Method and apparatus for controlling a multi-objective optimization process agent based on food production

Abstract

The application discloses a control method and device of a multi-objective optimization process intelligent agent based on food production. The method comprises the following steps: obtaining raw material information and standardizing the raw material information to form a raw material feature vector; a three-dimensional process matrix containing equipment nodes, equipment types and process parameters is constructed and embedded; a proxy model is trained using small sample data as an approximation of high-fidelity simulation or experiment; a multi-objective iterative optimization is performed in combination with a global optimization algorithm, and a computational fluid dynamics supplement or experimental verification is triggered in combination with an uncertainty evaluation strategy to realize an active learning closed loop of 'optimization-supplement-update'; the optimal process parameters are subjected to time sequence logic and virtual debugging verification through a digital twin system; and versioning is realized through a manufacturing execution system to realize the closed loop feedback of production data. The application solves the problems of strong coupling of traditional extraction process parameters, high experimental cost and poor cross-batch stability, and realizes multi-objective Pareto optimization of yield, energy consumption, time consumption and quality.

Description

Technical Field

[0001] This invention relates to the fields of food process engineering and intelligent manufacturing, specifically to a control method and apparatus for a multi-objective optimization process intelligence agent based on food production. Background Technology

[0002] The production of liquid foods involves multiple unit operations, such as crushing, enzymatic hydrolysis, ultrasonic extraction, filtration, concentration, sterilization, homogenization, drying, and filling. The coupling relationships between these processes are complex, and process parameters such as temperature, pressure, pH, shear, and flow rate of each piece of equipment require manual adjustment based on experience, making it difficult to simultaneously achieve multiple objectives such as extraction rate, energy consumption, processing time, and equipment capacity. Existing technologies rely on experience-based adjustments and point-by-point experiments, which are expensive and lack sufficient coverage, making it difficult to achieve true multi-objective Pareto optimality within an affordable cost. Raw material differences are not uniformly quantified, and the optimization-simulation-execution-feedback chain is fragmented, resulting in poor cross-batch stability and long development cycles. Furthermore, existing solutions are often single-objective or isolated simulation / MES management, lacking a unified mapping of "raw materials—process—objectives," proxy acceleration, and a closed-loop self-evolutionary system. Summary of the Invention

[0003] The purpose of this invention is to address the shortcomings of existing technologies by providing a control method and device for a multi-objective optimization process intelligence agent based on food production. This method and device can achieve one-click optimization of multiple objectives such as "maximizing output, minimizing energy consumption, minimizing time consumption, and maintaining stable quality" under conditions of multi-unit operation, strong coupling, and multi-objective conflict, within a limited experimental / simulation cost. Furthermore, it can be quickly deployed and self-evolved through MES / PLC.

[0004] The objective of this invention is achieved through the following technical solution:

[0005] This invention provides a control method for a multi-objective optimization process intelligence agent based on food production, comprising the following steps:

[0006] S1. Obtain the physicochemical characteristic data of the target batch of raw materials and perform standardization processing;

[0007] S2. Establish a three-dimensional process matrix representing the physical topology of the production line. Generate a binary mask tensor based on the selection state of equipment units in the three-dimensional process matrix. Based on the mask tensor, concatenate and fuse the three-dimensional process matrix with the raw material feature vector to generate the model input feature vector.

[0008] S3 Input the input feature vector into the pre-trained surrogate model to obtain the predicted values ​​of multi-dimensional process indicators and the uncertainty of the prediction; the pre-trained surrogate model is an ensemble learning model or a multi-task deep neural network model obtained by training through historical production data and offline fluid dynamics simulation database.

[0009] S4 If the prediction uncertainty is lower than a preset threshold, then a global optimization algorithm is executed based on the prediction value, and iterative optimization is performed until convergence and the optimal process configuration is output.

[0010] S5 uses a digital twin model to virtually verify the optimal process configuration;

[0011] S6 will send the verified process configuration scheme to the production system for execution and collect production data to update the proxy model.

[0012] According to one embodiment of this application, step S1 specifically comprises:

[0013] The physicochemical characteristic data of the target batch of raw materials are obtained, including continuous physicochemical characteristic data and discrete characteristic data; Z-score standardization is performed on the continuous physicochemical characteristic data; one-heat encoding is performed on the discrete characteristic data; the standardized continuous features and the encoded discrete features are concatenated in a predetermined order to generate the standardized feature vector.

[0014] According to one embodiment of this application, in step S2, the three-dimensional process matrix is ​​an N×M×P tensor structure, where N corresponds to the physical section node of the production line, M corresponds to the optional equipment unit physically connected under each node, and P corresponds to the readable and writable process register address in the PLC controller of each equipment unit.

[0015] According to one embodiment of this application, step S2 includes the following sub-steps:

[0016] S2.1 Generate a binary mask tensor with the same structure as the three-dimensional process matrix based on the selection status of the equipment units in the three-dimensional process matrix, wherein the value of the selected equipment unit and its associated entire process parameter space at the corresponding position in the mask tensor is 1, and the value of the unselected equipment unit and its associated parameter space at the corresponding position is 0.

[0017] S2.2 The standardized feature vector is concatenated with the expanded three-dimensional process matrix to obtain the concatenated vector;

[0018] S2.3 Calculate the Hadamard product of the concatenated vector and the expanded binary mask tensor, filter out the features corresponding to the unselected devices, and generate the model input vector.

