Intelligent production control method and device for amino acid formula powder

Abstract

The invention discloses an intelligent production control method and device for amino acid formula powder, and relates to the technical field of temperature control. Acquiring historical temperature data of the fermentation tank on a time sequence and determining a temperature fluctuation coefficient; environment parameters in the fermentation tank are obtained, a state evaluation coefficient is obtained through calculation according to the temperature fluctuation coefficient, and the fermentation stock solution is evaluated according to the state evaluation coefficient; if the state evaluation coefficient is greater than a state threshold value and the fermentation time reaches a first preset time, treating the fermentation stock solution and mixing according to a preset formula proportion to obtain amino acid formula powder; temperature fluctuation in the fermentation tank is accurately monitored by analyzing historical temperature data through a time sequence, the fermentation state is comprehensively evaluated in combination with environmental parameters, and stock solution treatment is ensured under the conditions of normal temperature and sufficient fermentation time, so that amino acid powder raw materials are efficiently and stably produced, the process is optimized, the production efficiency is improved, and the production cost is reduced. The fine control on the fermentation process is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of temperature control, and in particular relates to an intelligent production control method and device for amino acid formula powder. Background Art

[0002] As the basic unit of protein, amino acids play a vital role in the human body. They are not only involved in protein synthesis and metabolism, but also essential nutrients for physiological processes such as muscle growth, maintaining normal body functions, regulating blood sugar, improving immunity, promoting metabolism, helping detoxification, regulating emotions, and protecting heart health. The main method for large-scale industrial production of amino acids is through microbial fermentation.

[0003] The PID control algorithm is used to control the temperature of the amino acid fermentation tank, mainly using the PID controller to monitor and adjust the temperature in the fermentation tank in real time. According to the error between the set temperature value and the actual temperature value in the fermentation tank, the PID controller outputs the corresponding control signal through the calculation of the three links of proportion (P), integration (I) and differentiation (D), and adjusts the start and stop time and power output of the heating or cooling equipment, thereby realizing the control of the temperature in the fermentation tank. Ensure that the temperature in the fermentation tank fluctuates within the set range to provide a suitable temperature environment for the fermentation of amino acids.

[0004] During the fermentation production process of amino acids, the ambient temperature difference and the heat conduction of the tank will cause abnormal changes in the temperature inside the tank. The fermentation tank temperature control system has a lag, resulting in untimely response. Therefore, when using traditional algorithms for regulation, the fermentation tank temperature regulation lacks timeliness and stability. Summary of the invention

[0005] The purpose of the present invention is to solve the problem that the temperature inside the tank changes abnormally due to the ambient temperature difference and the heat conduction of the tank body, and the fermentation tank temperature control system has a lag that leads to untimely response. Therefore, in the process of using traditional algorithms for regulation, the fermentation tank temperature regulation lacks timeliness and stability. An intelligent production control method and device for amino acid formula powder are proposed.

[0006] In a first aspect of the present invention, a smart production control method for amino acid formula powder is first proposed, the method comprising: Acquire historical temperature data of the fermentation tank in time series, and determine the temperature fluctuation coefficient according to the historical temperature data; the historical temperature data includes historical ambient temperature, historical tank body temperature and historical tank internal temperature in any period; Acquire environmental parameters in the fermentation tank, calculate a state evaluation coefficient according to the environmental parameters and the temperature fluctuation coefficient, and evaluate the fermentation stock solution according to the state evaluation coefficient; the environmental parameters include: pH value, dissolved oxygen content and nutrient concentration; If the state assessment coefficient is greater than the state threshold and the fermentation time reaches the first preset time, the fermentation stock solution is processed to obtain an amino acid powder raw material, and the amino acid powder raw materials are mixed according to a preset formula ratio to obtain an amino acid formula powder.

[0007] Optionally, determining the temperature fluctuation coefficient according to the historical temperature data includes: Acquire historical product data and basic data, and input the historical temperature data, the basic data and the historical product data into a machine learning model for training; the historical temperature data and the historical product data correspond to each other at the same time in the time series; Using the output of the machine learning model as a model parameter, and iteratively updating the machine learning model according to the model parameter to obtain a fluctuation judgment model; Acquire the current tank temperature data and target product data, and input the tank temperature data and target product data into the fluctuation judgment model to obtain the predicted temperature data and predicted product data of the next cycle; The synthesis rate is calculated according to the time length of the next cycle and the predicted product data. If the synthesis rate is less than the synthesis threshold, the synthesis rate is determined to be abnormal; The time length of the next cycle is divided into time intervals, the predicted temperature difference of any time interval is calculated, the number of times the predicted temperature difference is greater than the temperature difference threshold is counted, and the temperature fluctuation coefficient is calculated based on the number.

[0008] Optionally, the fermentation stock solution is evaluated according to the state evaluation coefficient, and the temperature of the fermentation tank is controlled, and the temperature control is applied to a PID controller. The method includes: Step 1: determine the topological structure of the LSTM model, obtain the number of hyperparameters in the topological structure, and update the hyperparameters by the whale algorithm to obtain initial update parameters, and update the LSTM model according to the initial update parameters to obtain a target LSTM model; Step 2: Calculate the historical tank temperature data and the tank temperature data according to the preset temperature to obtain a historical temperature difference data set and a temperature difference data set; the temperature difference is the difference between the preset temperature and the tank temperature data; Step 3, inputting the tank temperature dataset, the preset temperature, the historical temperature difference dataset and the temperature difference dataset into the target LSTM model for training; Step 4: until the temperature difference data is less than the preset judgment value, the model result is output and the output model result is used as the PID control parameter, otherwise, return to step 2; Step five, controlling the temperature of the fermentation tank according to the PID control parameters.

