Abalone peptide production process optimization and intelligent decision-making method and system

Abstract

The invention relates to the technical field of abalone peptide production, and discloses an abalone peptide production process optimization and intelligent decision-making method and system, and the method comprises the steps: collecting production data through a multi-source sensor, and extracting features through a convolutional neural network; the genetic algorithm optimization model is combined with an adaptive mechanism and a multi-target fitness function to generate a process parameter optimization scheme; and the multi-objective production optimization model outputs optimal production process parameters by means of an improved particle swarm algorithm. In the hierarchical intelligent decision-making model, the planning layer realizes global resource allocation, the scheduling layer adjusts a process in real time based on a Bayesian network and the like, and the operation layer controls equipment through fuzzy logic and constructs quality closed-loop control. According to the method, the problems that parameter control depends on experience, data processing is difficult, optimization is one-sided, decision making is not intelligent and the like in traditional production are solved, data can be accurately processed, the process is optimized, decision making is intelligent, production efficiency, the raw material utilization rate and product quality are effectively improved, cost is reduced, and the application prospect is wide.

Description

Technical Field

[0001] The present invention relates to the technical field of abalone peptide production, and in particular to an abalone peptide production process optimization and intelligent decision-making method and system. Background Art

[0002] Abalone peptides are short peptides extracted from abalone with multiple biological activities, and have broad application prospects in the fields of food, medicine and health products. With the increasing market demand for abalone peptides, how to optimize its production process, improve production efficiency, raw material utilization and product quality has become a key research direction in this field.

[0003] In the traditional abalone peptide production process, the control of process parameters mainly relies on manual experience, which lacks precision and scientificity. For example, in the enzymatic reaction, the setting of parameters such as reaction temperature, pH value, enzyme addition amount and reaction time is often determined based on the operator's long-term accumulated experience. The difference in experience between different operators may lead to inconsistency in process parameter control, thereby affecting the production efficiency of abalone peptides and the stability of product quality. Moreover, manual experience makes it difficult to fully consider the interaction of various complex factors in the production process, and it is impossible to achieve the optimal configuration of process parameters.

[0004] At the same time, there are also major problems in the processing of multi-source sensor data. Although temperature sensors, pH sensors, enzyme activity sensors and other sensors are used to collect data during the production process, traditional data processing methods are difficult to effectively extract key information from these sensor data due to their high dimensionality, multimodality and dynamic changes. A large amount of redundant data not only increases the burden of data storage and transmission, but may also interfere with the accurate judgment of the production process, making it impossible to discover and solve potential problems in the production process in a timely manner.

[0005] In terms of production process optimization, most existing technologies use single-target optimization methods, such as simply pursuing the improvement of production efficiency or the improvement of raw material utilization, while ignoring other important factors. This one-sided optimization method may lead to the sacrifice of other indicators while improving a certain indicator, and fail to maximize production benefits. For example, excessively increasing the amount of enzyme to improve production efficiency may speed up the reaction, but it will lead to a decrease in raw material utilization, increase production costs, and may also have an adverse effect on product quality.

[0006] In addition, the application of intelligent decision-making in abalone peptide production is not mature enough. Current production decisions are often based on post-event data analysis and experience judgment, lacking real-time and foresight. When faced with emergencies or equipment failures in the production process, it is impossible to make effective decisions quickly, which can easily lead to production interruptions or product quality degradation. Moreover, traditional production control systems lack the ability to optimize and coordinate the production process globally, and information transmission between various production links is not smooth, making it difficult to achieve overall optimization of the production process. Summary of the invention

[0007] The object of the present invention is to provide a method and system for optimizing the production process of abalone peptides and intelligent decision-making, so as to solve the problems raised in the above-mentioned background technology.

[0008] To achieve the above object, the present invention provides the following technical solution: an abalone peptide production process optimization and intelligent decision-making method, the method comprising: Collecting real-time process data in the abalone peptide production process through multi-source sensors, wherein the multi-source sensors include a temperature sensor, a pH sensor, an enzyme activity sensor, and a pressure sensor; performing multi-scale feature extraction on the real-time process data based on a convolutional neural network to obtain process feature data; The process characteristic data is input into a pre-trained genetic algorithm optimization model, wherein the genetic algorithm optimization model adopts an adaptive crossover rate and mutation rate mechanism, iteratively optimizes the process parameter combination based on a multi-objective fitness function, and generates a process parameter optimization plan; Constructing a multi-objective production optimization model according to the process parameter optimization scheme, wherein the multi-objective production optimization model takes maximizing production efficiency and maximizing raw material utilization as optimization objectives, and adopts an improved particle swarm algorithm to globally optimize the production process, wherein the improved particle swarm algorithm introduces an inertia weight dynamic adjustment mechanism and a neighborhood search strategy; outputting optimal production process parameters based on the multi-objective production optimization model; A hierarchical intelligent decision-making model is established according to the optimal production process parameters, and the hierarchical intelligent decision-making model includes a planning layer, a scheduling layer and an operation layer, wherein the planning layer performs global resource allocation planning based on the process parameter optimization plan, the scheduling layer performs dynamic parameter scheduling based on the optimal production process parameters, and the operation layer realizes the execution control of the production equipment based on the fuzzy logic control algorithm; the production control instructions are output through the hierarchical intelligent decision-making model to realize the optimization and intelligent decision-making of the abalone peptide production process.

[0009] Preferably, the process characteristic data is input into a pre-trained genetic algorithm optimization model, the genetic algorithm optimization model adopts an adaptive crossover rate and mutation rate mechanism, and iteratively optimizes the process parameter combination based on a multi-objective fitness function, and generates a process parameter optimization scheme including: Acquire real-time process data, wherein the real-time process data includes reactor temperature, enzymatic solution pH value, enzyme activity concentration, raw material input rate and reaction pressure; construct a parameter space based on the real-time process data, and construct a decision space based on the process adjustable temperature change, pH adjustment amount, enzyme addition amount and pressure adjustment amount; A multi-objective fitness function is constructed based on the parameter space and the decision space, wherein the multi-objective fitness function includes a production efficiency item, a raw material utilization item, an energy consumption item, and a stability item, wherein the production efficiency item is calculated by the ratio of the product output per unit time to the target output, the raw material utilization item is calculated by the ratio of the actual product quality to the theoretical maximum product quality, the energy consumption item is calculated by the weighted square sum of the temperature adjustment amount, the pH adjustment amount, and the pressure adjustment amount, and the stability item is calculated by the Euclidean distance between the current process parameter and the historical optimal parameter; Construct an adaptive crossover rate and mutation rate mechanism, wherein the adaptive crossover rate is dynamically adjusted with the population diversity index, and the mutation rate changes exponentially with the increase in the number of iterations; select parent individuals based on the elite retention strategy, generate offspring individuals through crossover operations, and perform mutation operations on the offspring individuals; The non-dominated sorting method is used to stratify the population individuals, and the Pareto frontier solution set is screened based on the crowding distance. The optimal solution in the Pareto frontier solution set that satisfies the trade-off between production efficiency and raw material utilization is used as the process parameter optimization solution.