[0019] According to one embodiment of this application, in step S3, when the uncertainty of the prediction exceeds a preset threshold, computational fluid dynamics simulation is automatically triggered to perform physical field resampling, and the resampling result is used as the true value to update the surrogate model. According to one embodiment of this application, in step S4, the global optimization algorithm is one of genetic algorithm, particle swarm optimization algorithm, differential evolution algorithm, reinforcement learning, or Bayesian optimization.

[0020] The present invention also provides a control device for a multi-objective optimization process intelligence agent based on food production, comprising:

[0021] The data acquisition and standardization module is configured to acquire the characteristic data of the target batch of raw materials through the laboratory information management system, and to standardize and encode the characteristic data to generate a standardized feature vector of the target batch of raw materials.

[0022] The 3D process matrix construction module configures a 3D process matrix to characterize the physical topology of the production line.

[0023] The intelligent decision-making core module is communicatively connected to the data acquisition and standardization module and the three-dimensional process matrix construction module. It includes a global optimization algorithm engine, a feature mask encoder, and a proxy model, and is configured to realize process optimization based on mask mapping and active learning.

[0024] The simulation verification module is connected to the intelligent decision-making core module and is configured to perform virtual simulation and verification of the control logic, timing and performance of the optimal process configuration scheme through a digital twin model.

[0025] The execution and feedback module includes an MES terminal and a PLC control network, configured to send the verified process configuration scheme to the production system for execution, and collect production data to update the agent model.

[0026] The beneficial effects of this invention are:

[0027] A three-dimensional process matrix enables rapid cross-raw material migration and automated process path and parameter generation. A surrogate model replaces expensive evaluations, and uncertainty-driven replenishment balances efficiency and accuracy, shortening product development and process adjustment time. Optimized solutions maximize extraction rates while reducing energy consumption and time, helping to lower production costs. Digital twins and simulation verification ensure the feasibility and quality stability of each optimized solution, reducing trial-and-error costs. MES versioning enables one-click deployment of optimal solutions, reducing the risk of human input errors. Through data feedback from actual production processes, optimized solutions can be continuously updated and improved, gradually enhancing the intelligence and automation level of the production process. Attached Figure Description

[0028] The present invention will be further described below with reference to the accompanying drawings and embodiments;

[0029] Figure 1 This is a system architecture diagram of the multi-objective optimization process intelligent agent of the present invention;

[0030] Figure 2 This is a flowchart of the main control method of the present invention;

[0031] Figure 3 This is the three-dimensional process matrix of the present invention;

[0032] Figure 4 A flowchart for uncertainty-driven active learning;

[0033] Figure 5 This invention provides a schematic diagram of a control device for a multi-objective optimization process intelligence agent in food production. Detailed Implementation

[0034] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0035] like Figure 1 As shown, the control method of the multi-objective optimization process intelligence agent based on food production of the present invention includes:

[0036] 1. Data Collection and Standardization:

[0037] 1.1 Raw Material Data Acquisition: After the raw materials enter the production system, relevant data is first acquired through the Laboratory Information Management System (LIMS). This data includes:

[0038] Physicochemical properties: continuous variables, such as moisture content (%), sugar content (Brix), total acid (TA), pH value, etc.

[0039] Ingredient content: a continuous variable, such as the initial concentration of the target active ingredient (polyphenols, flavonoids);

[0040] Source information: discrete variables, including raw material variety, place of origin, harvesting season and batch number.

[0041] 1.2 Data Cleaning and Standardization: The collected data is cleaned, including outlier detection, missing value imputation, and unit conversion, to generate raw material feature vectors. For continuous features, Z-score standardization is used to eliminate dimensional differences, making the parameters of different raw materials comparable. For discrete features, one-hot encoding is used for transformation, such as encoding discrete features like raw material origin and harvest season, ensuring that the features can be processed by intelligent algorithms.

[0042] 1.3 The standardized continuous features and the discrete sparse vectors encoded by one-hot encoding are concatenated in sequence to generate a standardized raw material feature vector, which includes both continuous and discrete raw material features.

[0043] 2. Definition of three-dimensional matrix: Establish a three-dimensional process matrix with dimensions N×M×P, which includes "equipment nodes, equipment types, and process parameters".

[0044] This matrix is ​​not simply a data array, but a digital mapping container of the physical topology of the production line. Each element in the matrix has a unique hard mapping to the underlying control network of the workshop (such as the OPC UA address space). The X-axis (equipment node N) includes the process sequence nodes of the processing flow such as raw material pretreatment, extraction, filtration, purification, concentration, sterilization, drying, and packaging; the Y-axis (equipment type M) represents the different equipment types that can be used for each node, such as countercurrent extraction, ultrasonic extraction, and supercritical extraction equipment in the extraction process; the Z-axis (process parameter P) refers to the adjustable control parameters corresponding to each equipment, including temperature, pH, pressure, ultrasonic power, flow rate, residence time, motor speed, etc.

[0045] See the three-dimensional process matrix. Figure 3 Through this mapping, any modification to the matrix by the algorithm is equivalent to a reconstruction of the physical production line control logic, thus solving the engineering problem that traditional optimization algorithms cannot directly handle the strong coupling between equipment selection and parameter control.

