience, the sales trend analysis algorithm will be optimized regularly to improve the accuracy of sales forecasting, reduce forecast deviation, and provide more reliable demand data support for the dynamic capacity scheduling module. Function Optimization: Based on the actual needs of terminal sales management, the generation mode and query dimensions of sales ledgers will be optimized, and personalized management functions (such as store performance statistics and employee sales assessment data extraction) will be added to further improve the intelligence and efficiency of terminal sales management.

[0036] This module operates entirely within the closed-loop management logic of the invention process. It builds upon the sales-related raw data from the big data acquisition module, connects with the dynamic capacity scheduling module to provide core demand support, and collaborates with the warehousing and logistics management module to achieve coordinated delivery. This improves the utilization rate of sales data and provides crucial data support for the dynamic capacity scheduling module to minimize supply-demand discrepancies. Furthermore, it refines the entire collaborative management system, contributing to achieving the invention's goal of "optimized capacity and minimized costs." The warehousing and logistics management module receives warehousing and logistics parameters from the big data collection module and scheduling plans from the dynamic capacity scheduling module. It is responsible for the warehousing zoning management, inventory early warning, and loss control of raw materials and finished products, as well as logistics route optimization, transportation status management, and distribution coordination. It also collaborates with the terminal sales management module to achieve warehousing-logistics-sales coordination, ensuring timely raw material supply and efficient finished product delivery, and supporting the smooth execution of the capacity scheduling plan. Its simplified workflow is as follows: Step G1: Receive real-time parameters of warehousing (inventory quantity, temperature and humidity, etc.) and logistics (transportation location, temperature and humidity, etc.) transmitted by the big data acquisition module, as well as the capacity scheduling plan issued by the dynamic capacity scheduling module, and complete data synchronization and verification. Step G2: The warehouse implements zoned storage according to the scheduling plan and the characteristics of raw materials and finished products, monitors the inventory status in real time, triggers inventory warnings, and controls the loss of raw materials and finished products. Step G3: Based on the scheduling plan and the store demand and inventory balance data pushed by the terminal sales management module, the logistics side optimizes the delivery route, monitors the transportation status in real time, and ensures that delivery matches sales and production capacity requirements. Step G4 involves linking the terminal sales management module to synchronize delivery progress and inventory update information, forming a closed loop of warehousing-logistics-sales collaboration, and ensuring the implementation of the capacity scheduling plan.

[0037] Example 2 Please see Figure 4 A big data-based intelligent management method for the entire process of soybean products, applied to a big data-based intelligent management method for the entire process of soybean products, includes the following steps: Step 1: Full-process big data collection. Through the big data collection module, collect multi-source big data from each link of raw material procurement, pre-processing, processing and production, warehousing and logistics, and terminal sales, and transmit it to the data pre-processing module; Step 2: Data preprocessing. The data preprocessing module performs noise reduction, missing value imputation, normalization, and feature extraction on the collected big data to form a standardized big data dataset, which is then transmitted to the big data acquisition module, intelligent quality prediction module, dynamic capacity scheduling module, full-process monitoring module, warehousing and logistics management module, and terminal sales management module, respectively. Step 3: Core module operation. The intelligent quality prediction module uses the IA-LSTM algorithm to predict the quality of each key node and outputs the prediction results and intervention instructions. The dynamic capacity scheduling module uses the IMPSO algorithm to calculate the optimal capacity scheduling plan by combining the quality prediction results, market demand, raw material supply and other data. Step 4: Collaborative execution across the entire process. The intelligent quality prediction module, dynamic capacity scheduling module, full-process monitoring module, warehousing and logistics management module, and terminal sales management module execute full-process management according to quality intervention instructions and capacity scheduling plans, namely raw material procurement, production and processing, warehousing and logistics, and terminal sales. The full-process monitoring module monitors the execution status in real time. Step 5: Dynamic iterative optimization. Execution data from each stage is collected in real time and fed back to the data preprocessing module. After reprocessing, the data is input into the intelligent quality prediction module and the dynamic capacity scheduling module. The parameters of the IA-LSTM and IMPSO algorithms are adjusted to optimize the accuracy of quality prediction and the capacity scheduling scheme, thereby achieving continuous iterative upgrades of the entire process management.

