An intelligent air supply cooling control method and device

Abstract

The present invention discloses an intelligent air supply cooling control method and device. The method includes: S1: constructing a first relationship model among the tenderness of livestock and poultry meat, the air supply temperature, and the air supply speed; S2: constructing a second relationship model among the water loss rate of the livestock and poultry meat, the air supply temperature, and the air supply speed; S3: controlling the air supply temperature and the air supply speed according to the target tenderness and the target water loss rate, and cooling the livestock and poultry meat by using the first relationship model and the second relationship model. The present invention can accurately control the air supply temperature and the air supply speed according to quality indexes such as tenderness and water loss rate, improve the cooling speed and uniformity of livestock and poultry meat, and reduce the cooling energy consumption.

Description

Technical Field

[0001] The present invention relates to the technical field related to the cooling of livestock and poultry meat. More specifically, the present invention relates to an intelligent air supply cooling control method and device. Background Art

[0002] Fresh meat is the main form of meat consumption in China. A series of physiological and biochemical reactions occur inside the muscles of livestock and poultry after slaughter, and the quality of livestock and poultry meat gradually deteriorates. Cooling is a necessary step in the post-slaughter production process of livestock and poultry, which can reduce the temperature of livestock and poultry meat and inhibit the process of spoilage. Currently, the cooling of livestock and poultry meat generally adopts the form of air cooling. The air coolers are arranged at the top on one side of the cold storage, and the air supply is inaccurate and mostly uncontrollable. There are problems such as uneven cooling of livestock and poultry meat, slow cooling speed, and high energy consumption, resulting in a decline in the quality of livestock and poultry meat and causing great economic losses to production enterprises.

[0003] The air supply temperature and air supply speed are the main parameters affecting the cooling speed of livestock and poultry meat. Low temperature and high wind speed can significantly improve the cooling speed, but a lower temperature will generate greater energy consumption, and a higher wind speed will cause a greater water loss rate. Therefore, in the cooling process, it is necessary to balance the quality of livestock and poultry meat and the air supply parameters. The prior art cannot directly judge the air supply parameters from the quality of livestock and poultry meat. It is necessary to design a technical solution that can overcome the above defects, intelligently control parameters such as the air supply temperature and air supply speed according to the quality of livestock and poultry meat, and maintain the quality of livestock and poultry meat while reducing the cooling energy consumption. Summary of the Invention

[0004] An object of the present invention is to provide an intelligent air supply cooling control method and device, which can accurately control the air supply temperature and air supply speed according to quality indicators such as tenderness and water loss rate, improve the cooling speed and uniformity of livestock and poultry meat, and reduce the cooling energy consumption.

[0005] To achieve these objects and other advantages of the present invention, according to one aspect of the present invention, there is provided an intelligent air supply cooling control method for controlling the air supply cooling process of livestock and poultry meat, including: S1: constructing a first relationship model among the tenderness, air supply temperature, and air supply speed of livestock and poultry meat; S2: constructing a second relationship model among the water loss rate, air supply temperature, and air supply speed of the livestock and poultry meat; S3: controlling the air supply temperature and the air supply speed according to the target tenderness and target water loss rate, and using the first relationship model and the second relationship model to cool the livestock and poultry meat.

[0006] Further, the specific method for constructing the first relationship model in step S1 includes: S1-1: During the cooling process of livestock and poultry meat, temperature data is collected in real time through multi-point temperature sensors arranged on the surface and inside of the livestock and poultry meat, and the air supply temperature T and the air supply speed U are recorded synchronously; S1-2: The tenderness value SF under different air supply temperatures T and air supply speeds U is obtained through a tenderness meter; S1-3: Based on the thermodynamic model, calculate the temperature difference ΔT between the central temperature and the surface temperature of the livestock and poultry meat, and combine the air supply temperature T and the air supply speed U to construct a mapping relationship using a BP neural network: SF = f(T, U, ΔT).

[0007] Further, the specific method for constructing the mapping relationship in step S1-3 includes:

[0008] S1-3-1: Based on the thermodynamic model, calculate the temperature difference ΔT between the central temperature Tc and the surface temperature Ts of the livestock and poultry meat through the Fourier heat conduction equation, where the thermal conductivity k is dynamically adjusted according to the type of livestock and poultry meat;

[0009] S1-3-2: Take the air supply temperature T, the air supply speed U, and the temperature difference ΔT as input variables, and the tenderness value SF as the output variable to construct a BP neural network model;

[0010] S1-3-3: Use the Levenberg-Marquardt algorithm to train the BP neural network, where the number of hidden layer nodes H is dynamically determined according to the number of input variables n: ; where, m is the number of output variables, a is a constant, and the value range is 1-10;

[0011] S1-3-4: Optimize the neural network parameters through the cross-validation method, and use the mean square error MSE as the loss function to ensure that the model prediction accuracy MSE < 0.01.