[0046] Sparsity and Constraint Mechanism: A pre-set mutual exclusion mask within the matrix. When device m is selected at a node n... i At that time, the matrix automatically activates the Z-axis parameter space corresponding to the device, while simultaneously locking other devices m under the same node. j≠i The parameter space (set to invalid values ​​or zero) forms a structured process path decision tree. When the global optimization algorithm attempts to evaluate a process path, the mask tensor forcibly masks the neural connections of all unselected devices.

[0047] This method can significantly reduce the computational dimensionality of the model, accelerate convergence, and, more importantly, ensure the physical completeness and uniqueness of the feature vectors input to the proxy model, which is something that general machine learning methods cannot achieve.

[0048] 3. Global optimization algorithm generation and feature mask mapping

[0049] 3.1 Solution Vector Encoding: The equipment selection (Y-axis) and process parameters (Z-axis) in the 3D process matrix are mapped into a solution vector structure that can be processed by the global optimization algorithm. If a genetic algorithm (GA) is used, the matrix parameters are encoded as chromosomes; if a particle swarm optimization (PSO) algorithm is used, the matrix parameters are encoded as particle position vectors; if reinforcement learning (RL) is used, the matrix states are encoded as a state space. The system establishes a structure-aware parameter update mechanism to ensure that the discrete changes in equipment selection and the continuous changes in process parameters maintain the consistency of the physical topology during algorithm iteration.

[0050] 3.2 Feature Flattening and Mutual Exclusion Masking: An encoder is set up to concatenate and flatten the sparse "3D process matrix" and "raw material feature vector" to generate a fixed-length input feature vector X. input Because the three-dimensional matrix contains a large number of mutually exclusive device paths (i.e., selecting device A precludes selecting device B), it cannot be directly input into the proxy model. Therefore, a mutual exclusion masking mechanism is introduced. The concatenated vector and the mask tensor are subjected to an element-wise product operation, followed by a flattening operation, to generate a physically complete and sparse fixed-length input feature vector X. input When a genetic algorithm attempts to evaluate a process path, the mask tensor forcibly masks the neuron connections of all unselected devices.

[0051] This method can significantly reduce the computational dimensionality of the model, accelerate convergence, and, more importantly, ensure the physical completeness and uniqueness of the feature vectors input to the proxy model, which is something that general machine learning methods cannot achieve.

[0052] 4. Proxy Model Prediction and Active Learning

[0053] An ensemble learning model (such as XGBoost) or a multi-task deep neural network (Multi-Task DNN) is used as the surrogate model. The input is X. input It includes raw material attributes and process attributes; the output is a multi-dimensional vector Y (including output, energy consumption, time consumption and quality) and the uncertainty of the predicted value.

[0054] Surrogate model pre-training: Methods such as Kriging and radial basis function (RBF) are used to generate training samples based on existing CFD or experimental data, and 10-fold cross-validation is performed to ensure the stability and accuracy of the model. Ultimately, the surrogate model can provide fast and accurate predictions for unknown operating conditions and provides the model's prediction variance as a measure of uncertainty.

[0055] Uncertainty Assessment: During the optimization process, when the uncertainty of the predicted results for certain parameter combinations exceeds a preset threshold, the system automatically invokes fluid dynamics (CFD) simulation software to generate a physical simulation task for that specific process matrix parameter. If the uncertainty is less than or equal to the preset threshold, the prediction result is deemed reliable, and the system proceeds to the next step of fitness calculation. The uncertainty threshold ranges from 3% to 15%, preferably from 5% to 10%.

[0056] Replenishment calculation and self-calibration: After each replenishment, the CFD simulation results are returned to the system, added to the training set, and the surrogate model performs incremental learning. After the model is updated, predictions are re-made for the current population.

[0057] 5. Multi-objective evolutionary optimization and convergence determination

[0058] Objective function design: The optimization objectives include extraction rate (Y), energy consumption (E), time consumption (T), and quality (Q). A multi-objective fitness function is designed to comprehensively consider these objectives, using weighted coefficients (ω1, ω2, ω3, ω4) to balance the relationships between the objectives.

[0059]

[0060] in, For extraction rate, For energy consumption, Because it takes time, For quality, The corresponding maximum or minimum boundary conditions, ω1, ω2, ω3, and ω4 are the corresponding weight coefficients, which can be dynamically adjusted according to the production strategy.

[0061] Intelligent optimization algorithms with global search capabilities (including but not limited to genetic algorithms, particle swarm optimization algorithms, differential evolution algorithms, or deep reinforcement learning) are used to traverse the high-dimensional mixed space of process parameters to find the parameter combination that optimizes the objective function.

[0062] Taking the genetic algorithm as an example, the convergence determination is as follows: Check whether the current population meets the convergence conditions (such as reaching the maximum number of iterations or fitness stagnation). If it has not converged, perform the selection, crossover, and mutation operations of the genetic algorithm to generate the next generation of the three-dimensional process matrix population, and return to step 3 to repeat the process. If it has converged, output the Pareto optimal solution set and proceed to the next verification stage.

[0063] Taking genetic algorithms as an example, the flowchart of uncertainty-driven active learning is shown below. Figure 4 .