[0038] This invention relates to the field of soybean product processing and intelligent management, aiming to solve the problems of fragmented management, lagging quality control, and rigid capacity scheduling in traditional soybean product management, and to provide a full-process collaborative management system and method. This invention includes modules for big data collection, data preprocessing, intelligent quality prediction, dynamic capacity scheduling, full-process monitoring, warehousing and logistics management, and terminal sales management. By collecting multi-source data from the entire process, and after preprocessing, an improved IA-LSTM algorithm is used to predict and intervene in the quality of key nodes. An improved IMPSO algorithm is used to generate a dynamic capacity scheduling scheme, linking warehousing and logistics with terminal sales to achieve full-process collaboration. This invention improves the quality stability and capacity utilization of soybean products, reduces production costs and supply-demand discrepancies, and is applicable to various intelligent full-process management scenarios for soybean products.

[0039] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A big data-based intelligent management system for the entire process of soybean products, characterized in that, It includes a big data acquisition module, a data preprocessing module, an intelligent quality prediction module, a dynamic capacity scheduling module, a full-process monitoring module, a warehousing and logistics management module, and a terminal sales management module. The big data acquisition module is deployed at all stages of the entire process, namely raw material procurement points, preprocessing workshops, production workshops, warehouses, logistics vehicles, and terminal stores. It collects multi-source big data, including but not limited to raw material parameters, processing parameters, equipment parameters, environmental parameters, warehousing parameters, logistics parameters, and sales parameters, to provide data input for subsequent modules. The data preprocessing module performs noise reduction, missing value imputation, normalization, and feature extraction on the collected multi-source big data to form a standardized big data set, providing reliable standardized data input for intelligent quality prediction module, dynamic capacity scheduling module, etc. The full-process monitoring module, based on IoT technology, receives real-time data from the big data acquisition module and the data preprocessing module, connects to the intelligent quality prediction module and the dynamic capacity scheduling module and terminal equipment at each stage, monitors the operating status of each stage of the entire process in real time, identifies abnormal data and issues graded warnings, and links relevant modules and equipment to achieve rapid handling and closed-loop management of anomalies, ensuring that quality prediction intervention instructions and capacity scheduling plans are implemented and avoiding resource waste and capacity loss caused by the expansion of anomalies. The intelligent quality prediction module, based on the standardized feature parameters provided by the data preprocessing module, predicts the product quality in advance at the four key nodes of raw material screening, soaking, boiling, and molding of soybean products, outputs the quality prediction results and intervention instructions, and feeds them back to the full-process monitoring module and dynamic capacity scheduling module. The dynamic capacity scheduling module uses the improved multi-objective particle swarm optimization algorithm IMPSO to integrate the full-process data such as raw material supply, quality prediction, and market demand provided by the big data acquisition module, intelligent quality prediction module, and terminal sales management module. Combined with preset constraints, it calculates the optimal capacity scheduling scheme and dynamically adjusts it to minimize production costs, maximize capacity utilization, and minimize supply and demand deviation. The warehousing and logistics management module receives warehousing and logistics parameters from the big data collection module and scheduling schemes from the dynamic capacity scheduling module. It is responsible for the warehousing zoning management, inventory early warning, and loss control of raw materials and finished products of soybean products, as well as logistics route optimization, transportation status management, and distribution coordination. It also links with the terminal sales management module to achieve warehousing-logistics-sales coordination, ensuring timely supply of raw materials and efficient delivery of finished products, and supporting the smooth execution of the capacity scheduling scheme. The terminal sales management module receives terminal sales data from the big data collection module, integrates sales information from all terminals, and automatically generates standardized sales ledgers, enabling automated management of sales ledgers and traceability of sales data. It analyzes sales trends and market demand preferences, extracts core demand data, provides accurate market demand support for the dynamic capacity scheduling module, and links with the warehousing and logistics management module and the full-process monitoring module to achieve synergy between sales, production, and logistics, facilitating precise matching of supply and demand.