[0009] Optionally, the hyperparameters are updated by a whale algorithm to obtain initial update parameters, and the method includes: The whale individuals are screened according to the fitness to obtain a judgment group, and the whale individuals are randomly selected and the fitness of the whale individuals is determined; the judgment group includes the fitness of the first best whale, the second best whale and the third best whale; If the fitness of the individual whale belongs to any one of the judgment groups, the second condition of rand is judged; If the fitness of the individual whale belongs to any one of the judgment groups, it is determined that rand meets the second condition, and rand < the preset threshold, then the Levy flight strategy is executed; If the fitness of the individual whale belongs to any one of the judgment groups, it is determined that rand meets the second condition, and rand ≥ the preset threshold, then the random walk strategy is executed; If the fitness of the whale individual does not belong to any of the judgment groups, it is determined that rand does not meet the second condition and the original whale optimization strategy is executed.

[0010] Optionally, after the temperature of the fermentation tank is controlled according to the PID control parameters, the following steps are included: The temperature system of the fermentation tank is controlled according to the PID control parameters, and when a second preset time is reached, the second environmental parameters and the second temperature fluctuation coefficient in the fermentation tank are obtained, and a second state evaluation coefficient is calculated according to the second environmental parameters and the second temperature fluctuation coefficient, and the fermentation stock solution is evaluated according to the second state evaluation coefficient.

[0011] In a second aspect of the present invention, a smart production control device for amino acid formula powder is provided, comprising: a temperature fluctuation coefficient acquisition module, a state evaluation coefficient evaluation module and an amino acid formula powder production module: The temperature fluctuation coefficient acquisition module is used to acquire the historical temperature data of the fermentation tank in time series, and determine the temperature fluctuation coefficient according to the historical temperature data; the historical temperature data includes the historical ambient temperature, the historical tank body temperature and the historical tank internal temperature of any period; The state evaluation coefficient evaluation module is used to obtain environmental parameters in the fermentation tank, calculate the state evaluation coefficient according to the environmental parameters and the temperature fluctuation coefficient, and evaluate the fermentation stock solution according to the state evaluation coefficient; the environmental parameters include: pH value, dissolved oxygen content and nutrient concentration; The amino acid formula powder production module is used to process the fermentation liquid to obtain amino acid powder raw materials if the state evaluation coefficient is greater than the state threshold and the fermentation time reaches a first preset time, and mix the amino acid powder raw materials according to a preset formula ratio to obtain amino acid formula powder.

[0012] Optionally, the temperature fluctuation coefficient acquisition module further includes: a model training module, a model updating module, a model prediction module, a success rate judgment module and a temperature fluctuation coefficient calculation module: The model training module is used to obtain historical product data and basic data, and input the historical temperature data, the basic data and the historical product data into the machine learning model for training; the historical temperature data and the historical product data have a one-to-one correspondence at the same moment in the time series; The model updating module is used to use the output of the machine learning model as a model parameter, and iteratively update the machine learning model according to the model parameter to obtain a fluctuation judgment model; The model prediction module is used to obtain the tank temperature data and target product data at the current moment, and input the tank temperature data and target product data into the fluctuation judgment model to obtain the predicted temperature data and predicted product data of the next cycle; The synthesis rate judgment module is used to calculate the synthesis rate according to the time length of the next cycle and the predicted product data, and if the synthesis rate is less than the synthesis threshold, it is determined that the synthesis rate is abnormal; The temperature fluctuation coefficient calculation module is used to divide the time length of the next cycle to obtain time intervals, calculate the predicted temperature difference of any time interval, count the number of times the predicted temperature difference is greater than the temperature difference threshold, and calculate the temperature fluctuation coefficient based on the number.

[0013] Optionally, the state evaluation coefficient evaluation module further includes: an initialization and updating parameter module, a training data acquisition module, a training data input module, a parameter judgment module and a PID parameter determination module: The initialization update parameter module is used to determine the topological structure of the LSTM model, obtain the number of hyperparameters in the topological structure, and update the hyperparameters through the whale algorithm to obtain initial update parameters, and update the LSTM model according to the initial update parameters to obtain a target LSTM model; The training data acquisition module is used to set a preset temperature, and calculate the historical tank temperature data and the tank temperature data according to the preset temperature to obtain a historical temperature difference data set and a temperature difference data set; the temperature difference is the difference between the preset temperature and the tank temperature data; The training data input module is used to input the tank temperature data set, the preset temperature, the historical temperature difference data set and the temperature difference data set into the target LSTM model for training; The parameter judgment module is used to output the model result and use the output model result as the PID control parameter until the temperature difference data is less than the preset judgment value, otherwise, return to step 2; The PID parameter determination module is used to control the temperature of the fermentation tank according to the PID control parameters.

[0014] Optionally, the initialization update parameter module includes: an update mode confirmation module, an update mode judgment module, a first update mode module, a second update mode module and a third update mode module: The update mode confirmation module is used to screen the whale individuals according to the fitness to obtain a judgment group, randomly select the whale individual and determine the fitness of the whale individual; the judgment group includes the fitness of the first best whale, the second best whale and the third best whale; The update mode judgment module is used to judge whether the second condition of rand is satisfied if the fitness of the individual whale belongs to any one of the judgment groups; The first update mode module is used to determine that rand satisfies the second condition if the fitness of the individual whale belongs to any one of the judgment groups, and rand < a preset threshold, and then execute the Levy flight strategy; The second update mode module is used to determine that rand satisfies the second condition if the fitness of the whale individual belongs to any one of the judgment groups, and rand ≥ a preset threshold, and then execute a random walk strategy; The third update mode module is used to determine that rand does not meet the second condition and execute the original whale optimization strategy if the fitness of the whale individual does not belong to any of the judgment groups.

[0015] Optionally, the PID parameter determination module is also used to: control the temperature system of the fermentation tank according to the PID control parameters, obtain the second environmental parameters and the second temperature fluctuation coefficient in the fermentation tank when a second preset time is reached, calculate the second state evaluation coefficient according to the second environmental parameters and the second temperature fluctuation coefficient, and evaluate the fermentation stock solution according to the second state evaluation coefficient.