[0064] 6. Digital twin simulation verification

[0065] Digital Twin Construction: The digital twin system is built on a multiphysics simulation platform for joint simulation of fluid flow, heat transfer, and mass transfer behavior during the extraction process. The multiphysics simulation platform can be COMSOL Multiphysics, ANSYS Fluent, OpenFOAM, or an equivalent platform, and can work collaboratively with process-level and control-level simulation platforms (such as MATLAB / Simulink). A virtual production line model is established, and the optimal three-dimensional process matrix output by the GA is parsed into control commands for the digital twin for virtual testing. Testing includes: process cycle time and capacity assessment, energy and material consumption prediction, quality index simulation, and control timing and logic verification (such as valve opening and closing timing, PID control logic). Only solutions that pass digital twin verification are marked as "executable versions." If verification fails, constraints are fed back to the GA engine for re-optimization.

[0066] 7. Closed-loop control and iteration

[0067] MES System Integration: Optimal recipes and process parameters are distributed to the PLC registers of production equipment via the Manufacturing Execution System (MES), driving the actions of actuators such as frequency converters and control valves, thus achieving real-time deployment of process parameters. Each process recipe generates a version number and effective time in the MES and undergoes electronic signature verification.

[0068] Production data feedback: During the production process, MES collects real-time process data and quality indicators and feeds this data back to the optimization system for subsequent model retraining and optimization.

[0069] Example 1: Multi-objective process optimization for flavonoid extraction from pear peel residue

[0070] This embodiment applies the aforementioned multi-objective optimization process intelligence for food production to optimize the extraction of flavonoid active substances from pear peel and pomace, a byproduct of pear juice processing. Pear peel and pomace are rich in flavonoids, but due to the different origins and ripeness of the pears, the cellulose structure of the peel and pomace varies significantly, severely affecting the dissolution efficiency. Traditional fixed processes struggle to achieve both high extraction rates and low energy consumption.

[0071] The specific implementation steps of this embodiment are as follows:

[0072] 1. Raw material data collection and standardization: When a batch of pear peel residue raw material enters the production line, the intelligent agent first obtains the physicochemical data of that batch of pear peel residue raw material through the LIMS system interface and performs preprocessing, including the following steps:

[0073] Continuous feature collection: Focus on collecting raw material particle size, moisture content, and cellulose content of bran and pomace.

[0074] Discrete feature acquisition: Acquire classification label information that is associated with the batch, such as the origin of the raw materials (e.g., "Dangshan, Anhui") and the harvest season (e.g., "late-maturing").

[0075] Data preprocessing: For continuous features (particle size, moisture content, fiber content), Z-score normalization is used, as shown in the following formula:

[0076]

[0077] Where x is a certain physicochemical index of the sample, μ is the average value of the physicochemical index of all samples, and σ is the standard deviation of the physicochemical index of all samples.

[0078] For discrete features (origin, season), one-hot encoding is used to generate sparse vectors. The standardized continuous features are then concatenated with the one-hot encoded discrete feature sparse vectors in sequence to finally generate a standardized raw material feature vector V. pear This feature vector includes both continuous and discrete features of raw materials.

[0079] 2. Three-dimensional matrix construction of pear extraction process

[0080] The system initializes a three-dimensional process matrix with dimensions N×M×P, where N corresponds to the physical process node of the production line, M corresponds to the selectable equipment unit physically connected to each node, and P corresponds to the readable and writable process register address in the PLC controller of each equipment unit. In this embodiment, the focus is on intelligent configuration for the "extraction section (Node 2)":

[0081] Equipment node (X-axis): Locked to Node 2 (extraction section).

[0082] Equipment type (Y-axis): The agent selects between "Dynamic Countercurrent Extraction Tank" and "High-Power Ultrasonic Extraction Tank". Assume that the current round of optimization genetic algorithm (GA) activates "Ultrasonic Extraction Tank (Device M=1)".

[0083] Process parameters (Z-axis): Activate the space of 4 key adjustable parameters corresponding to this equipment: Ultrasonic power (P1): range [100,800] W; Ethanol concentration (P2): range [40,100] %; Extraction temperature (P3): range [30,60] ℃; Material-liquid ratio (P4): range [1:5,1:50].

[0084] Generate a binary mask tensor with the same structure as the three-dimensional process matrix, wherein the value of the selected device unit and its associated entire process parameter space at the corresponding position in the mask tensor is 1, and the value of the unselected device unit and its associated parameter space at the corresponding position is 0.

[0085] 3. Proxy Model Prediction and Active Optimization

[0086] Feature mapping: mapping the raw material feature vector V pear After concatenating the vectors with the aforementioned three-dimensional matrix using a mutual exclusion masking mechanism, the Hadamard product (element-wise product) operation is performed to filter out feature interference from invalid physical paths, and then input into the pre-trained XGBoost surrogate model.

[0087] Multi-objective optimization: The surrogate model outputs three predicted target values: extraction rate ( Energy consumption ), Ethanol consumption ( The genetic algorithm iteratively optimizes based on the following fitness function:

[0088]

[0089] In the formula, This represents the maximum extraction rate. To minimize energy consumption, This represents the minimum ethanol consumption.

[0090] Intelligent Decision Making: Given the high cellulose content (>15%) detected in this batch of raw materials, the surrogate model predicts that the traditional "low power + high temperature" extraction method will not achieve sufficient extraction efficiency. In this embodiment, a genetic algorithm (GA) is preferably used: the algorithm iterates through selection, crossover, and mutation operations. With a population size of 100, a crossover probability of 0.8, and a mutation probability of 0.05, 120 generations of iterative searching are performed. The genetic algorithm converges to a new set of parameters: significantly increasing the ultrasonic power to 750W while simultaneously reducing the temperature to 45°C, controlling the ethanol concentration to 65%, and controlling the material-to-liquid mass-to-volume ratio to 1:20. At this point, the prediction uncertainty given by the surrogate model is below the threshold, indicating the feasibility of the proposed solution. The uncertainty threshold is set based on the allowable error of the process system, and in this embodiment, it is preferably approximately 10%. Ensuring both prediction reliability and avoiding CFD computational overload are key technical parameters for the intelligent process optimization efficiency of this invention.