[0010] Preferably, the method for outputting optimal production process parameters based on the multi-objective production optimization model includes: Constructing a multi-objective optimization function, wherein the multi-objective optimization function includes a production efficiency objective function and a raw material utilization objective function, wherein the production efficiency objective function is calculated by the ratio of reaction time to product output, and the raw material utilization objective function is calculated by the ratio of raw material consumption to product quality; Constructing constraint conditions based on the multi-objective optimization function, wherein the constraint conditions include temperature range constraint, pH range constraint, enzyme activity threshold constraint and pressure safety threshold constraint; The particle swarm algorithm is used to encode the production process parameters. Each particle represents a set of process parameter combinations. The parameter values ​​are adjusted based on the particle position update formula. The inertia weight dynamic adjustment mechanism adjusts the weight value according to the particle convergence degree. The neighborhood search strategy guides the population update through the local optimal particle. A dynamic learning factor is introduced, and the dynamic learning factor changes piecewise linearly with the increase of the number of iterations, and the contribution weights of the global optimal particle and the local optimal particle to the speed update are adjusted by the dynamic learning factor; Based on the improved particle swarm algorithm, iterative optimization is performed, the production efficiency and raw material utilization rate of the parameter combination generated in each iteration are evaluated, the non-dominated solution is stored in the optimization solution set, and finally the parameter combination with the highest comprehensive score in the Pareto front is selected as the optimal production process parameter.

[0011] Preferably, the planning layer performs global resource allocation planning based on the process parameter optimization solution, including: A virtual production environment model is constructed using digital twin technology. The virtual production environment model is driven by real-time process data and equipment status data to simulate the dynamic behavior of the reactor and the material flow process. A resource allocation model is constructed based on a mixed integer programming method, wherein the resource allocation model aims to maximize equipment utilization and minimize energy consumption, and the constraints include raw material supply cycle, equipment capacity limit, and production batch continuity; The production task is decomposed into multiple subtasks, and the resource allocation matrix is ​​generated based on the task priority and equipment load status. The resource allocation problem is solved through the Lagrangian relaxation algorithm to generate a global equipment scheduling plan and raw material delivery plan.

[0012] Preferably, the scheduling layer performs real-time process adjustment based on the optimal production process parameters, including: A dynamic process adjustment model is constructed using a Bayesian network, which uses real-time sensor data as observation nodes, process parameter adjustment amounts as decision nodes, and describes the dependency relationship between parameters through a conditional probability table; A state transition model is constructed based on the Markov decision process. The state space is defined as the current process parameters and equipment status, and the action space is defined as the adjustment amount of temperature, pH value and pressure. The optimal adjustment strategy is solved through the value iteration algorithm. A sliding time window mechanism is introduced to perform weighted average processing on historical process data, eliminate instantaneous noise interference, and generate smoothed process parameter correction instructions.

[0013] Preferably, the operation layer realizes precise control of production equipment based on fuzzy logic control algorithm, including: Establishing a multi-input multi-output fuzzy rule base for production equipment, wherein the input variables of the fuzzy rule base include temperature deviation, pH deviation and pressure deviation, and the output variables of the fuzzy rule base include heating power, acid-base pump flow and pressure valve opening; The input and output variables are fuzzified using triangular membership functions, and the fuzzy rules are defined as IF-THEN forms. The precise control quantity is obtained by defuzzification using the centroid method. An anti-saturation compensation mechanism is constructed. When the control quantity exceeds the physical limit of the equipment, a proportional-integral-differential auxiliary controller is used to compensate for the error to ensure the stable output of the actuator.

[0014] Preferably, the above-mentioned abalone peptide production process optimization and intelligent decision-making method also includes: A quality prediction model is constructed based on a long short-term memory (LSTM) network. The quality prediction model takes historical process data and real-time sensor data as input and outputs the distribution of product peptide chain length and the predicted value of biological activity index. The attention mechanism is used to enhance the feature weights of key process parameters, local time series patterns are extracted through time series sliding windows, and quality prediction results are generated in combination with global process status; The quality prediction results are fed back to the scheduling layer to dynamically correct the process adjustment strategy to achieve closed-loop quality control.

[0015] Preferably, the virtual production environment model is constructed by the following steps: The finite element method was used to model the fluid dynamics in the reactor, and the temperature field and concentration field distribution were solved by the Navier-Stokes equation. The collision and dissolution process of the raw material particles was simulated based on the discrete element method, and the product generation rate was calculated in combination with the enzymatic reaction kinetic equation. The model parameters are updated through real-time data assimilation technology, and the Kalman filter algorithm is used to fuse sensor data and simulation results.

[0016] Preferably, the training of the Bayesian network includes: The expectation maximization (EM) algorithm is used to fill in the missing process data, and the posterior distribution of latent variables is estimated by the Gibbs sampling method; Based on KL divergence, the network structure learning objective function is constructed, and the node connection relationship is optimized using a greedy search strategy to generate a network topology with the minimum description length. The network uncertainty is evaluated through the Monte Carlo dropout method, and the confidence threshold of the process adjustment strategy is dynamically adjusted.