[0016] Beneficial effects of the present invention: The present invention proposes an intelligent production control method for amino acid formula powder, which obtains historical temperature data of a fermentation tank in a time series and determines a temperature fluctuation coefficient; obtains environmental parameters in the fermentation tank and calculates a state evaluation coefficient according to the temperature fluctuation coefficient, and evaluates the fermentation stock solution according to the state evaluation coefficient; if the state evaluation coefficient is greater than a state threshold, and the fermentation time reaches a first preset time, the fermentation stock solution is processed and mixed according to a preset formula ratio to obtain the amino acid formula powder; the time series historical temperature data is analyzed to accurately monitor the temperature fluctuation in the fermentation tank, and the fermentation state is comprehensively evaluated in combination with environmental parameters to ensure that the stock solution is processed when the temperature is normal and the fermentation time is sufficient, thereby efficiently and stably producing amino acid powder raw materials, optimizing the process and improving production efficiency, and realizing refined control of the fermentation process. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present invention will be further described below in conjunction with the accompanying drawings.

[0018] Figure 1 A flowchart of a smart production control method for amino acid formula powder is provided for an embodiment of the present invention; Figure 2 A structural schematic diagram of an intelligent production control device for amino acid formula powder is provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0019] 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. The term "and / or" herein is only a description of the association relationship of the associated objects, indicating that there can be three relationships. For example, A and B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the present invention, the description of "first", "second", etc. is only used for descriptive purposes, and cannot be understood as indicating or implying its relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" can explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the ability of ordinary technicians in the field to implement. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0020] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0021] The embodiment of the present invention provides a smart production control method for amino acid formula powder. Figure 1 , Figure 1 A flowchart of a smart production control method for amino acid formula powder provided by an embodiment of the present invention. The method comprises the following steps: S101, obtaining historical temperature data of the fermentation tank in time series, and determining a temperature fluctuation coefficient according to the historical temperature data; S102, obtaining environmental parameters in the fermentation tank, calculating a state evaluation coefficient according to the environmental parameters and the temperature fluctuation coefficient, and evaluating the fermentation stock solution according to the state evaluation coefficient; S103, if the state evaluation coefficient is greater than the state threshold, and the fermentation time reaches the first preset time, the fermentation stock solution is processed to obtain an amino acid powder raw material, and the amino acid powder raw materials are mixed according to a preset formula ratio to obtain an amino acid formula powder; Historical temperature data include historical ambient temperature, historical tank body temperature and historical tank internal temperature in any period; environmental parameters include: pH value, dissolved oxygen content and nutrient concentration; An intelligent production control method for amino acid formula powder provided by an embodiment of the present invention accurately monitors temperature fluctuations in a fermentation tank by analyzing historical temperature data through time series, and comprehensively evaluates the fermentation status in combination with environmental parameters to ensure that the stock solution is processed when the temperature is normal and the fermentation time is sufficient, thereby efficiently and stably producing amino acid powder raw materials, and accurately preparing amino acid formula powder that meets the preset formula ratio. The entire process improves production efficiency and realizes refined control of the fermentation process.

[0022] In one implementation, whether the temperature in the tank is abnormal at the current moment is determined based on the historical temperature data in the time series. If the temperature is abnormal, it is necessary to further obtain environmental parameters to evaluate the state in the tank at the current moment (if the state evaluation coefficient is greater than the state threshold, the temperature is normal, that is, normal fermentation; if it is less than the state threshold, the temperature is abnormal and temperature control is required). Through subsequent temperature regulation, the amino acid production is adjusted. When the fermentation time reaches the preset time (for example, 24 hours), the fermentation stock solution can be processed (bacteria separation, preliminary extraction, purification, and final drying and finished product), and various amino acid powder raw materials are mixed according to the preset formula ratio to obtain amino acid formula powder.

[0023] In one implementation, by obtaining the historical temperature data of the fermentation tank in time series and determining the temperature fluctuation coefficient based on this data, the temperature change during the fermentation process can be accurately quantified. This helps to detect abnormal temperature fluctuations in a timely manner, thereby preventing the fermentation process from getting out of control due to unstable temperature, which in turn causes a decrease in fermentation efficiency.

[0024] In one implementation, by obtaining the environmental parameters in the fermentation tank and combining them with the temperature fluctuation coefficient to calculate the state assessment coefficient, the state of the fermentation stock solution can be comprehensively and objectively assessed. Environmental parameters such as pH value, dissolved oxygen content, and nutrient concentration are all key factors affecting the fermentation process. Combining these parameters with the temperature fluctuation coefficient can more accurately reflect the overall state of the fermentation process. Sa=(α+0.1*T norm )*(PH norm +DO norm )+(β-0.1*T norm )*(Nc norm +T norm ), where Sa is the state assessment coefficient, T norm is the temperature fluctuation coefficient, PH norm ,DO norm and Nc norm is the environmental parameter, PH norm is pH value, DO norm is the dissolved oxygen content (unit: %), Nc norm is the nutrient concentration (unit: g / L), α and β are constant proportional coefficients, and the above environmental parameters and temperature fluctuation coefficients are normalized.

[0025] For example, here α is 0.3 and β is 0.2: parameter Original value Normalized values pH 5.5 0.75 Dissolved oxygen content (%) 22 0.033 Nutrient concentration (g / L) 4.8 0.955 Temperature fluctuation coefficient 0.583 0.583 Among them, the temperature fluctuation coefficient (the total number of times is 12 times, the number of abnormal times is 7 times, calculated) Sa=(0.3+0.1*0.583)*(0.75+0.033)+(0.2-0.1*0.583)*(0.955+0.583)≈0.499.