[0012] Further, the dynamic determination method of the number of hidden layer nodes H in step S1-3-3 also includes: During each iterative training process, dynamically adjust the value of a according to the mean square error MSE of the current model: If the MSE decline rate for three consecutive iterations is lower than 5%, then let , otherwise, if MSE rises, then let ; Map the adjusted number of hidden layer nodes H to an integer, and limit its minimum value to 5 and maximum value to 50 to control the network complexity.

[0013] Further, the specific method for constructing the second relationship model in step S2 includes: S2-1: Real-time monitoring of the quality change during the cooling process of livestock and poultry meat through a high-precision weight sensor, calculating the water loss rate DL, and synchronously recording the air supply temperature T, air supply speed U, and ambient humidity RH; S2-2: Based on the thermodynamics-moisture migration coupling model, decomposing the water loss rate DL into surface evaporation loss and internal moisture migration loss, taking the air supply temperature T, air supply speed U, and ambient humidity RH as input variables and the water loss rate DL as the output variable, constructing a non-linear regression model optimized by a genetic algorithm: , where , t is the cooling time; where k1 and k2 are dynamic parameters, which are real-time optimized according to the type of livestock and poultry meat through a particle swarm algorithm.

[0014] Further, a feature vector F=(f1,f2,f3) is established according to the muscle fiber density, myoglobin content, and thermal denaturation threshold of the type of livestock and poultry meat, and F is mapped to the parameter space (k1,k2) through a genetic algorithm-particle swarm hybrid optimization, where:

[0015] The genetic algorithm adopts tournament selection, uniform crossover, and adaptive mutation operations, globally searching the parameter range k1∈[0.1,1.2], k2∈[0.8,2.0], and the particle swarm algorithm is based on the inertia weight decreasing strategy for local optimization, with a convergence error ≤0.05, iter represents the current iteration number, itermax represents the maximum iteration number of the algorithm operation; the weight adjustment formula is: , where ρ muscle is the muscle density, T is the air supply temperature, k i is obtained through genetic algorithm-particle swarm hybrid optimization, and k i ′ is the final dynamic parameter.

[0016] Further, step S3 includes: S3-1: According to the target tenderness and the target water loss rate , a candidate parameter set of the air supply temperature T and air supply speed U is generated through a multi-objective optimization algorithm, where the optimization objective function is:

[0017] In the formula, SF and DL are respectively calculated in real time by the first relationship model and the second relationship model, α and β are weight coefficients, which are dynamically allocated according to the type of livestock and poultry meat through the entropy weight method, and α + β = 1; S3-2: Based on the real-time collected surface temperature Ts, center temperature Tc, and ambient humidity RH of the livestock and poultry meat, the candidate parameter set is roll-optimized through model predictive control, and the optimal solution satisfying the following constraint conditions is screened: T min ≤T≤T max , U min ≤U≤Umax , |dT / dt| ≤ 5 °C / min; where, T min and T max is set to -35 °C to 10 °C, U min and U max is set to 0.5 m / s to 8 m / s.

[0018] Furthermore, the step S3 further includes: correcting the model parameters at intervals of Δt time. The specific method is: performing residual analysis on the actual tenderness SF' at the current moment and the model prediction value SF. If the residual , then trigger the online fine-tuning of the BP neural network; according to the deviation between the measured water loss rate DL' and the model prediction value DL, update the dynamic parameters k1 and k2 in the second relationship model through an adaptive Kalman filter.

[0019] According to another aspect of the present invention, a control device is further provided, including: a refrigeration module, which includes a refrigeration unit and a air supply duct. A plurality of air supply outlets are provided on the air supply duct for respectively supplying air to a plurality of livestock and poultry meats; a control module, which pre-sets a first relationship model and a second relationship model, and is used to control the air supply temperature and air supply speed of the air supply outlets according to the target tenderness and target water loss rate, and cool the livestock and poultry meats by using the first relationship model and the second relationship model.

[0020] Furthermore, it further includes a conveying track and a conveying device connected to the conveying track. The conveying device is used to load or clamp livestock and poultry meats. The air supply duct is arranged above the conveying track, and the air supply outlets are arranged facing the conveying device.

[0021] The present invention has at least the following beneficial effects:

[0022] The present invention constructs a first relationship model among tenderness, air supply temperature, and air supply speed, and constructs a second relationship model among the water loss rate of livestock and poultry meats, air supply temperature, and air supply speed. Furthermore, according to the target tenderness and target water loss rate, the air supply temperature and air supply speed are controlled by using the first relationship model and the second relationship model to quickly cool the livestock and poultry meats.