[0017] Preferably, the present invention also includes an abalone peptide production process optimization and intelligent decision-making system, which is used to implement the above-mentioned abalone peptide production process optimization and intelligent decision-making method, including: Data acquisition and feature extraction module, which is used to collect real-time process data through multi-source sensors, extract process feature data based on convolutional neural network, and generate process parameter optimization scheme through genetic algorithm optimization model; A production process optimization module is used to construct a multi-objective production optimization model according to the process parameter optimization scheme, and output the optimal production process parameters using an improved particle swarm algorithm; The intelligent decision-making execution module is used to realize global resource allocation, real-time process adjustment and precise equipment control based on the hierarchical intelligent decision-making model, and output production control instructions.

[0018] Compared with the prior art, the present invention has the following beneficial effects: By using multi-source sensors to collect process data in the abalone peptide production process in real time and using convolutional neural networks to extract multi-scale features, key information can be accurately extracted from high-dimensional, multi-modal dynamic data. Compared with the problem that traditional data processing methods are susceptible to redundant data interference, this invention not only reduces the burden of data storage and transmission, but also provides accurate and reliable data support for subsequent process optimization and decision-making, greatly improving the accuracy of judgment on the production process.

[0019] The genetic algorithm optimization model adopts an adaptive crossover rate and mutation rate mechanism, combined with a multi-objective fitness function, which can comprehensively integrate multiple factors such as production efficiency, raw material utilization, energy consumption and stability, and iteratively optimize the process parameter combination. Different from the traditional method of setting process parameters based on manual experience and failing to fully consider the interaction of complex factors, the present invention can generate a better process parameter optimization scheme, achieve a balance between various production indicators, significantly improve production efficiency and raw material utilization, while reducing energy consumption and production costs.

[0020] A multi-objective production optimization model with the goal of maximizing production efficiency and raw material utilization was constructed, and an improved particle swarm algorithm was used for global optimization. The algorithm introduced a dynamic adjustment mechanism for inertia weight, a neighborhood search strategy, and a dynamic learning factor, and was able to output optimal production process parameters under various constraints such as temperature, pH value, enzyme activity, and pressure. This changed the limitations of single-objective optimization in existing technologies, achieved overall optimization of the production process, effectively avoided the problem of indicator imbalance caused by one-sided optimization, and further improved production efficiency.

[0021] The planning layer of the hierarchical intelligent decision-making model uses digital twin technology and mixed integer programming methods to achieve reasonable allocation and planning of global resources, generate scientific equipment scheduling plans and raw material delivery plans, and ensure the orderly progress of production; the scheduling layer is based on Bayesian networks and Markov decision processes, combined with a sliding time window mechanism, and can dynamically and accurately adjust process parameters according to real-time data to respond to uncertainties in the production process in a timely manner; the operation layer uses fuzzy logic control algorithms to achieve precise control of production equipment, and builds an anti-saturation compensation mechanism to ensure stable operation of the equipment. In addition, the quality prediction model combines attention mechanisms and time series analysis to feed back quality prediction results to the scheduling layer to form a closed-loop quality control. The entire intelligent decision-making and control system realizes real-time and intelligent regulation of the production process, improves production stability and product quality, and has stronger real-time and forward-looking features than traditional decision-making methods based on post-analysis and experience judgment. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A working principle diagram of the abalone peptide production process optimization and intelligent decision-making method of the present invention; Figure 2 Working principle diagram for generating process parameter optimization scheme for genetic algorithm optimization model; Figure 3 Schematic diagram of the working principle of outputting optimal production process parameters for a multi-objective production optimization model. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] See also Figure 1-3 The present invention provides a technical solution: an abalone peptide production process optimization and intelligent decision-making method, the method comprising: Using multi-source sensors such as temperature sensors, pH sensors, enzyme activity sensors, and pressure sensors, the process data of abalone peptide production is collected in real time, such as reactor temperature, enzymatic solution pH, enzyme activity concentration, raw material input rate, and reaction pressure. The collected real-time process data is input into the convolutional neural network, and multi-scale feature extraction is performed on the data through the convolutional layer, pooling layer and other components of the convolutional neural network to obtain process feature data that can reflect the key information of the production process.

[0025] The extracted process feature data is input into the pre-trained genetic algorithm optimization model. The model adopts an adaptive crossover rate and mutation rate mechanism, dynamically adjusts the crossover rate according to the population diversity index, and makes the mutation rate exponentially decay with the increase of the number of iterations. Based on the multi-objective fitness function including production efficiency, raw material utilization, energy consumption and stability, the process parameter combination is iteratively optimized. After multiple rounds of selection, crossover and mutation operations, the non-dominated sorting method is used to stratify the population individuals, and the Pareto frontier solution set is screened based on the crowding distance, and the optimal solution that satisfies the production efficiency-raw material utilization trade-off is selected as the process parameter optimization solution.

[0026] According to the generated process parameter optimization scheme, a multi-objective production optimization model with the optimization objectives of maximizing production efficiency and maximizing raw material utilization is constructed. The improved particle swarm algorithm is used to globally optimize the production process, which introduces a dynamic adjustment mechanism of inertia weight and a neighborhood search strategy. By constructing a multi-objective optimization function including the production efficiency objective function and the raw material utilization objective function, and setting conditions such as temperature range constraints, pH range constraints, enzyme activity threshold constraints, and pressure safety threshold constraints, the production process parameters are encoded and iteratively optimized, and the optimal production process parameters are finally output.

[0027] A hierarchical intelligent decision-making model is established based on the optimal production process parameters. The model includes a planning layer, a scheduling layer, and an operation layer. The planning layer uses digital twin technology to build a virtual production environment model, and builds a resource allocation model based on a mixed integer programming method. The production task is decomposed into multiple subtasks, and a global equipment scheduling plan and a raw material delivery plan are generated to achieve global resource allocation planning. The scheduling layer uses a Bayesian network to build a dynamic process adjustment model, builds a state transition model based on a Markov decision process, and introduces a sliding time window mechanism to achieve real-time process adjustment. The operation layer establishes a multi-input and multi-output fuzzy rule library for production equipment, uses a triangular membership function to fuzzify the input and output variables, and obtains precise control quantities through the centroid method. An anti-saturation compensation mechanism is constructed to achieve precise control of production equipment. The production control instructions are output through a hierarchical intelligent decision-making model to achieve optimization and intelligent decision-making of the abalone peptide production process. In addition, a quality prediction model is built based on a long short-term memory (LSTM) network, and an attention mechanism is used to enhance the feature weights of key process parameters. The quality prediction results are fed back to the scheduling layer, and the process adjustment strategy is dynamically corrected to achieve closed-loop quality control.