[0026] In one implementation, various amino acid powder raw materials are mixed according to a preset formula ratio to obtain amino acid formula powder, and various amino acid raw materials that meet quality standards are selected, such as lysine, methionine, threonine, tryptophan, etc. These raw materials are usually in powder form, which is convenient for mixing and processing; according to the formula requirements, other ingredients such as glucose syrup, mixed vegetable oil, vitamins, minerals, etc. may also need to be added; according to the target audience and nutritional needs of the product, the specific content ratio of various amino acids and other ingredients is determined, and then accurate weighing tools are used to accurately weigh various raw materials according to the formula ratio; the weighed raw materials are placed in a mixing device and fully mixed to ensure that the various ingredients are evenly distributed. During the mixing process, it may be necessary to control parameters such as mixing time and mixing speed to achieve the best mixing effect; the evenly mixed amino acid formula powder is packaged and stored in a dry, cool and ventilated place to avoid moisture and deterioration; amino acid formula powder ingredients, such as: glucose syrup powder, vegetable oil (sunflower oil, medium chain triglycerides, rapeseed oil), calcium hydrogen phosphate, L-leucine, potassium citrate, L-lysine acetate, L-glutamine, maltodextrin, L-proline, L-arginine, citric acid fatty acid glyceride, L-valine, glycine, L-isoleucine, L-aspartic acid, L-threonine, L-phenylalanine, L-tyrosine, L-serine, L-histidine, L-alanine, sodium chloride, L- Cystine, magnesium chloride, L-tryptophan, choline bitartrate, L-methionine, calcium citrate, magnesium L-aspartate, arachidonic acid oil, docosahexaenoic acid oil inositol, L-ascorbic acid, taurine, ferrous sulfate, zinc sulfate, L-carnitine, disodium 5'-uridine monophosphate, cytidine 5' monophosphate, niacinamide, disodium 5'-inosinic acid, adenosine 5' monophosphate, calcium D-pantothenate, disodium 5'-guanylate, dl-α-tocopheryl acetate, ascorbyl palmitate, cupric sulfate, thiamine hydrochloride, pyridoxine hydrochloride, manganese sulfate, riboflavin, vitamin A acetate, dl-α-tocopherol, potassium iodide, mixed tocopherol concentrate, folic acid, chromium chloride, phytonadione, sodium selenite, sodium molybdate, D-biotin, vitamins , cyanocobalamin. The added amount of medium chain triglycerides is 80.0g / 1000g. The lactose content is less than 0.5g / 100g; it is only for illustration.

[0027] In one embodiment, step S101 includes: Obtain historical product data and basic data, and input the historical temperature data, basic data and historical product data into the machine learning model for training; the historical temperature data and the historical product data correspond one to one at the same time in the time series; The output of the machine learning model is used as a model parameter, and the machine learning model is iteratively updated according to the model parameters to obtain a fluctuation judgment model; Obtaining the current tank temperature data and target product data, and inputting the tank temperature data and target product data into the fluctuation judgment model to obtain the predicted temperature data and predicted product data of the next cycle; The synthesis rate is calculated according to the time length of the next cycle and the predicted product data. If the synthesis rate is less than the synthesis threshold, the synthesis rate is determined to be abnormal; The time length of the next cycle is divided into time intervals, the predicted temperature difference of any time interval is calculated, the number of times the predicted temperature difference is greater than the temperature difference threshold is counted, and the temperature fluctuation coefficient is calculated based on the number.

[0028] In one implementation, the machine learning model is as follows: 1. Long Short-Term Memory Network (LSTM): LSTM is a special recurrent neural network (RNN) that can effectively handle long-term dependencies in time series data. It is particularly suitable for processing time series data such as historical temperature data and historical product data, and can learn the time correlation in the data; 2. Convolutional Neural Network Combined with Long Short-Term Memory Network (CNN-LSTM) CNN-LSTM combines the feature extraction capability of convolutional neural network (CNN) and the time series modeling capability of LSTM. CNN can be used to extract the features of historical temperature data and historical product data, while LSTM can process the time series characteristics of these features; 3. Autoregressive Moving Average Model (ARIMA) ARIMA is a classic statistical time series model suitable for the prediction of univariate time series data. It models time series data through methods such as difference, autoregression and moving average, and is suitable for the prediction of historical temperature data; 4. XGBoost: XGBoost is an ensemble learning algorithm based on gradient boosting, suitable for regression and classification tasks. It can handle large-scale data sets and performs well in feature importance evaluation and model interpretability. XGBoost can be used to predict temperature and product data, and the model performance can be optimized by cross-validation parameter adjustment; 5. Models based on attention mechanism: The attention mechanism can help the model better focus on important features in the data. Models combined with time series decomposition (such as adaptive time series decomposition) (such as ATFSAS) can effectively handle trend and periodic components in time series data; 6. Multivariate LSTM: Multivariate LSTM can process multiple related time series variables (such as temperature and product data) at the same time and capture the relationship between them. This model is particularly suitable for predicting multivariate time series data.

[0029] In one implementation, historical temperature data, basic data (capacity of the fermentation tank, stirring speed, ventilation volume, culture medium composition, etc., which are fixed and do not need to be re-entered later) and historical product data are integrated and input into the machine learning model for training, which can make full use of existing data resources, improve the accuracy and efficiency of model training, and lay a solid foundation for subsequent temperature and product predictions. Obtaining the current tank temperature data and target product data, and using the fluctuation judgment model to predict the next cycle, can achieve real-time monitoring and abnormal detection of the production process, and timely discover and deal with potential problems; because the fermentation temperature is directly related to the amino acid production rate, in this machine learning model, the correlation between the product content and the temperature can be determined.

[0030] In one implementation, the synthesis rate is calculated based on the time length of the next cycle and the predicted product data, that is, predicted product data ÷ time length = synthesis rate. The synthesis rate helps to determine whether the amino acid production in the current fermentation tank is normal. If it is abnormal, it is necessary to judge the temperature in the tank; and the temperature fluctuation coefficient is: the number of times the predicted temperature difference is greater than the temperature difference threshold is taken as the number of abnormal temperatures, and the time length of the next cycle is divided to obtain the time interval as the total number of times. The temperature fluctuation coefficient = abnormal temperature number ÷ total number, which reflects the abnormal temperature fluctuation under the time length of the next cycle. When abnormal temperature fluctuation occurs, it is necessary to evaluate the state of the tank.

[0031] In one embodiment, step S102 includes: Step 1: determine the topological structure of the LSTM model, obtain the number of hyperparameters in the topological structure, and update the hyperparameters through the whale algorithm to obtain initial update parameters. According to the initial update parameters, the LSTM model is updated to obtain the target LSTM model. Step 2: Calculate the historical tank temperature data and the tank temperature data according to the preset temperature to obtain a historical temperature difference data set and a temperature difference data set; the temperature difference is the difference between the preset temperature and the tank temperature data; Step 3: Input the in-tank temperature dataset, preset temperature, historical temperature difference dataset, and temperature difference dataset into the target LSTM model for training; Step 4: until the temperature difference data is less than the preset judgment value, the model result is output and the output model result is used as the PID control parameter, otherwise, return to step 2; Step 5: Control the temperature of the fermentation tank according to the PID control parameters.