[0023] During the cooling process, the present invention can perform precise air supply according to the quality of livestock and poultry meats, reducing the problems of slow cooling speed and uneven temperature distribution during the cooling process. It can realize intelligent control of the cooling process of different types of livestock and poultry meats, reduce the energy consumption of the cooling system, avoid economic losses caused by reduced cooling quality, and reduce the production economic cost of enterprises.

[0024] Other advantages, objectives, and features of the present invention will be partially reflected by the following description, and partially will also be understood by those skilled in the art through the research and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a flowchart of an embodiment of the present application;

[0026] Figure 2 is a schematic structural diagram of a cooling device of an embodiment of the present application from one angle;

[0027] Figure 3 is a schematic structural diagram of the cooling device of an embodiment of the present application from another angle. Detailed implementation manners

[0028] The following further describes the present invention in detail with reference to the accompanying drawings so that those skilled in the art can implement it according to the description in the specification.

[0029] It should be understood that terms such as "having", "including", and "comprising" used in the embodiments of the present application do not exclude the presence or addition of one or more other elements or combinations thereof. All directional indications (such as up, down, left, right, front, back...) in the embodiments of the present application are only used to explain the relative positional relationship and movement conditions between components in a specific posture. If the specific posture changes, the directional indications will also change accordingly. When an element is referred to as being "fixed to" or "disposed on" another element, it can be directly on the other element or there may be an intermediate element at the same time. When an element is referred to as being "connected" to another element, it can be directly connected to the other element or can also be indirectly connected to the other element through an intermediate element. The descriptions involving "first", "second", etc. in the embodiments of the present application are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include at least one of such features.

[0030] It should be noted that the technical solutions between various embodiments of the present application can be combined with each other, but it must be based on the ability of those of ordinary skill in the art 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 protection scope required by the present application.

[0031] As Figure 1 shown, an embodiment of the present application provides an intelligent air supply cooling control method, including:

[0032] S1: Construct a first relationship model among the tenderness, air supply temperature, and air supply speed of livestock and poultry meat;

[0033] Exemplarily, pre-conduct a cooling test to obtain the tenderness data of livestock and poultry meat at different air supply temperatures and air supply speeds, and then use methods such as regression models, machine learning, and neural networks to construct the first relationship model;

[0034] S2: Construct a second relationship model among the water loss rate of the livestock and poultry meat, the air supply temperature, and the air supply speed;

[0035] Exemplarily, during the cooling test, obtain the water loss rate data of the livestock and poultry meat under different air supply temperatures and air supply speeds, and then use methods such as regression models, machine learning, and neural networks to construct the second relationship model;

[0036] S3: According to the target tenderness and the target water loss rate, use the first relationship model and the second relationship model to control the air supply temperature and the air supply speed, and cool the livestock and poultry meat;

[0037] Exemplarily, determine the target tenderness and the target water loss rate according to the quality requirements of the livestock and poultry meat, then use the first relationship model and the second relationship model to inversely deduce the air supply temperature and the air supply speed, and use the obtained air supply temperature and air supply speed to cool the livestock and poultry meat to be cooled;

[0038] In this embodiment, the air supply temperature, the air supply speed, etc. are dynamically adjusted according to the target water loss rate and the target tenderness of the livestock and poultry meat, reducing the blindness of the cooling process, being able to achieve the purpose of maintaining the quality of the livestock and poultry meat, and thus can help reduce the losses of production enterprises.

[0039] In another embodiment, during the cooling process of the livestock and poultry meat, temperature data can be collected in real time through multi-point temperature sensors arranged on the surface and inside of the livestock and poultry meat. The surface temperature sensors can be arranged in the subcutaneous tissue layer of the livestock and poultry meat, evenly distributed at intervals of 5 cm to 10 cm; the internal temperature sensors can be inserted into the center position of the muscle, with a depth of 1 / 3 to 1 / 2 of the meat block thickness. The measurement of the air supply temperature T can use a PT100 platinum resistance temperature sensor, with a range of -50°C to 50°C and an accuracy of ±0.1°C; the air supply speed U can be measured by a hot-wire anemometer, with a range of 0.1 m / s to 10 m / s. The assembly positions of the temperature sensors and the anemometer need to be 0.5 m to 1 m away from the air outlet to avoid the direct impact of the air flow affecting the measurement accuracy. Data synchronous recording can be achieved through a data acquisition card (such as NI-9213), and the sampling frequency is set to 1 Hz to 10 Hz.

[0040] The determination of the tenderness value SF can use a tenderness meter. Cook the cooled meat sample in water at 71°C for 35 min (the center temperature of the sample reaches 71°C), then cool it in running water for 30 min, cut it into small meat strips of 1.0 cm × 1.0 cm × 1.5 cm, and use the tenderness meter to measure the shear force of the sample. In the experiment, the livestock and poultry meat samples of the same batch need to be grouped, with the sample size of each group being 20 to 30 pieces to ensure statistical significance (p < 0.05).