[0091] 4. Digital Twin Verification and Closed-Loop Execution

[0092] Virtual verification: The optimal parameter combination (750W, 65%, 45℃, 1:20) was input into a digital twin constructed in the COMSOL Multiphysics platform. Simulation results show that at this power, the tank vibration frequency is within a safe range, and there is no safety hazard caused by excessively rapid ethanol evaporation.

[0093] Physical execution: The MES system generates the recipe version. Parameters are written to the PLC address via the OPC UA protocol.

[0094] 5. Comparison of Implementation Results

[0095] Compared with traditional fixed empirical parameters (500W, 80%, 60℃, 1:40), the extraction rate is improved: the total flavonoid yield in pear peel residue increases from 1.2% to 1.45%; energy consumption is reduced: due to the lower extraction temperature and targeted shortening of time, the energy consumption per batch decreases by 18%; solvent is saved: the concentration of ethanol used is reduced, and the solvent consumption is reduced by 12% by optimizing the material-liquid ratio.

[0096] Example 2: Multi-objective dynamic optimization of Polygonatum polysaccharide extraction process

[0097] Polygonatum ( Polygonatum sibiricum Polysaccharides are the core active ingredients in functional beverages and oral liquids. Traditional extraction methods typically use fixed parameters (such as constant high temperature and long-term decoction), which not only consumes a lot of energy but also makes it difficult to cope with batch-to-batch differences in raw materials from different origins, often resulting in large fluctuations in extraction rates and severe degradation of heat-sensitive components.

[0098] This embodiment utilizes the aforementioned multi-objective optimization process intelligence agent to perform "one-click" dynamic optimization for the Polygonatum polysaccharide extraction process. The specific steps are as follows:

[0099] 1. Collection and standardization of multi-source raw material data

[0100] Dried rhizomes of Polygonatum were selected as raw materials, and samples were collected from 10 batches from major domestic producing areas, including southeastern Guizhou, Zhaotong in Yunnan, Loudi in Hunan, and Huangshan in Anhui. Upon warehousing of the raw materials, key physicochemical data were entered into the LIMS system: Continuous characteristics: moisture content (%), crude polysaccharide content (%), initial pH value, reducing sugar content, etc. Discrete characteristics: origin of raw materials (e.g., "southern Guizhou"), harvest season (e.g., "autumn harvest").

[0101] Data preprocessing: For continuous features, the Z-score standardization formula is used to eliminate the influence of dimensions.

[0102]

[0103] Where x is a certain physicochemical index of the sample, μ is the average value of the physicochemical index of all samples, and σ is the standard deviation of the physicochemical index of all samples.

[0104] For discrete features (origin, season), one-hot encoding is used to generate sparse feature vectors.

[0105] The standardized continuous features and the discrete sparse vectors encoded by one-hot encoding are concatenated sequentially to generate the standardized raw material feature vector V. polygonatum .

[0106] 2. Three-dimensional matrix construction and initialization of the extraction process:

[0107] The system establishes an N×M×P three-dimensional process matrix. This embodiment focuses on the intelligent configuration of the "polysaccharide extraction section". Matrix structure definition:

[0108] X-axis (equipment node): locked at the extraction section.

[0109] Y-axis (equipment type): Activate the "ultrasonic assisted extraction tank" according to the production line configuration.

[0110] Z-axis (process parameters): Maps the four key adjustable control parameters of the equipment: ultrasonic power (P1): range [100, 800] W; extraction time (P2): range [5, 30] min; ethanol concentration (P3): range [40, 100]%; extraction temperature (P4): range [30, 60] ℃.

[0111] Matrix initialization strategy: An initial matrix library is constructed based on historical experimental datasets (containing the extraction effects of raw materials from different origins under different processes). For missing process points, Design of Experiments (DOE) methods (such as Box-Behnken design) are used to fill in the initial data for pre-training the agent model, ensuring that the agent has basic predictive capabilities in the initial stage.

[0112] 3. Proxy model prediction and multi-objective evolutionary optimization

[0113] The raw material feature vector V polygonatum Combine with the 3D process matrix and input the proxy model.

[0114] Multi-objective optimization settings: Three optimization objectives are set: Maximize extraction rate (Y): Improve the utilization rate of core raw materials. Minimize energy consumption (E): Reduce electricity and steam consumption. Minimize time consumption (T): Improve production line turnover efficiency. Simultaneously, a penalty function is introduced to prevent component degradation: When the temperature > 80℃ and the time > 20 min, the fitness function applies a heavy penalty to simulate a thermosensitive protection mechanism.