[0032] In one implementation, steps 1-5 are extensions of the original step 1. The complete update process is as follows: Step 1, determine the topology of the LSTM model, obtain the number of hyperparameters in the topology, initialize the dimension in the target whale algorithm to the number of hyperparameters, and randomly initialize the position of the whale population; Step 2, determine the fitness function and calculate the fitness, and judge whether the current whale population is the optimal population according to the fitness. If the fitness is not greater than the fitness threshold, execute step 5; if the fitness is greater than the fitness threshold, execute step 3; Step 3, calculate the fitness of the whale population and perform an update operation on the whale population; Step 4, update the position of each whale individual in the whale population and the fitness of the whale population, and determine the optimal population; Step 5, using the optimal population as the initial update parameter of the LSTM model to update the target LSTM model; Step 6, calculating the historical tank temperature data and the tank temperature data according to the preset temperature to obtain a historical temperature difference data set and a temperature difference data set; the temperature difference is the difference between the preset temperature and the tank temperature data; Step 7, input the tank temperature dataset, preset temperature, historical temperature difference dataset and temperature difference dataset into the target LSTM model for training; Step 8, until the temperature difference data is less than the preset judgment value, the model result is output and the output model result is used as the PID control parameter, otherwise, return to step 6; Step 9, controlling the temperature of the fermentation tank according to the PID control parameters.

[0033] In one implementation, LSTM is a recursive neural network that can handle nonlinear relationships in time series data and helps extract the complex dynamic characteristics of temperature control during the fermentation process. LSTM can capture long-term dependencies in time series through its unique gating mechanism (input gate, forget gate, and output gate); update the parameters of the LSTM model in combination with the whale optimization algorithm (for example: 1. The number of neurons in the hidden layer: The number of neurons in the hidden layer of the LSTM network, which directly affects the complexity and fitting ability of the model. Optimizing the number of neurons in the hidden layer can find a balance point that can avoid overfitting of the model and ensure that the model has sufficient ability to capture complex patterns in the data; 2. Learning rate: The learning rate is a very important hyperparameter in the LSTM training process. 3. Number of training iterations: This is the number of iterations in the model training process, which determines the adequacy of the model training. Optimizing the number of training iterations can avoid insufficient or excessive model training, thereby improving the model's predictive performance. That is, the LSTM model can dynamically adjust its parameters according to real-time data to adapt to various changes that may occur during the fermentation process, thereby improving adaptability. By training the LSTM model to predict the temperature difference and adjusting the PID control parameters accordingly, the accuracy and stability of temperature control can be improved.

[0034] In one implementation, the PID control parameter refers to the proportional gain , integral gain , differential gain ; In the above step 2, if the fitness is not greater than the fitness threshold, execute step 5 (the fitness here can be the fitness corresponding to the maximum number of iterations or the fitness not greater than the fitness threshold): If the fitness function value of the current whale individual is ≤ the fitness threshold, it means that the individual has found a good enough solution, and the condition can be considered to be met, or if the algorithm has run the maximum number of iterations, even if the fitness threshold is not reached, further iterations will be stopped to avoid an infinite loop; if the fitness is greater than the fitness threshold, execute step 3: Fitness does not meet the standard: The fitness function value of the current whale individual is still higher than the preset threshold, indicating that there is still room for improvement and iterative optimization needs to continue. The number of iterations does not meet the standard: If the algorithm has not reached the preset maximum number of iterations, even if the current solution is not good enough, it will continue to iterate to find a better solution.

[0035] In one implementation, step eight also has a stop condition, and when the maximum number of iterations is met, the model results will be directly output to prevent an infinite loop; in step seven, the tank temperature data set, the preset temperature, the historical temperature difference data set, and the temperature difference data set are input into the target LSTM model for training. These input data are used to calculate the output of the LSTM network, which is then set as the three adjustable parameters of the PID controller. The purpose of this step is to use the LSTM network's ability to process time series data to predict and adjust the controller parameters, thereby optimizing the temperature control process.

[0036] In one implementation, by applying the whale optimization algorithm to the initial parameter optimization of the LSTM network and combining it with the PID controller, the method can adaptively adjust the controller parameters in the n-dimensional search space to achieve precise control of complex systems (such as temperature control); by combining the whale optimization algorithm and the PID controller parameter optimization method of the LSTM network, by adaptively adjusting the input and output of the LSTM network, precise and intelligent control of the temperature control system is achieved, and the response speed and control accuracy of the system are improved.

[0037] In one embodiment, the hyperparameters are updated by the whale algorithm to obtain initial update parameters, including: The whale individuals are screened according to their fitness to obtain a judgment group, and a whale individual is randomly selected and the fitness of the whale individual is determined; the judgment group includes the fitness of the first best whale, the second best whale, and the third best whale; If the fitness of the individual whale belongs to any of the judgment groups, the second condition of rand is judged; If the fitness of the individual whale belongs to any one of the judgment groups, it is determined that rand meets the second condition, and rand < the preset threshold, then the Levy flight strategy is executed; If the fitness of the whale individual belongs to any one of the judgment groups, it is determined that rand meets the second condition, and rand ≥ the preset threshold, then the random walk strategy is executed; If the fitness of the whale individual does not belong to any of the judgment groups, it is determined that rand does not meet the second condition and the original whale optimization strategy is executed.

[0038] In one implementation, the Levy flight strategy: generates a random step size through Levy distribution, selects an arbitrary direction as the moving direction, and updates the whale individual according to the random step size and moving direction; the random walk strategy: randomly selects individuals in the whale population to generate movement probabilities and performs population traversal, determines the movement mode of the individual according to the movement probability, executes a jump to obtain a probability distribution, and uses the probability distribution as the input for the next jump for iterative update; In one implementation, the judgment group includes the fitness of the first best whale, the second best whale and the third best whale. The judgment group includes three fitnesses. The whale individuals are arranged in descending order according to the fitness, and the first three fitnesses are selected as components of the judgment group; rand is the function call result of a random number generator, which is used to generate a random floating point number between 0 and 1; rand is compared with a preset threshold, which is , where Y is the preset threshold, t is the number of iterations, is the exploration iteration ratio, and max is the maximum number of iterations.