2. The intelligent management system for the entire process of soybean products based on big data as described in claim 1, characterized in that: The multi-source big data collected in the big data acquisition module specifically includes the following: Raw material parameters: Real-time collection of data including but not limited to the moisture content of soybeans. Impurity content Protein content The data includes: origin, quantity received, arrival time, and storage period. Among these, moisture content, impurity content, and protein content directly correspond to the characteristic parameters of the raw material screening node in the intelligent quality prediction module. These parameters are collected, recorded in real time, and labeled with the raw material batch to ensure traceability. Processing parameters are also included. Real-time collection of soaking temperature Soaking time Boiling temperature Boiling time slurry pH value Molding pressure Molding time Molding environment temperature Sterilization temperature The parameters correspond one-to-one with the 16 characteristic parameters of the intelligent quality prediction module, and the acquisition frequency is synchronized with the process. Equipment parameters: Real-time collection of operating speed, energy consumption, runtime, and fault codes of production equipment at each stage, providing equipment operating condition data for the dynamic capacity scheduling module and abnormal early warning data for the full-process monitoring module; Environmental parameters: Synchronously collect temperature, humidity, and air quality parameters from various workshops, warehouses, and logistics vehicles. Among them, workshop environmental parameters are related to the stability of processing parameters, and temperature and humidity parameters of warehouses and logistics vehicles are related to product storage and transportation quality, providing support for the intelligent quality prediction module and the warehouse logistics management module. Warehouse parameters: Real-time collection of inventory quantities, locations, inbound times, outbound times, and losses of raw materials and finished products, providing data for inventory early warning and zone management in the warehouse logistics management module, and providing input for raw material supply capacity and finished product inventory data in the dynamic capacity scheduling module; Logistics parameters: Real-time collection of data including but not limited to the location of logistics vehicles, transportation speed, temperature and humidity of the vehicle compartment, transportation time, and loading and unloading records, providing data for route optimization in the warehouse logistics management module and supporting finished product distribution scheduling in the dynamic capacity scheduling module; Sales parameters: Real-time collection of finished product sales, sales price, sales time, product batches, and demand feedback from terminal stores, providing data for sales trend analysis in the terminal sales management module, and providing input for market demand forecasting and capacity adjustment in the dynamic capacity scheduling module.

3. The intelligent management system for the entire process of soybean products based on big data as described in claim 1, characterized in that: The specific implementation steps of the intelligent quality prediction module are as follows: A1. Key prediction nodes and feature parameter selection: Based on the characteristics of the entire soybean product processing process, four key nodes affecting quality were identified: raw material screening, soaking, boiling, and forming. Each node corresponds to several core feature parameters, which serve as input features for quality prediction, as detailed below: Raw material screening node: Characteristic parameters include soybean moisture content Impurity content Protein content Origin compatibility ; Immersion node: Characteristic parameters include immersion temperature Soaking time , water hardness The mass ratio of soybeans to water ; Boiling point: Characteristic parameters include boiling temperature Boiling time slurry pH value Stirring speed ; Molding node: Characteristic parameters include molding pressure Molding time Molding environment temperature Sterilization temperature ; After filtering, the input feature vector is obtained. There are a total of 16 feature parameters, and the standardized feature values ​​are provided by the data preprocessing module. Step A2: Construction of the improved IA-LSTM algorithm: An improved LSTM network is adopted, and an attention mechanism is introduced to focus on feature parameters with high weights that have a significant impact on quality, thus constructing the IA-LSTM algorithm. Step A3, Algorithm Training and Accuracy Verification: Training sample preparation: Collect big data on the entire process of soy products and corresponding quality testing data from the past 3-5 years, and select 10,000 valid samples, of which 8,000 are used as the training set and 2,000 are used as the test set; each sample contains standardized values ​​of 16 feature parameters and corresponding actual quality grade values. ; Algorithm training: Input the training set into the IA-LSTM algorithm, use the Adam optimization algorithm to adjust the network parameters, and minimize the loss function. Complete the algorithm training to obtain a well-trained quality prediction model; Accuracy verification: Input the test set into the trained quality prediction model, calculate the prediction accuracy, and if the prediction error... If the sample proportion is ≥95%, the quality prediction model is deemed qualified; if it is not qualified, adjust the attention weight calculation parameters and LSTM network structure, and retrain until the accuracy requirements are met. Step A4: Real-time quality prediction and anomaly intervention: Real-time prediction: The real-time standardized feature vector output by the data preprocessing module. Input the trained IA-LSTM model and output the quality prediction level values ​​of the four key nodes respectively. At the same time, output the prediction confidence level. ; Quality rating and intervention: A preset quality rating threshold, i.e., excellent: ; Qualified: 80≤ ; Warning: Unqualified: ; If the prediction is a warning, a warning signal is sent to the full-process monitoring module, prompting staff to adjust the process parameters of the corresponding node; if the prediction is a non-conformity, a shutdown command is immediately triggered to prevent non-conforming products from flowing into the next stage, and feedback is also sent to the dynamic capacity scheduling module to adjust the subsequent capacity plan.