[0115] Intelligent Optimization Process: LIMS data shows that a batch of raw materials from "Zhaotong, Yunnan" has a dense texture and low moisture content. Traditional Pain Points: Traditional fixed processes (60℃, 60min) cause gelatinization of the outer polysaccharide layer, hindering internal dissolution, and prolonged high temperatures lead to partial polysaccharide hydrolysis. Intelligent Agent Decision-Making: After receiving the raw material characteristics, the process agent uses a surrogate model as the objective function evaluator and performs a global search within the process parameter space using a genetic algorithm. During the search, the surrogate model outputs target predicted values ​​such as extraction rate and energy consumption, while simultaneously calculating the corresponding uncertainty index. The uncertainty threshold is set according to the allowable error of the process system, preferably approximately 8% in this embodiment. For parameter combinations with uncertainties exceeding the preset threshold, the system automatically triggers CFD simulation for high-precision recalculation and returns the results to the surrogate model. Guided by the surrogate model, the genetic algorithm avoids the high temperature and long time range and converges to a new set of parameters: high ultrasonic power (700W): using strong mechanical waves to break the tight fiber structure and accelerate dissolution; medium and low temperature (45℃): avoiding polysaccharide thermal degradation; short time (18min): achieving the target extraction rate with a significant reduction in energy consumption.

[0116] 4. Digital Twin and MES Closed-Loop Execution Logic Verification: The optimal parameter combination (700W, 18min, 45℃, 60% ethanol) passed digital twin verification. The system confirmed that the high-power ultrasound would not cause the tank's resonant frequency to exceed the limit, and the 18-minute cycle time perfectly matched the downstream concentration section, with no risk of blockage. Closed-Loop Distribution: The MES system generates the batch formula. Parameters are automatically written to the PLC register.

[0117] 5. Implementation Results The application of this patented method solves three major challenges in the extraction of Polygonatum polysaccharides: Improved quality and stable yield: The standard deviation of extraction rates between batches of raw materials from different origins is reduced by 65%, achieving "a thousand materials, a thousand processes, but consistent quality." Cost reduction and efficiency improvement: Compared with traditional fixed processes, the extraction time per batch is shortened by 40% (from 30 minutes to approximately 18 minutes), and overall energy consumption is reduced by 25%. Activity retention: Due to the avoidance of prolonged high temperatures, the average molecular weight distribution of polysaccharides in the extract is more concentrated, and the retention rate of heat-sensitive functional components is increased by 15%.

[0118] Example 3: Multi-objective dynamic control of UHT instantaneous sterilization of fresh milk

[0119] Ultra-high temperature (UHT) sterilization of fresh milk is a core step in liquid milk production. Traditional production typically employs an "over-sterilization" strategy (such as setting a fixed temperature of 137°C for 4 seconds) to ensure shelf-life safety. However, this often leads to elevated levels of heat-sensitive indicators such as furosemine, resulting in a "cooked" taste, and also results in significant energy waste. Furthermore, as production progresses, milk residue (fouling) forms on the heat exchanger surface, causing a dynamic decrease in the heat transfer coefficient, making it difficult to maintain a constant sterilization intensity with fixed parameters.

[0120] This embodiment applies the aforementioned intelligent agent system to perform dynamic balance optimization of "safety-quality-energy consumption" in the UHT section. The specific steps are as follows:

[0121] 1. Real-time collection and standardization of raw milk data

[0122] After each batch of raw milk enters the storage tank, the system automatically obtains the physicochemical indicators of that batch from LIMS:

[0123] Key physicochemical indicators: fat content (Fat%), protein content (Protein%), non-fat milk solids (SNF), initial total bacterial count (CFU / mL), and acidity (°T).

[0124] Data preprocessing: Z-score standardization was performed on the continuous variables mentioned above. "Production runtime" was introduced as a special input feature because thermal efficiency decreases non-linearly as runtime increases, leading to thicker scale buildup.

[0125] 2. Three-dimensional matrix construction of the UHT sterilization section

[0126] The system initializes a three-dimensional process matrix with dimensions N×M×P and locks the "sterilization section (Node X=5)".

[0127] X-axis (equipment node): sterilization section.

[0128] Y-axis (equipment type): The system activates "Tube Sterilizer Group B (Device Y=2)" according to the shift production schedule.

[0129] Z-axis (process parameters): Maps the key adjustable parameters in the PLC of this equipment: sterilization temperature (P1): range [135.0, 142.0] ℃; material flow rate (P2): range [5000, 12000] L / h; homogenization pressure (P3): range [15, 25] MPa; back pressure (P4): range [2.0, 4.0] bar.

[0130] 3. Considering the scaling effect, the surrogate model prediction is to sequentially concatenate the "raw milk physicochemical data characteristics + runtime" and the "three-dimensional process matrix" and input them into the surrogate model.

[0131] Multi-objective optimization settings:

[0132] Maximize aseptic assurance (F0 value): Ensure F0 > 5.0 to meet commercial aseptic requirements; Minimize heat damage (furfurine value): Ensure furfurine < 12 mg / 100g protein to maintain fresh taste; Minimize steam consumption (E): Reduce steam consumption per ton of milk.

[0133] Uncertainty-driven active resampling:

[0134] After the equipment has been running continuously for 4 hours, due to scaling on the heat exchange tube walls, the uncertainty in the surrogate model's prediction of the "actual material heating temperature" exceeds a threshold. This uncertainty threshold is set based on the allowable error of the process system, and in this embodiment, it is preferably about 5%. The system automatically triggers a CFD simulation to simulate the "tube wall scaling thermal resistance model" under the current protein content. The simulation results show that the current scaling layer has caused a 5% decrease in heat transfer efficiency. The system feeds this "true value" back to the model, which then adjusts its prediction logic, no longer blindly trusting the set temperature, but instead predicting the actual temperature obtained by the fluid.