[0039] In one implementation, the Levy flight strategy simulates the long-distance random search behavior of animals when looking for food. The Levy flight strategy can effectively explore the solution space while maintaining search efficiency, thereby increasing the probability of finding the global optimal solution. In the whale algorithm, the Levy distribution is first used to generate a heavy-tailed distribution step size (the step size determines the distance that the individual whale moves in the search space), and a direction is randomly selected (it can be the current position to the position of the best solution, or other random directions), and the position of the individual whale is updated according to the calculated step size and direction (if the new position exceeds the boundary of the search space, it will be adjusted to the boundary).

[0040] In one implementation, the random walk strategy simulates the random search behavior of animals in a local area. While maintaining the randomness of the search process, the random walk strategy also introduces a certain directionality, which helps the algorithm to explore the solution space more effectively. In the whale algorithm, first, any individual in the whale group is started to traverse the entire group. The random individual randomly generates a movement probability G (G represents the probability of jumping to any individual in the group). There are two ways for the individual to move: when the movement probability is G, it randomly jumps to any individual in the group, and when the movement probability is 1-G, it moves to an adjacent individual. After each jump, a probability distribution reflecting the probability of each individual in the found group can be obtained. This probability distribution is then used as the input for the next walk, and the process is repeated. Under predetermined conditions, the probability distribution tends to converge, and a stable probability distribution is obtained after convergence. The random walk generates new solutions using crossover and hybrid mutations, which can enhance the diversity of the group and effectively avoid falling into the local optimal solution, thereby improving the local search ability of the target algorithm. Applying the random walk strategy in the swarm intelligence algorithm can speed up the search for the best solution.

[0041] In one implementation, the original whale optimization strategy is: surround prey, bubble net attack, and search and prey. The execution of the Levy flight strategy and the random walk strategy in WOA provides global search and local search capabilities respectively. Levy flight helps the algorithm escape from the local optimum through long-distance jumps, while random walks help the algorithm conduct detailed searches in local areas through short-distance random moves. The combination of these two strategies enables WOA to effectively explore the solution space while maintaining search efficiency, thereby increasing the probability of finding the global optimal solution.

[0042] In one embodiment, after the fermentation tank is temperature controlled according to the PID control parameters, the method includes: The temperature system of the fermentation tank is controlled according to the PID control parameters. When the second preset time is reached, the second environmental parameters and the second temperature fluctuation coefficient in the fermentation tank are obtained. The second state evaluation coefficient is calculated according to the second environmental parameters and the second temperature fluctuation coefficient. The fermentation stock solution is evaluated according to the second state evaluation coefficient.

[0043] In one implementation, by conducting a secondary evaluation of the fermentation tank, it is possible to confirm whether the temperature conditions in the fermentation tank are in the optimal state and determine whether the temperature adjustment is effective. If it is not effective, temperature control needs to be performed again; the appropriate temperature helps accelerate the growth and metabolism of microorganisms and improve fermentation efficiency. Optimized temperature conditions help shorten the fermentation cycle, reduce production time, and improve overall production efficiency.

[0044] In one implementation, the re-evaluation of the fermenter can collect a large amount of data such as environmental parameters and temperature fluctuation coefficients, which provide data support for subsequent process optimization and continuous improvement.

[0045] Based on the same inventive concept, the present invention also provides an intelligent production control device for amino acid formula powder. Figure 2 , Figure 2 A schematic diagram of the structure of an intelligent production control device for amino acid formula powder provided by an embodiment of the present invention, comprising: a temperature fluctuation coefficient acquisition module, a state evaluation coefficient evaluation module and an amino acid formula powder production module: The temperature fluctuation coefficient acquisition module is used to obtain the historical temperature data of the fermentation tank in time series and determine the temperature fluctuation coefficient according to the historical temperature data; the historical temperature data includes the historical ambient temperature, the historical tank body temperature and the historical tank internal temperature in any period; The state evaluation coefficient evaluation module is used to obtain the environmental parameters in the fermentation tank, calculate the state evaluation coefficient according to the environmental parameters and the temperature fluctuation coefficient, and evaluate the fermentation stock solution according to the state evaluation coefficient; the environmental parameters include: pH value, dissolved oxygen content and nutrient concentration; The amino acid formula powder production module is used to process the fermentation liquid to obtain amino acid powder raw materials if the state evaluation coefficient is greater than the state threshold and the fermentation time reaches the first preset time, and mix the amino acid powder raw materials according to the preset formula ratio to obtain amino acid formula powder.

[0046] An intelligent production control device for amino acid formula powder provided by an embodiment of the present invention accurately monitors temperature fluctuations in a fermentation tank by analyzing historical temperature data through time series, and comprehensively evaluates the fermentation status in combination with environmental parameters to ensure that the stock solution is processed when the temperature is normal and the fermentation time is sufficient, thereby efficiently and stably producing amino acid powder raw materials and accurately preparing amino acid formula powder that meets the preset formula ratio. The entire process improves production efficiency and realizes refined control of the fermentation process.

[0047] In one embodiment, the temperature fluctuation coefficient acquisition module further includes: a model training module, a model updating module, a model prediction module, a success rate determination module and a temperature fluctuation coefficient calculation module: The model training module is used to obtain historical product data and basic data, and input the historical temperature data, basic data and historical product data into the machine learning model for training; the historical temperature data and the historical product data correspond to each other at the same time in the time series; A model updating module is used to use the output of the machine learning model as a model parameter, and iteratively update the machine learning model according to the model parameters to obtain a fluctuation judgment model; The model prediction module is used to obtain the current tank temperature data and target product data, and input the tank temperature data and target product data into the fluctuation judgment model to obtain the predicted temperature data and predicted product data of the next cycle; A synthesis rate judgment module is used to calculate the synthesis rate according to the time length of the next cycle and the predicted product data. If the synthesis rate is less than the synthesis threshold, the synthesis rate is judged to be abnormal; The temperature fluctuation coefficient calculation module is used to divide the time length of the next cycle to obtain time intervals, calculate the predicted temperature difference of any time interval, count the number of times the predicted temperature difference is greater than the temperature difference threshold, and calculate the temperature fluctuation coefficient based on the number.