[0028] The present invention will be further described below in conjunction with Examples 1 to 5: Embodiment 1:

[0029] This embodiment describes in detail the specific process of generating a process parameter optimization solution by a genetic algorithm optimization model. By constructing a reasonable parameter space, decision space and multi-objective fitness function, and adopting an adaptive crossover rate and mutation rate mechanism and an elite retention strategy, it is ensured that a process parameter optimization solution that meets production needs can be obtained, which specifically includes: (1) Constructing parameter space and decision space: Real-time collection of reactor temperature during abalone peptide production , pH value of enzymatic solution , enzyme activity concentration , Raw material input rate and reaction pressure These data constitute real-time process data. Parameter space is constructed based on these real-time process data. The temperature change that can be adjusted according to the process , pH adjustment amount , enzyme addition amount And pressure adjustment Construct a decision space. For example, at a certain moment, the reactor temperature According to production experience and process requirements, the temperature can be adjusted within the range of , then the temperature change The value range of is [-5, 5], and the other parameters are similarly determined to determine their respective ranges of variation, thereby constructing a complete decision space.

[0030] (2) Constructing a multi-objective fitness function: Production efficiency item: Production efficiency item Product output per unit time With target output The ratio of is calculated as follows: For example, over a period of time The actual product output is 50 grams and the target output is 80 grams. The production efficiency item in this time period is .

[0031] Raw material utilization item: Raw material utilization item The actual product quality Theoretical maximum product mass The ratio of is calculated as follows: Assuming that the theoretical maximum product mass is 100 grams according to the composition of the raw materials and the chemical reaction formula, and the actual product mass is 70 grams, then the raw material utilization rate term .

[0032] Energy consumption items: Energy consumption items Adjust the amount by temperature , pH adjustment amount And pressure adjustment The weighted square sum of is calculated as follows: ,in , , is the weight coefficient, which is determined according to the degree of influence of each parameter on energy consumption in the production process. For example, after experiments and data analysis, it is determined , , , during an adjustment , , , then the energy consumption item .

[0033] Stability term: Stability term It is calculated by the Euclidean distance between the current process parameters and the historical optimal parameters. Assume that the current process parameter vector is , the historical optimal parameter vector is , then the stability term For example, the current process parameters are temperature , pH value is 7.2, and the temperature is the best parameter in history. , pH value is 7.0, then the stability term . Multi-objective fitness function Taking all the above into consideration, it can be expressed as ,in , , , The weight of each item is determined according to actual production needs to balance the relationship between different objectives.

[0034] (3) Adaptive crossover rate and mutation rate mechanism and genetic operation: Adaptive crossover rate: Adaptive crossover rate Dynamically adjusted with population diversity index. Population diversity index can be measured by calculating the degree of difference between individuals in a population, such as using methods such as Hamming distance of individual coding. When the population diversity is high, the crossover rate is appropriately reduced to avoid excessive destruction of excellent individuals; when the population diversity is low, the crossover rate is increased to promote the generation of new individuals. Assuming that the population diversity index The value is between 0 and 1, and the lower limit of the crossover rate is , the upper limit is , then the adaptive crossover rate calculation formula can be .For example, , , when the population diversity index hour, .

[0035] Mutation rate: Mutation rate With the number of iterations The increase is exponentially decaying, and the formula is ,in is the initial mutation rate is the attenuation coefficient, determined experimentally. For example, , , when the number of iterations hour, .

[0036] Genetic operation: Select parent individuals based on the elite retention strategy, that is, retain a part of individuals with higher fitness in the current population and directly enter the next generation. Then, perform a crossover operation on the selected parent individuals according to the adaptive crossover rate to generate offspring individuals. For example, using the single-point crossover method, randomly select two parent individuals, randomly select a crossover point on their code string, exchange the part after the crossover point, and generate two offspring individuals. Perform mutation operations on offspring individuals according to the mutation rate. For example, for binary-coded individuals, perform an inversion operation on each bit in the individual code with the mutation rate as the probability.

[0037] (4) Optimization scheme for screening process parameters: Use the non-dominated sorting method to stratify the individuals in the population. Non-dominated sorting is to divide the individuals in the population into different levels according to the dominance relationship. No less inferior than individuals on all goals , and is superior to individuals in at least one goal , then we call it individual Dominant Individual . Individuals that are not dominated by any other individuals are divided into the first layer. After removing the first layer individuals from the population, repeat the above operation for the remaining individuals to obtain the second layer, and so on. The Pareto frontier solution set is screened based on the crowding distance. The crowding distance is used to measure the degree of crowding of individuals in their level. When calculating the crowding distance of each individual, first calculate the sum of the distances between the individual and the adjacent individuals in each target dimension, and then normalize the sum of the distances of all target dimensions. Individuals with larger crowding distances are selected to form the Pareto frontier solution set, because these individuals are more evenly distributed in the target space and have better diversity. The optimal solution that satisfies the trade-off between production efficiency and raw material utilization is selected from the Pareto frontier solution set as the process parameter optimization solution. For example, according to actual production needs, a comprehensive evaluation index of production efficiency and raw material utilization can be set, such as , select from the Pareto frontier solution set The solution with the largest value is taken as the final process parameter optimization solution. Embodiment 2:

[0038] This embodiment describes in detail how to iteratively optimize production process parameters based on a multi-objective production optimization model by constructing optimization functions and constraints, using an improved particle swarm algorithm, combining an inertia weight dynamic adjustment mechanism, a neighborhood search strategy, and a dynamic learning factor, and outputting optimal production process parameters that can simultaneously meet the maximization of production efficiency and the maximization of raw material utilization, so as to provide accurate parameter guidance for actual production. The specific method includes: ①Construct a multi-objective optimization function: Production efficiency objective function: Production efficiency objective function By reaction time and product yield The ratio of is calculated as follows: For example, in a production process, the reaction time is 5 hours and the product output is 200 grams, then the production efficiency objective function value is g / h.