[0135] 4. Evolutionary Optimization and Decision Making

[0136] Based on the revised model, the genetic algorithm (GA) generates the optimal combination of process parameters:

[0137] Original strategy: Set 137℃ and flow rate 8000 L / h.

[0138] New strategy: The system detected that the raw milk in this batch had a low background of microorganisms and good thermal stability; the agent decided to reduce the sterilization temperature to 135.5℃ and increase the flow rate to 8500 L / h.

[0139] Expected results: While ensuring that F0 meets the standards, furosine production will decrease by 15% and production capacity will increase by 6%.

[0140] 5. Digital twin verification and PLC distribution

[0141] Virtual verification: The digital twin simulates the entire process of fluid passing through the heated tube.

[0142] Verification point 1: Confirm that the F0 value calculated by integrating the fluid centerline temperature meets the safety standards.

[0143] Verification Point 2: Verify that under high flow rates, the pipeline pressure drop will not exceed the pump's head limit. The maximum allowable head of the transfer pump is approximately 30 m of water column, corresponding to a pressure of approximately 0.30 MPa. Considering safety redundancy in industrial operation, the system sets the upper limit of the pump's effective allowable pressure drop to 80% of the maximum head, i.e., approximately 0.24 MPa.

[0144] Closed-loop execution:

[0145] After successful verification, the MES generates the recipe and writes it into the Siemens S7-1500 PLC via OPC UA. The PID setting of the steam regulating valve is updated to 135.5℃, and the frequency of the feed variable frequency pump is updated to correspond to a flow rate of 8500 L / h.

[0146] 6. Implementation Results

[0147] After introducing this method, product quality is improved: the furosine content in finished milk is reduced by an average of 12%-18%, significantly improving the "freshness" and taste of the product and eliminating the burnt taste caused by overheating; the operating cycle is extended: by dynamically adjusting the homogenization pressure and temperature difference, the scaling rate of the heat exchanger is slowed down, and the single continuous production time (CIP cycle) is extended by 1.5 hours; energy saving and consumption reduction: overheating is avoided, and the steam consumption per unit product is reduced by 8%.

[0148] Example 4: Alternative Scheme for Process Optimization Based on Particle Swarm Optimization (PSO)

[0149] To verify the universality of the "three-dimensional process matrix + surrogate model" architecture proposed in this invention, this embodiment uses the multi-objective particle swarm optimization algorithm (MOPSO) instead of the genetic algorithm in the previous embodiment to perform optimization tests on the same pear peel and residue extraction process.

[0150] The difference in the specific implementation methods lies in:

[0151] Encoding method: The equipment selection (discrete variable on the Y-axis) and process parameters (continuous variable on the Z-axis) in the three-dimensional process matrix are encoded as position vectors of "particles". Among them, the discrete variables adopt binary particle swarm optimization (BPSO) logic, and the velocity is mapped to the selection probability of [0,1] through the sigmoid function; the continuous variables adopt standard PSO logic.

[0152] Update mechanism: Instead of performing crossover and mutation operations, the velocity and position of particles are updated based on the individual historical best solution (Pbest) and the global best solution (Gbest).

[0153] Speed ​​update formula:

[0154] Position update formula:

[0155] In the above velocity update formula and position update formula, and These are the velocity component and position component of the i-th particle in the d-th iteration, respectively. Inertial weights are used to balance the algorithm's global search and local exploitation capabilities. and The learning factor (acceleration constant) represents the particle's direction of motion. and Learning step size weights; and A random number that follows a uniform distribution in [0,1].

[0156] Optimization Results: Under the same surrogate model and mutual exclusion masking mechanism, the PSO algorithm can also converge to the Pareto front. Tests show that compared with the genetic algorithm, PSO improves the convergence speed by about 15% when dealing with continuous parameters (such as temperature and power), proving that the system architecture proposed in this invention is compatible with multiple global optimization algorithms.

[0157] Those skilled in the art should understand that although this embodiment uses a genetic algorithm (GA) as an example to illustrate the multi-objective optimization process in detail, the present invention is not limited thereto. Without departing from the core idea of ​​the present invention (i.e., based on a three-dimensional matrix structure and a closed-loop surrogate model), any heuristic algorithm or machine learning algorithm with global optimization capabilities can replace the genetic algorithm.

[0158] Specifically, alternative algorithms include, but are not limited to:

[0159] Swarm intelligence algorithms include: Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), Artificial Bee Colony Optimization (ABC), and Grey Wolf Optimization (GWO).

[0160] Evolutionary algorithm variants: such as differential evolution (DE), evolutionary strategy (ES), simulated annealing (SA);

[0161] Learning-based algorithms include Deep Reinforcement Learning (DRL), Deep Q-Network (DQN), and Proximal Policy Optimization (PPO).

[0162] Model-based algorithms include Bayesian optimization (BO) and tree-based Pareto estimation (TPE).

[0163] Any technical solution that uses the above algorithm to iteratively update the three-dimensional process matrix described in this invention and combines a mutual exclusion mask mechanism and a proxy model for evaluation should be covered within the protection scope of this invention.

[0164] This application also provides a control device for a multi-objective optimization process intelligence agent based on food production, such as... Figure 5 As shown, the device includes: a data acquisition and standardization module 301, a three-dimensional process matrix construction module 302, an intelligent decision-making core module 303, a simulation verification module 304, and an execution and feedback module 305.