[0048] In one embodiment, the state evaluation coefficient evaluation module further includes: an initialization update parameter module, a training data acquisition module, a training data input module, a parameter judgment module and a PID parameter determination module: Initialization update parameter module, used to determine the topological structure of the LSTM model, obtain the number of hyperparameters in the topological structure, and update the hyperparameters through the whale algorithm to obtain the initial update parameters. The LSTM model is updated according to the initial update parameters to obtain the target LSTM model. The training data acquisition module is used to set a preset temperature, calculate the historical tank temperature data and the tank temperature data according to the preset temperature to obtain a historical temperature difference data set and a temperature difference data set; the temperature difference is the difference between the preset temperature and the tank temperature data; A training data input module is used to input the in-tank temperature dataset, the preset temperature, the historical temperature difference dataset, and the temperature difference dataset into the target LSTM model for training; The parameter judgment module is used to output the model result and use the output model result as the PID control parameter until the temperature difference data is less than the preset judgment value, otherwise, return to step 2; The PID parameter determination module is used to control the temperature of the fermentation tank according to the PID control parameters.

[0049] In one embodiment, the initialization update parameter module includes: an update mode confirmation module, an update mode judgment module, a first update mode module, a second update mode module and a third update mode module: The update mode confirmation module is used to screen the whale individuals according to the fitness to obtain a judgment group, randomly select a whale individual and determine the fitness of the whale individual; the judgment group includes the fitness of the first best whale, the second best whale and the third best whale; Update mode judgment module, used to judge whether the second condition of rand is satisfied if the fitness of the individual whale belongs to any one of the judgment groups; The first update mode module is used to determine that rand satisfies the second condition if the fitness of the individual whale belongs to any one of the judgment groups, and rand < a preset threshold, and then execute the Levy flight strategy; The second update mode module is used to determine that rand satisfies the second condition if the fitness of the whale individual belongs to any one of the judgment groups, and rand ≥ a preset threshold, and then execute a random walk strategy; The third update mode module is used to determine that rand does not meet the second condition and execute the original whale optimization strategy if the fitness of the whale individual does not belong to any of the judgment groups.

[0050] In one embodiment, the PID parameter determination module is also used to: control the temperature system of the fermentation tank according to the PID control parameters, obtain the second environmental parameters and the second temperature fluctuation coefficient in the fermentation tank when the second preset time is reached, calculate the second state evaluation coefficient according to the second environmental parameters and the second temperature fluctuation coefficient, and evaluate the fermentation stock solution according to the second state evaluation coefficient.

[0051] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A smart production control method for amino acid formula powder, characterized in that: The method comprises: Acquire historical temperature data of the fermentation tank in time series, and determine the temperature fluctuation coefficient according to the historical temperature data; the historical temperature data includes historical ambient temperature, historical tank body temperature and historical tank internal temperature in any period; Acquire environmental parameters in the fermentation tank, calculate a state evaluation coefficient according to the environmental parameters and the temperature fluctuation coefficient, and evaluate the fermentation stock solution according to the state evaluation coefficient; the environmental parameters include: pH value, dissolved oxygen content and nutrient concentration; If the state assessment coefficient is greater than the state threshold and the fermentation time reaches the first preset time, the fermentation stock solution is processed to obtain an amino acid powder raw material, and the amino acid powder raw materials are mixed according to a preset formula ratio to obtain an amino acid formula powder.

2. The intelligent production control method for amino acid formula powder according to claim 1, characterized in that: Determining the temperature fluctuation coefficient according to the historical temperature data includes: Acquire historical product data and basic data, and input the historical temperature data, the basic data and the historical product data into a machine learning model for training; the historical temperature data and the historical product data correspond to each other at the same time in the time series; Using the output of the machine learning model as a model parameter, and iteratively updating the machine learning model according to the model parameter to obtain a fluctuation judgment model; Acquire the current tank temperature data and target product data, and input the tank temperature data and target product data into the fluctuation judgment model to obtain the predicted temperature data and predicted product data of the next cycle; The synthesis rate is calculated according to the time length of the next cycle and the predicted product data. If the synthesis rate is less than the synthesis threshold, the synthesis rate is determined to be abnormal; The time length of the next cycle is divided into time intervals, the predicted temperature difference of any time interval is calculated, the number of times the predicted temperature difference is greater than the temperature difference threshold is counted, and the temperature fluctuation coefficient is calculated based on the number.

3. The intelligent production control method for amino acid formula powder according to claim 1, characterized in that: The fermentation stock solution is evaluated according to the state evaluation coefficient, and the temperature of the fermentation tank is controlled, and the temperature control is applied to a PID controller. The method includes: Step 1: determine the topological structure of the LSTM model, obtain the number of hyperparameters in the topological structure, and update the hyperparameters by the whale algorithm to obtain initial update parameters, and update the LSTM model according to the initial update parameters to obtain a target LSTM model; Step 2: Calculate the historical tank temperature data and the tank temperature data according to the preset temperature to obtain a historical temperature difference data set and a temperature difference data set; the temperature difference is the difference between the preset temperature and the tank temperature data; Step 3, inputting the tank temperature dataset, the preset temperature, the historical temperature difference dataset and the temperature difference dataset into the target LSTM model for training; Step 4: until the temperature difference data is less than the preset judgment value, the model result is output and the output model result is used as the PID control parameter, otherwise, return to step 2; Step five, controlling the temperature of the fermentation tank according to the PID control parameters.