[0039] Raw material utilization objective function: Raw material utilization objective function By raw material consumption And product quality The ratio of is calculated as follows: Assuming that the mass of raw materials consumed is 300 grams and the mass of products obtained is 200 grams, the objective function value of raw material utilization is

[0040] ②Construction constraints: Temperature range constraint: According to the abalone peptide production process requirements, the temperature in the reactor The minimum temperature should be within a certain range. , the maximum temperature is , then the temperature range constraint is For example, the production process requires the temperature to be Between ,

[0041] pH range constraint: Enzymatic hydrolysis has specific requirements for pH value. The lower limit of the pH value of the enzymatic hydrolysis solution is , the upper limit is , then the pH range is constrained to For example, the suitable pH range is 6.5-7.5. ,

[0042] Enzyme activity threshold constraint: enzyme activity concentration A certain threshold must be reached to ensure the normal reaction. The enzyme activity threshold is , then the enzyme activity threshold constraint is . Assume that the enzyme activity threshold is determined experimentally to be , then the enzyme activity concentration during the production process must be greater than or equal to this value.

[0043] Pressure safety threshold constraint: response pressure The pressure safety threshold cannot exceed the safety tolerance of the equipment. , then the pressure safety threshold constraint is For example, if the pressure safety threshold of the equipment is 5MPa, then the reaction pressure Need to meet .

[0044] ③ Particle swarm algorithm coding and parameter adjustment: Coding: Particle swarm algorithm is used to encode the production process parameters. Each particle represents a set of process parameter combinations, such as temperature, pH value, enzyme addition amount, raw material input rate, etc. Assume that the position vector of each particle is ,in Indicates process parameters.

[0045] Particle position update formula: Particle position update formula is ,in is the particle velocity, and the calculation formula is . is the inertia weight, , is the learning factor, , is a random number between [0, 1], For particles The best historical position, is the global optimal position.

[0046] Dynamic adjustment mechanism of inertia weight: The inertia weight is adjusted according to the degree of particle convergence. When the particle converges quickly and approaches the local optimal solution, the inertia weight is reduced to make the particle focus more on local search; when the particle converges slowly, the inertia weight is increased to promote the particle global search. The degree of particle convergence can be measured by the distance variance between particles. The lower limit of inertia weight is , the upper limit is , then the inertia weight adjustment formula can be ,in , is the preset maximum and minimum distance variance. For example, , , , , when the current particle distance variance is calculated hour, .

[0047] Neighborhood search strategy: The neighborhood search strategy guides the population update through the local optimal particle. The entire population is divided into multiple neighborhoods, and each particle only communicates with other particles in the neighborhood. Find the local optimal particle in each neighborhood , when updating the particle speed, in addition to considering the global optimal particle , also consider the local optimal particle , that is, the speed update formula becomes ,in The learning factor for the neighborhood search, is a random number between [0, 1].

[0048] Introducing dynamic learning factors: Dynamic learning factors , With the number of iterations The increase is piecewise linear. In the early stage of iteration, it is hoped that particles are more inclined to global search. value, reduce In the later stages of iteration, we hope that particles will pay more attention to local search and increase value, reduce Assume that the total number of iterations is , the iterative process can be divided into The length of each segment is . Take the example of dividing into three sections: In the first paragraph hour, , t; In the second paragraph hour, , ; In the third paragraph hour, , .

[0049] in , , , is the pre-set minimum and maximum learning factor, e.g. , , , .

[0050] Iterative optimization and optimal solution selection: Iterative optimization is performed based on the improved particle swarm algorithm. In each iteration, the generated parameter combinations are used to calculate the production efficiency and raw material utilization rate according to the multi-objective optimization function. The non-dominated solutions that meet all constraints and are not dominated by other solutions are stored in the optimization solution set. After multiple iterations, the parameter combination with the highest comprehensive score in the Pareto frontier is selected as the optimal production process parameter. The comprehensive score can be used to assign different weights to production efficiency and raw material utilization rate according to actual production needs, such as ,in , The weight is determined according to the actual production situation, and the parameter combination with the largest Score value is selected as the final optimal production process parameter. Embodiment 3:

[0051] This embodiment describes in detail how the planning layer performs global resource allocation planning based on the process parameter optimization scheme. By building a virtual production environment model to simulate the production process, using the mixed integer programming method to build a resource allocation model, decomposing the production tasks and generating a resource allocation matrix, and finally obtaining a global equipment scheduling plan and a raw material delivery plan, the reasonable allocation and efficient use of production resources are achieved, ensuring the smooth progress of the production process.

[0052] Build a virtual production environment model: Modeling in the reactor: The finite element method is used to model the fluid dynamics in the reactor. According to the Navier-Stokes equation ,in is the fluid density, is the flow velocity vector, For time, For pressure, is the dynamic viscosity, is an external force. By solving this equation, the temperature field and concentration field distribution in the reactor are obtained, and the flow and reaction of the materials in the reactor are simulated. For example, after the geometric shape of the reactor, the initial conditions of the materials and the boundary conditions are given, the reactor is meshed using finite element software, the Navier-Stokes equations are discretized, and the distribution of the temperature field and concentration field at different times is obtained through numerical calculation.

[0053] Raw material particle simulation: Based on the discrete element method, the collision and dissolution process of raw material particles is simulated. The raw material particles are regarded as discrete individuals, and the interaction forces between particles, such as gravity, friction, collision force, etc., are considered to establish the particle motion equation. Combined with the enzymatic reaction kinetic equation, such as the Michaelis equation (in is the reaction rate, is the maximum reaction rate, is the substrate concentration, is the Michaelis constant), and the product generation rate is calculated. By simulating the movement, dissolution and reaction of raw material particles in the reactor, the material changes in the production process can be more accurately predicted.

[0054] Model parameter update: Update model parameters through real-time data assimilation technology. Use Kalman filter algorithm to fuse sensor data and simulation results. Assume that the state equation is , the observation equation is ,in is the state vector, is the state transfer matrix, is the process noise, is the observation vector, is the observation matrix, To observe noise, the Kalman filter algorithm is used to continuously correct the prediction results of the model based on the data collected by the sensor in real time, so that the virtual production environment model is more in line with the actual production situation.

[0055] Constructing resource allocation model: Constructing resource allocation model based on mixed integer programming method, with the goal of maximizing equipment utilization and minimizing energy consumption. Suppose the equipment set is , the task set is ,equipment The utilization rate is , the energy consumption is , then the objective function is .