4. The intelligent management system for the entire process of soybean products based on big data as described in claim 1, characterized in that: The specific construction method of the IA-LSTM algorithm in step A2 is as follows: Attention mechanism design: Calculate the attention weight for each feature parameter. , The weight calculation formula is: ,in For the first The importance scores of each feature parameter are calculated by the fully connected layer. ; This is the weight matrix. This is the bias term, obtained by training with historical quality data; Ensure that the weight allocation is reasonable; LSTM network improvement: Incorporating attention weights With the corresponding feature parameters Multiply to obtain the weighted eigenvector. The key features are then fed into the LSTM network to replace the original feature input of the traditional LSTM, thereby enhancing the influence of key features on the prediction results. Loss function optimization: An improved mean squared error loss function is adopted, introducing a quality deviation penalty term to avoid batch quality problems caused by excessive prediction deviation. The loss function calculation formula is as follows: in, The value of the loss function; This represents the number of training samples; For the first The actual quality grade value of each sample; This represents the predicted quality level value of the IA-LSTM algorithm. This is the penalty coefficient, used to penalize samples with large prediction bias.

5. The intelligent management system for the entire process of soybean products based on big data as described in claim 1, characterized in that: The specific implementation steps of the dynamic capacity scheduling module are as follows: B1. Determination of scheduling objectives and constraints: Multi-objective scheduling objectives: Three core scheduling objectives are constructed: minimizing production costs, maximizing capacity utilization, and minimizing supply-demand deviation. A weighted summation method is used to construct the multi-objective optimization function, the formula of which is:

1. The parameters are explained as follows: The smaller the overall scheduling objective function value, the better the scheduling effect. The cost per unit of time is the production cost, which includes raw material costs, equipment energy costs, and labor costs. Due to capacity utilization deviation, The smaller the value, the higher the capacity utilization rate; Due to supply and demand discrepancies, The smaller the value, the higher the degree of supply and demand matching; , , Let the target weight coefficient satisfy... Adjustments will be made based on production priorities; regular production: , , Peak demand season: , , ; Constraints: Based on the actual production of soy products, four types of constraints are set to ensure the feasibility of the scheduling plan: Raw material constraints: Actual production capacity ≤ raw material supply capacity, i.e. ,in Actual production capacity Raw material supply capacity is provided by the big data collection module; Equipment constraints: Actual production capacity ≤ rated production capacity, and equipment operating time ≤ maximum continuous operating time. , Hour); Quality constraint: Actual production capacity ≤ quality-predicted qualified production capacity, i.e. ,in The qualified production capacity predicted by the intelligent quality prediction module; Time constraints: The scheduling cycle is matched with the production cycle to ensure that capacity scheduling is synchronized with actual production; Step B2, Construction of the improved IMPSO algorithm: The IMPSO algorithm is constructed by introducing dynamic inertia weights and adaptive learning factors, as detailed below: Dynamic inertia weight design: Inertia weight The weights are dynamically adjusted according to the number of iterations. In the early stages, the weights are increased to speed up the global search, and in the later stages, the weights are decreased to improve the accuracy of the local search. The formula is as follows: ,in , which is the maximum inertia weight; , where the minimum inertia weight; This represents the current iteration number. That is, the maximum number of iterations; Adaptive learning factor design: learning factors That is, individual cognitive factors and That is, group social factors. and The formula is as follows: (Adaptive adjustment based on the number of iterations) ; ,in, , , , Early stage big, Smaller, enhancing individual search capabilities, later Small, Larger size enhances the group's convergence ability; Particle encoding and fitness function: Real-number encoding is used, with each particle corresponding to a set of production scheduling schemes; the fitness function is the reciprocal of the optimization objective function, combined with constraint penalty terms, as shown in the formula: ,in, This represents the particle fitness value; the larger the value, the better the scheduling scheme. This is the penalty coefficient; To determine the degree of constraint violation, when all constraints are satisfied... When a certain type of constraint is violated This is the cumulative value corresponding to the degree of violation; Step B3: Algorithm Iteration and Optimal Scheduling Scheme Selection: Parameter initialization: Set IMPSO algorithm parameters and population size. Maximum number of iterations Particle velocity range Production capacity scheduling range Randomly initialize the particle population, with each particle corresponding to a set of capacity scheduling schemes. Filter the initial population according to constraints and remove particles that violate the constraints. Population Iterative Update: Calculate the fitness value of each particle. Record the individual optimal position of each particle. and the optimal position of the population Based on dynamic inertia weights and adaptive learning factor , Update the particle's velocity and position using the following formula: , ,in For the first The particle velocity and position in the next iteration; , Use random numbers; repeat the iteration until the maximum number of iterations is reached. ; Optimal solution selection and verification: After the iteration, output the optimal position of the population. The corresponding capacity scheduling scheme is verified to see if it meets all constraints. If it does, it is determined to be the optimal capacity scheduling scheme. If it does not meet the constraints, the algorithm parameters are adjusted and the process is repeated until the optimal scheme that meets the constraints is found. Step B4: Issuance and Dynamic Adjustment of Scheduling Plan: Plan distribution: The optimal capacity scheduling plan, including capacity allocation, equipment operating time, and raw material consumption plan for each production link, is distributed to the full-process monitoring module, production workshop terminal, and raw material procurement module to guide each link to execute according to the scheduling plan; Dynamic adjustment: Real-time data updates from the big data acquisition module, intelligent quality prediction module, and terminal sales management module are received. If situations such as insufficient raw material supply, sudden changes in market demand, or a decrease in qualified production capacity occur, the IMPSO algorithm is immediately restarted to quickly iterate and calculate a new optimal scheduling scheme, thereby achieving real-time dynamic adjustment of production capacity and ensuring that production capacity is always matched with supply and demand, quality, and equipment operating conditions.

6. A big data-based intelligent management method for the entire process of soybean products, applied to the big data-based intelligent management method for the entire process of soybean products as described in any one of claims 1-5, characterized in that, Includes the following steps: Step 1: Full-process big data collection. Through the big data collection module, collect multi-source big data from each link of raw material procurement, pre-processing, processing and production, warehousing and logistics, and terminal sales, and transmit it to the data pre-processing module; Step 2: Data preprocessing. The data preprocessing module performs noise reduction, missing value imputation, normalization, and feature extraction on the collected big data to form a standardized big data dataset, which is then transmitted to the big data acquisition module, intelligent quality prediction module, dynamic capacity scheduling module, full-process monitoring module, warehousing and logistics management module, and terminal sales management module, respectively. Step 3: Core module operation. The intelligent quality prediction module uses the IA-LSTM algorithm to predict the quality of each key node and outputs the prediction results and intervention instructions. The dynamic capacity scheduling module uses the IMPSO algorithm to calculate the optimal capacity scheduling plan by combining the quality prediction results, market demand, raw material supply and other data. Step 4: Collaborative execution across the entire process. The intelligent quality prediction module, dynamic capacity scheduling module, full-process monitoring module, warehousing and logistics management module, and terminal sales management module execute full-process management according to quality intervention instructions and capacity scheduling plans, namely raw material procurement, production and processing, warehousing and logistics, and terminal sales. The full-process monitoring module monitors the execution status in real time. Step 5: Dynamic iterative optimization. Execution data from each stage is collected in real time and fed back to the data preprocessing module. After reprocessing, the data is input into the intelligent quality prediction module and the dynamic capacity scheduling module. The parameters of the IA-LSTM and IMPSO algorithms are adjusted to optimize the accuracy of quality prediction and the capacity scheduling scheme, thereby achieving continuous iterative upgrades of the entire process management.

Images