[0165] The data acquisition and standardization module 301 is used to acquire the characteristic data of the target batch of raw materials through the laboratory information management system, and to standardize and encode the characteristic data to generate the standardized feature vector of the target batch of raw materials.

[0166] The three-dimensional process matrix construction module 302 is used to establish a three-dimensional process matrix that characterizes the physical topology of the production line.

[0167] The intelligent decision-making core module 303 is communicatively connected to the data acquisition and standardization module and the three-dimensional process matrix construction module. It includes a global optimization algorithm engine, a feature mask encoder and a proxy model, and is used to realize process optimization based on mask mapping and active learning.

[0168] The simulation verification module 304 is connected to the intelligent decision-making core module and is used to perform virtual simulation and verification of the control logic, timing and performance of the optimal process configuration scheme through a digital twin model.

[0169] The execution and feedback module 305 includes an MES terminal and a PLC control network, which is used to send the verified process configuration scheme to the production system for execution and to collect production data to update the agent model.

[0170] Although embodiments of this application have been shown and described above, it is understood that these embodiments are merely illustrative and do not constitute a limitation on the scope of protection of this application. Without departing from the essential spirit of this application, those skilled in the art can make various changes, modifications, substitutions, or adjustments to the above embodiments within the scope defined by this application. It should be noted that although the embodiments herein are illustrated using the field of food raw materials as an example, the ideas and methods are equally applicable to other process optimization fields with similar characteristics.

Claims

1. A control method for a multi-objective optimization process intelligent agent based on food production, characterized in that, Includes the following steps: S1. Obtain the physicochemical characteristic data of the target batch of raw materials and perform standardization processing to generate a standardized feature vector; S2. Establish a three-dimensional process matrix that represents the physical topology of the production line, and generate a binary mask tensor with the same structure as the three-dimensional process matrix based on the selection state of the equipment unit in the three-dimensional process matrix. Based on the mask tensor, the standardized feature vector is concatenated with the expanded three-dimensional process matrix to obtain the concatenated vector; the Hadamard product of the concatenated vector and the expanded binary mask tensor is calculated to filter out the features corresponding to the unselected devices and generate the model input feature vector. S3. Input the input feature vector into the pre-trained surrogate model to obtain the predicted values ​​of multi-dimensional process indicators and the uncertainty of the prediction; the pre-trained surrogate model is an ensemble learning model or a multi-task deep neural network model obtained by training through historical production data and offline fluid dynamics simulation database; wherein, when the uncertainty of the prediction exceeds a preset threshold, computational fluid dynamics simulation is automatically triggered to perform physical field supplementation, and the supplementation result is used as the true value to update the surrogate model; S4. If the prediction uncertainty is lower than a preset threshold, then a global optimization algorithm is executed based on the prediction value, and iterative optimization is performed until convergence and the optimal process configuration is output. S5. The optimal process configuration is virtually verified using a digital twin model; S6. The verified process configuration scheme is sent to the production system for execution, and production data is collected to update the proxy model; The three-dimensional process matrix is ​​an N×M×P tensor structure, where N corresponds to the physical section node of the production line, M corresponds to the optional equipment unit physically connected under each node, and P corresponds to the readable and writable process register address in the PLC controller of each equipment unit.

2. The control method for a multi-objective optimization process intelligent agent based on food production according to claim 1, characterized in that, Step S1 specifically involves: acquiring the physicochemical characteristic data of the target batch of raw materials, the physicochemical characteristic data including continuous physicochemical characteristic data and discrete characteristic data; performing Z-score standardization on the continuous physicochemical characteristic data; performing one-heat encoding on the discrete characteristic data; and concatenating the standardized continuous features and the encoded discrete features in a predetermined order to generate the standardized feature vector.

3. The control method for a multi-objective optimization process intelligent agent based on food production according to claim 1, characterized in that, In step S2, the value of the selected device unit and its associated entire process parameter space at the corresponding position in the mask tensor is 1, and the value of the unselected device unit and its associated parameter space at the corresponding position is 0.

4. The control method for a multi-objective optimization process intelligent agent based on food production according to claim 1, characterized in that, In step S4, the global optimization algorithm is one of the following: genetic algorithm, particle swarm optimization algorithm, differential evolution algorithm, reinforcement learning, or Bayesian optimization.

5. An apparatus for implementing the control method of a multi-objective optimization process intelligence agent based on food production as described in any one of claims 1-4, characterized in that, include: The data acquisition and standardization module is configured to acquire feature data of the target batch of raw materials, and to standardize and encode the feature data to generate a standardized feature vector of the target batch of raw materials. The 3D process matrix construction module configures a 3D process matrix to characterize the physical topology of the production line. The intelligent decision-making core module is communicatively connected to the data acquisition and standardization module and the three-dimensional process matrix construction module. It includes a global optimization algorithm engine, a feature mask encoder, and a proxy model, and is configured to realize process optimization based on mask mapping and active learning. The simulation verification module is connected to the intelligent decision-making core module and is configured to perform virtual simulation and verification of the control logic, timing and performance of the optimal process configuration scheme through a digital twin model. The execution and feedback module includes an MES terminal and a PLC control network, configured to send the verified process configuration scheme to the production system for execution, and collect production data to update the agent model.

Images