[0056] The constraints include: Raw material supply cycle constraint: Assume that the raw material supply cycle is The amount of raw materials supplied each time is , the consumption rate of raw materials in the production process is , then it needs to satisfy ,in is the number of supplies, For the task The amount of raw material required.

[0057] Equipment capacity limit constraints: Equipment The capacity is ,Task On the device The capacity occupied during processing is ,but .

[0058] Production batch continuity constraint: If the production task needs to be carried out in batches, the minimum interval between adjacent batches is , it is necessary to ensure that the production arrangements of adjacent batches meet the time interval requirements.

[0059] Generate resource allocation plan: Decompose the production task into multiple subtasks, and generate a resource allocation matrix based on task priority and equipment load status. Task priority can be determined based on factors such as order urgency and product profit, and equipment load status can be measured by indicators such as the current working time and remaining capacity of the equipment. For example, for the task and equipment , the elements in the resource allocation matrix It can be defined as: If the task Assign to device On production, ;otherwise The resource allocation problem is solved by the Lagrangian relaxation algorithm, which introduces Lagrangian multipliers to integrate constraints into the objective function and transforms the original problem into a series of unconstrained sub-problems for solution. Finally, a global equipment scheduling plan is generated to determine the production task arrangement of each equipment at different times, as well as the raw material delivery plan, and to clarify the raw material delivery quantity and delivery time at each production stage. Embodiment 4:

[0060] This embodiment explains in detail how the scheduling layer makes real-time process adjustments based on the optimal production process parameters. By constructing a Bayesian network dynamic process adjustment model and a Markov decision process state transition model, combined with a sliding time window mechanism, it is possible to make timely and accurate adjustments to process parameters based on real-time sensor data, adapt to the uncertainty in the production process, and ensure the stability of the production process and product quality.

[0061] Constructing a Bayesian network dynamic process adjustment model Network structure setting: Use Bayesian network to build a dynamic process adjustment model with real-time sensor data, such as temperature sensor data , pH sensor data , enzyme activity sensor data As observation nodes, process parameter adjustment, such as temperature adjustment , pH adjustment amount , Pressure adjustment As decision nodes, the dependency between parameters is described by the conditional probability table. For example, for the temperature adjustment and temperature sensor data , conditional probability table It can represent the probability of needing different temperature adjustments under different temperature sensor measurement values.

[0062] Network training: The expectation maximization (EM) algorithm is used to fill in the missing process data. Assuming that some temperature sensor data is missing, the EM algorithm estimates the expected value of the missing data based on the current model parameters in the E step (expectation step), and re-estimates the model parameters using the estimated complete data in the M step (maximization step), and iterates until convergence. The Gibbs sampling method is used to estimate the posterior distribution of hidden variables to process hidden variables in the network. The network structure learning objective function is constructed based on the KL divergence. ,in is the true distribution, The greedy search strategy is used to optimize the node connection relationship. Starting from an initial network structure, by continuously adding, deleting or modifying edges, we try to find the network topology that minimizes the KL divergence and generates the network topology with the minimum description length. The Monte Carlo dropout method is used to evaluate network uncertainty. Some nodes are randomly discarded during the training process. Different network structures are obtained through multiple training. The network uncertainty is evaluated based on the differences in the prediction results of these network structures, and the confidence threshold of the process adjustment strategy is dynamically adjusted.

[0063] Constructing a state transition model of Markov decision process: Constructing a state transition model based on Markov decision process, defining the state space as the current process parameters and equipment state, and setting the state vector ,in is the temperature, is the pH value, For pressure, is the equipment status (such as whether the equipment is operating normally, the current load of the equipment, etc.). The action space is the adjustment amount of temperature, pH value and pressure, that is, The optimal adjustment strategy is solved by the value iteration algorithm. The core formula of the value iteration algorithm is:

[0064] in Status The optimal value function of For the status Execute an action Transfer to state The probability of From the state Execute an action Transfer to state Rewards received, is a discount factor, ranging from 0 to 1, reflecting the importance of future rewards. By continuously iterating and updating the value function, the optimal action in each state is finally obtained, that is, the optimal adjustment strategy.

[0065] Sliding time window mechanism: Introduce a sliding time window mechanism to perform weighted average processing on historical process data. Suppose the time window size is , the current time is , the historical process data is , the weight is ( , and the closer to the current moment, the greater the weight, such as ), then the smoothed process parameter value In this way, instantaneous noise interference is eliminated, smoothed process parameter correction instructions are generated, and frequent and unreasonable process adjustments caused by noise are avoided. For example, for temperature data, after weighted average processing in the sliding time window, a more stable temperature correction value is obtained, providing more reliable data support for subsequent process adjustments. Embodiment 5:

[0066] This embodiment describes in detail how the operation layer can achieve precise control of production equipment based on fuzzy logic control algorithms, and how to achieve closed-loop quality control through quality prediction models. By establishing a fuzzy rule base, performing fuzzification and defuzzification processing, and building an anti-saturation compensation mechanism, it is ensured that production equipment can accurately execute control instructions. At the same time, the results of the quality prediction model feedback are used to dynamically adjust the process to improve the stability and consistency of product quality.

[0067] ① Equipment control based on fuzzy logic control algorithm: Establish a fuzzy rule base: Establish a multi-input and multi-output fuzzy rule base for production equipment, and the input variables include temperature deviation ( To set the temperature, is the actual measured temperature), pH deviation ( To set the pH value, is the actual measured pH value) and pressure deviation ( To set the pressure, is the actual measured pressure), the output variables include heating power , Acid and alkali pump flow And pressure valve opening .

[0068] Fuzzy processing: Use the triangle membership function to perform fuzzy processing on the input and output variables. Taking temperature deviation as an example, assuming that the temperature deviation range is [-5, 5], define fuzzy sets such as "negative large", "negative small", "zero", "positive small", and "positive large", and the vertex coordinates of the triangle membership function are determined according to the actual situation. For example, the vertex coordinates of the triangle membership function of "negative large" can be (-5, -5, -3), and the vertex coordinates of "negative small" can be (-3, -1, 1), etc. The actual temperature deviation value is mapped to the corresponding fuzzy set through the membership function to obtain the fuzzified result.