4. The intelligent production control method for amino acid formula powder according to claim 3, characterized in that: The hyperparameters are updated by using the whale algorithm to obtain initial update parameters, and the method includes: The whale individuals are screened according to the fitness to obtain a judgment group, and the whale individuals are randomly selected and the fitness of the whale individuals is determined; the judgment group includes the fitness of the first best whale, the second best whale and the third best whale; If the fitness of the individual whale belongs to any one of the judgment groups, the second condition of rand is judged; If the fitness of the individual whale belongs to any one of the judgment groups, it is determined that rand meets the second condition, and rand < the preset threshold, then the Levy flight strategy is executed; If the fitness of the individual whale belongs to any one of the judgment groups, it is determined that rand meets the second condition, and rand ≥ the preset threshold, then the random walk strategy is executed; If the fitness of the whale individual does not belong to any of the judgment groups, it is determined that rand does not meet the second condition and the original whale optimization strategy is executed.

5. The intelligent production control method for amino acid formula powder according to claim 3, characterized in that: After the temperature of the fermentation tank is controlled according to the PID control parameters, the method further comprises: The temperature system of the fermentation tank is controlled according to the PID control parameters, and when a second preset time is reached, the second environmental parameters and the second temperature fluctuation coefficient in the fermentation tank are obtained, and a second state evaluation coefficient is calculated according to the second environmental parameters and the second temperature fluctuation coefficient, and the fermentation stock solution is evaluated according to the second state evaluation coefficient.

6. An intelligent production control device for amino acid formula powder, characterized in that: The device comprises: a temperature fluctuation coefficient acquisition module, a state evaluation coefficient evaluation module and an amino acid formula powder production module: The temperature fluctuation coefficient acquisition module is used to acquire the historical temperature data of the fermentation tank in time series, and determine the temperature fluctuation coefficient according to the historical temperature data; the historical temperature data includes the historical ambient temperature, the historical tank body temperature and the historical tank internal temperature of any period; The state evaluation coefficient evaluation module is used to obtain environmental parameters in the fermentation tank, calculate the state evaluation coefficient according to the environmental parameters and the temperature fluctuation coefficient, and evaluate the fermentation stock solution according to the state evaluation coefficient; the environmental parameters include: pH value, dissolved oxygen content and nutrient concentration; The amino acid formula powder production module is used to process the fermentation liquid to obtain amino acid powder raw materials if the state evaluation coefficient is greater than the state threshold and the fermentation time reaches a first preset time, and mix the amino acid powder raw materials according to a preset formula ratio to obtain amino acid formula powder.

7. The intelligent production control device for amino acid formula powder according to claim 6, characterized in that: The temperature fluctuation coefficient acquisition module also includes: a model training module, a model updating module, a model prediction module, a rate determination module and a temperature fluctuation coefficient calculation module: The model training module is used to obtain historical product data and basic data, and input the historical temperature data, the basic data and the historical product data into the machine learning model for training; the historical temperature data and the historical product data have a one-to-one correspondence at the same moment in the time series; The model updating module is used to use the output of the machine learning model as a model parameter, and iteratively update the machine learning model according to the model parameter to obtain a fluctuation judgment model; The model prediction module is used to obtain the tank temperature data and target product data at the current moment, and input the tank temperature data and target product data into the fluctuation judgment model to obtain the predicted temperature data and predicted product data of the next cycle; The synthesis rate judgment module is used to calculate the synthesis rate according to the time length of the next cycle and the predicted product data, and if the synthesis rate is less than the synthesis threshold, it is determined that the synthesis rate is abnormal; The temperature fluctuation coefficient calculation module is used to divide the time length of the next cycle to obtain time intervals, calculate the predicted temperature difference of any time interval, count the number of times the predicted temperature difference is greater than the temperature difference threshold, and calculate the temperature fluctuation coefficient based on the number.

8. The intelligent production control device for amino acid formula powder according to claim 6, characterized in that: The state evaluation coefficient evaluation module also includes: an initialization update parameter module, a training data acquisition module, a training data input module, a parameter judgment module and a PID parameter determination module: The initialization update parameter module is used to determine the topological structure of the LSTM model, obtain the number of hyperparameters in the topological structure, and update the hyperparameters through the whale algorithm to obtain initial update parameters, and update the LSTM model according to the initial update parameters to obtain a target LSTM model; The training data acquisition module is used to set a preset temperature, and calculate the historical tank temperature data and the tank temperature data according to the preset temperature to obtain a historical temperature difference data set and a temperature difference data set; the temperature difference is the difference between the preset temperature and the tank temperature data; The training data input module is used to input the tank temperature data set, the preset temperature, the historical temperature difference data set and the temperature difference data set into the target LSTM model for training; The parameter judgment module is used to output the model result and use the output model result as the PID control parameter until the temperature difference data is less than the preset judgment value, otherwise, return to step 2; The PID parameter determination module is used to control the temperature of the fermentation tank according to the PID control parameters.

9. The intelligent production control device for amino acid formula powder according to claim 8, characterized in that: The initialization update parameter module includes: an update mode confirmation module, an update mode judgment module, a first update mode module, a second update mode module and a third update mode module: The update mode confirmation module is used to screen the whale individuals according to the fitness to obtain a judgment group, randomly select the whale individual and determine the fitness of the whale individual; the judgment group includes the fitness of the first best whale, the second best whale and the third best whale; The update mode judgment module is used to judge whether the second condition of rand is satisfied if the fitness of the individual whale belongs to any one of the judgment groups; The first update mode module is used to determine that rand satisfies the second condition if the fitness of the individual whale belongs to any one of the judgment groups, and rand < a preset threshold, and then execute the Levy flight strategy; The second update mode module is used to determine that rand satisfies the second condition if the fitness of the whale individual belongs to any one of the judgment groups, and rand ≥ a preset threshold, and then execute a random walk strategy; The third update mode module is used to determine that rand does not meet the second condition and execute the original whale optimization strategy if the fitness of the whale individual does not belong to any of the judgment groups.

10. The intelligent production control device for amino acid formula powder according to claim 8, characterized in that: The PID parameter determination module is also used to: control the temperature system of the fermentation tank according to the PID control parameters, obtain the second environmental parameters and the second temperature fluctuation coefficient in the fermentation tank when the second preset time is reached, calculate the second state evaluation coefficient according to the second environmental parameters and the second temperature fluctuation coefficient, and evaluate the fermentation stock solution according to the second state evaluation coefficient.

Images