[0069] Defuzzification and control quantity calculation: The accurate control quantity is obtained by defuzzification through the centroid method. For output variables, such as heating power, it is assumed that the multiple fuzzy output quantities obtained by fuzzy reasoning are , and its membership degree is , then the precise control amount In this way, fuzzy control decisions are converted into actual executable control signals to control the operation of production equipment.

[0070] Anti-saturation compensation mechanism: Build an anti-saturation compensation mechanism. When the controlled quantity exceeds the physical limit of the equipment, a proportional integral derivative (PID) auxiliary controller is used to compensate for the error. The physical limit is , when the calculated heating power When the PID controller is based on the error Calculate compensation amount ,in , , These are the parameters of the PID controller. By adjusting these parameters, the output of the actuator can be stabilized within the tolerable range of the equipment.

[0071] ②Quality prediction and feedback control: Constructing a quality prediction model: A quality prediction model is constructed based on a long short-term memory (LSTM) network, with historical process data, such as temperature, pH value, enzyme activity, raw material input rate, and other data in the past period of time, as well as real-time sensor data as input. The memory unit in the LSTM network can effectively capture the long-term dependencies in time series data. Assume that the input data sequence is After layer-by-layer processing by the LSTM network, the output product peptide chain length distribution Prediction value of bioactivity index .

[0072] Application of attention mechanism: The attention mechanism is used to enhance the feature weights of key process parameters. When the LSTM network processes data, the attention weight of each input feature is calculated. , ,in It is calculated by an attention scoring function, which can be based on the correlation between the input features and the network hidden state. For example: ,in is the hidden state of the previous moment. The input features are weighted by the attention weight to highlight the process parameter features that have a greater impact on product quality. Then the local time series pattern is extracted through the time series sliding window. Assuming that the sliding window size is, at each time step, the current moment is taken forward. The data of each time step is taken as a window data, and the changing trend and characteristics of the data in the window are analyzed. Combined with the global process status, such as the current equipment operation status, overall production progress and other information, a more accurate quality prediction result is generated.

[0073] Closed-loop quality control: Feedback the quality prediction results to the scheduling layer to dynamically correct the process adjustment strategy. If the quality prediction model predicts that the product peptide chain length distribution or biological activity index deviates from the expected target, the scheduling layer will re-evaluate the current process parameters based on the feedback information, and adjust the adjustment amount of process parameters such as temperature, pH value, and pressure in combination with the Bayesian network dynamic process adjustment model and the Markov decision process state transition model, so that the production process is carried out in a direction that is conducive to improving product quality. For example, if the product biological activity index is predicted to be low, and analysis shows that it is related to the current low reaction temperature, the scheduling layer will appropriately increase the temperature adjustment amount, thereby achieving closed-loop quality control, continuously optimizing the production process, and ensuring stable product quality.

[0074] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0075] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An abalone peptide production process optimization and intelligent decision-making method, characterized in that: The method comprises: Collecting real-time process data in the abalone peptide production process through multi-source sensors, wherein the multi-source sensors include a temperature sensor, a pH sensor, an enzyme activity sensor, and a pressure sensor; performing multi-scale feature extraction on the real-time process data based on a convolutional neural network to obtain process feature data; The process characteristic data is input into a pre-trained genetic algorithm optimization model, wherein the genetic algorithm optimization model adopts an adaptive crossover rate and mutation rate mechanism, iteratively optimizes the process parameter combination based on a multi-objective fitness function, and generates a process parameter optimization plan; Constructing a multi-objective production optimization model according to the process parameter optimization scheme, wherein the multi-objective production optimization model takes maximizing production efficiency and maximizing raw material utilization as optimization objectives, and adopts an improved particle swarm algorithm to globally optimize the production process, wherein the improved particle swarm algorithm introduces an inertia weight dynamic adjustment mechanism and a neighborhood search strategy; outputting optimal production process parameters based on the multi-objective production optimization model; A hierarchical intelligent decision-making model is established according to the optimal production process parameters, and the hierarchical intelligent decision-making model includes a planning layer, a scheduling layer and an operation layer, wherein the planning layer performs global resource allocation planning based on the process parameter optimization plan, the scheduling layer performs dynamic parameter scheduling based on the optimal production process parameters, and the operation layer realizes the execution control of the production equipment based on the fuzzy logic control algorithm; the production control instructions are output through the hierarchical intelligent decision-making model to realize the optimization and intelligent decision-making of the abalone peptide production process.

2. The method for optimizing the production process and intelligent decision-making of abalone peptide according to claim 1, characterized in that: The process characteristic data is input into a pre-trained genetic algorithm optimization model. The genetic algorithm optimization model adopts an adaptive crossover rate and mutation rate mechanism to iteratively optimize the process parameter combination based on a multi-objective fitness function to generate a process parameter optimization solution including: Acquire real-time process data, wherein the real-time process data includes reactor temperature, enzymatic solution pH value, enzyme activity concentration, raw material input rate and reaction pressure; construct a parameter space based on the real-time process data, and construct a decision space based on the process adjustable temperature change, pH adjustment amount, enzyme addition amount and pressure adjustment amount; A multi-objective fitness function is constructed based on the parameter space and the decision space, wherein the multi-objective fitness function includes a production efficiency item, a raw material utilization item, an energy consumption item, and a stability item, wherein the production efficiency item is calculated by the ratio of the product output per unit time to the target output, the raw material utilization item is calculated by the ratio of the actual product quality to the theoretical maximum product quality, the energy consumption item is calculated by the weighted square sum of the temperature adjustment amount, the pH adjustment amount, and the pressure adjustment amount, and the stability item is calculated by the Euclidean distance between the current process parameter and the historical optimal parameter; Construct an adaptive crossover rate and mutation rate mechanism, wherein the adaptive crossover rate is dynamically adjusted with the population diversity index, and the mutation rate changes exponentially with the increase in the number of iterations; select parent individuals based on the elite retention strategy, generate offspring individuals through crossover operations, and perform mutation operations on the offspring individuals; The non-dominated sorting method is used to stratify the population individuals, and the Pareto frontier solution set is screened based on the crowding distance. The optimal solution in the Pareto frontier solution set that satisfies the trade-off between production efficiency and raw material utilization is taken as the process parameter optimization solution.

3. The abalone peptide production process optimization and intelligent decision-making method according to claim 1, characterized in that: The method for outputting optimal production process parameters based on the multi-objective production optimization model includes: Constructing a multi-objective optimization function, wherein the multi-objective optimization function includes a production efficiency objective function and a raw material utilization objective function, wherein the production efficiency objective function is calculated by the ratio of reaction time to product output, and the raw material utilization objective function is calculated by the ratio of raw material consumption to product quality; Constructing constraint conditions based on the multi-objective optimization function, wherein the constraint conditions include temperature range constraint, pH range constraint, enzyme activity threshold constraint and pressure safety threshold constraint; The particle swarm algorithm is used to encode the production process parameters. Each particle represents a set of process parameter combinations. The parameter values ​​are adjusted based on the particle position update formula. The inertia weight dynamic adjustment mechanism adjusts the weight value according to the particle convergence degree. The neighborhood search strategy guides the population update through the local optimal particle. A dynamic learning factor is introduced, and the dynamic learning factor changes piecewise linearly with the increase of the number of iterations, and the contribution weights of the global optimal particle and the local optimal particle to the speed update are adjusted by the dynamic learning factor; Based on the improved particle swarm algorithm, iterative optimization is performed, the production efficiency and raw material utilization rate of the parameter combination generated in each iteration are evaluated, the non-dominated solution is stored in the optimization solution set, and finally the parameter combination with the highest comprehensive score in the Pareto front is selected as the optimal production process parameter.

4. The method for optimizing the production process and intelligent decision-making of abalone peptide according to claim 1, characterized in that: The planning layer performs global resource allocation planning based on the process parameter optimization solution, including: A virtual production environment model is constructed using digital twin technology. The virtual production environment model is driven by real-time process data and equipment status data to simulate the dynamic behavior of the reactor and the material flow process. A resource allocation model is constructed based on a mixed integer programming method, wherein the resource allocation model aims to maximize equipment utilization and minimize energy consumption, and the constraints include raw material supply cycle, equipment capacity limit, and production batch continuity; The production task is decomposed into multiple subtasks, and the resource allocation matrix is ​​generated based on the task priority and equipment load status. The resource allocation problem is solved through the Lagrangian relaxation algorithm to generate a global equipment scheduling plan and raw material delivery plan.

5. The abalone peptide production process optimization and intelligent decision-making method according to claim 1, characterized in that: The scheduling layer performs real-time process adjustments based on the optimal production process parameters, including: A dynamic process adjustment model is constructed using a Bayesian network, which uses real-time sensor data as observation nodes, process parameter adjustment amounts as decision nodes, and describes the dependency relationship between parameters through a conditional probability table; A state transition model is constructed based on the Markov decision process. The state space is defined as the current process parameters and equipment status, and the action space is defined as the adjustment amount of temperature, pH value and pressure. The optimal adjustment strategy is solved through the value iteration algorithm. A sliding time window mechanism is introduced to perform weighted average processing on historical process data, eliminate instantaneous noise interference, and generate smoothed process parameter correction instructions.

6. The abalone peptide production process optimization and intelligent decision-making method according to claim 1, characterized in that: The operation layer realizes precise control of production equipment based on fuzzy logic control algorithm, including: Establishing a multi-input multi-output fuzzy rule base for production equipment, wherein the input variables of the fuzzy rule base include temperature deviation, pH deviation and pressure deviation, and the output variables of the fuzzy rule base include heating power, acid-base pump flow and pressure valve opening; The input and output variables are fuzzified using triangular membership functions, and the fuzzy rules are defined as IF-THEN forms. The precise control quantity is obtained by defuzzification using the centroid method. An anti-saturation compensation mechanism is constructed. When the control quantity exceeds the physical limit of the equipment, a proportional-integral-differential auxiliary controller is used to compensate for the error to ensure the stable output of the actuator.

7. The abalone peptide production process optimization and intelligent decision-making method according to claim 1, characterized in that: Also includes: A quality prediction model is constructed based on a long short-term memory (LSTM) network. The quality prediction model takes historical process data and real-time sensor data as input and outputs the distribution of product peptide chain length and the predicted value of biological activity index. The attention mechanism is used to enhance the feature weights of key process parameters, local time series patterns are extracted through time series sliding windows, and quality prediction results are generated in combination with global process status; The quality prediction results are fed back to the scheduling layer to dynamically correct the process adjustment strategy to achieve closed-loop quality control.

8. The method for optimizing the production process and intelligent decision-making of abalone peptide according to claim 4, characterized in that: The virtual production environment model is constructed by the following steps: The finite element method was used to model the fluid dynamics in the reactor, and the temperature field and concentration field distribution were solved by the Navier-Stokes equation. The collision and dissolution process of the raw material particles was simulated based on the discrete element method, and the product generation rate was calculated in combination with the enzymatic reaction kinetic equation. The model parameters are updated through real-time data assimilation technology, and the Kalman filter algorithm is used to fuse sensor data and simulation results.

9. The method for optimizing the production process and intelligent decision-making of abalone peptide according to claim 5, characterized in that: The training of the Bayesian network includes: The expectation maximization (EM) algorithm is used to fill in the missing process data, and the posterior distribution of latent variables is estimated by the Gibbs sampling method; Based on KL divergence, the network structure learning objective function is constructed, and the node connection relationship is optimized using a greedy search strategy to generate a network topology with the minimum description length. The network uncertainty is evaluated through the Monte Carlo dropout method, and the confidence threshold of the process adjustment strategy is dynamically adjusted.

10. An abalone peptide production process optimization and intelligent decision-making system, used to implement the method according to any one of claims 1 to 9, characterized in that: include: Data acquisition and feature extraction module, which is used to collect real-time process data through multi-source sensors, extract process feature data based on convolutional neural network, and generate process parameter optimization scheme through genetic algorithm optimization model; A production process optimization module is used to construct a multi-objective production optimization model according to the process parameter optimization scheme, and output the optimal production process parameters using an improved particle swarm algorithm; The intelligent decision-making execution module is used to realize global resource allocation, real-time process adjustment and precise equipment control based on the hierarchical intelligent decision-making model, and output production control instructions.

Images