[0041] The difference ΔT between the core temperature Tc and the surface temperature Ts of livestock and poultry meat can be calculated using the Fourier heat conduction equation. The thermal conductivity coefficient k ranges from 0.4 W / (m·K) to 0.6 W / (m·K) and is dynamically adjusted based on the type of meat (e.g., pork, beef). The input variables of the BP neural network model include T (-35°C to 10°C), U (0.5 m / s to 8 m / s), and ΔT (1°C to 10°C). The output variable is SF. The number of hidden layer nodes in the neural network can be set to 8 to 15, the activation function is sigmoid, and the number of training cycles is 500 to 1000. The model training tool can be MATLAB Neural Network Toolbox, and the mean square error (MSE) must be less than 0.01.

[0042] This example utilizes a multi-point sensor layout and precise measurement equipment to ensure the real-time and reliability of temperature and wind speed data, providing high-quality input for model construction. Combining thermodynamic models with neural networks enables tenderness prediction using multi-parameter coupling, providing a theoretical basis for optimizing air supply parameters.

[0043] In another embodiment, when calculating the difference ΔT between the core and surface temperatures of livestock and poultry meat, the Fourier heat conduction equation can be used. The thermal conductivity coefficient k can be set in the range of 0.4 W / (m·K) to 0.6 W / (m·K), and can be dynamically adjusted based on the type of livestock and poultry meat. For example, the thermal conductivity coefficient k for pork can be 0.45 W / (m·K), and for beef can be 0.55 W / (m·K). The calculation process can be performed using finite element analysis software (such as COMSOL Multiphysics). The meshing accuracy can be set to 1 mm to 5 mm, and the time step can be 10 seconds to 60 seconds. The temperature sensor can be a PT100 platinum resistance probe. The surface temperature sensor should be placed 1 cm to 2 cm from the edge of the meat, and the core temperature sensor should be inserted to a depth of 1 / 3 to 1 / 2 of the meat thickness.

[0044] The input variables include supply air temperature T (range -35°C to 10°C), supply air velocity U (range 0.5 m / s to 8 m / s), and temperature difference ΔT (range 1°C to 10°C). The output variable is tenderness value SF. The number of input layer nodes in the BP neural network model is fixed at 3, and the number of output layer nodes is 1. Data normalization can be performed using the Z-score method. The training dataset can be divided into training and test sets in a 7:3 ratio.

[0045] The dynamic calculation formula for the number of hidden layer nodes H is: , where n=3 (number of input variables), m=1 (number of output variables), and the value of a ranges from 1 to 10. During the training process, if the mean square error (MSE) decreases less than 5% for three consecutive iterations, then ; If the MSE increases, then let , and limit the minimum value of H to 5 and the maximum value to 50. The Levenberg-Marquardt algorithm can be implemented through MATLAB's Neural Network Toolbox. The initial learning rate is set to 0.001, and the maximum number of iterations is 1000. The activation function can be selected as the Sigmoid function, and the output layer uses a linear function.

[0046] Cross-validation can adopt the K-fold cross-validation method, and the value of K is set to 5 to 10. The threshold of the mean square error MSE needs to be less than 0.01. If not up to the standard, the number of hidden layer nodes or the number of training times needs to be readjusted. The optimized model parameters can be exported through MATLAB's Regression Learner toolbox and embedded in the embedded processor of the control system for operation. The prediction error analysis of the test set can be carried out by calculating the mean absolute percentage error (MAPE), and it is required that MAPE ≤ 3%.

[0047] In this embodiment, by dynamically adjusting the thermal conductivity coefficient k, it adapts to the thermodynamic characteristics of different meats, ensures the accuracy of ΔT calculation, and provides reliable input for the neural network. Standardize the range and dimension of the input data, improve the model convergence speed and generalization ability, and avoid training bias caused by magnitude differences. Dynamically adjust the number of hidden layer nodes, balance the model complexity and computational efficiency, prevent overfitting or underfitting, and ensure the prediction accuracy. Through cross-validation and strict error thresholds, ensure the robustness of the model and meet the deployment requirements of the real-time control system.

[0048] In another embodiment, during the iterative training process of the BP neural network, the calculation parameter a of the number of hidden layer nodes can be dynamically adjusted according to the change of the mean square error (MSE). The specific adjustment rule is: if the MSE decline rate in 3 consecutive iterations is lower than 5% (threshold), then let ; If the MSE increases, then let . During the training process, the change of MSE can be monitored in real time through MATLAB's NeuralNetwork Toolbox, and the code for triggering the condition judgment can be embedded in the training loop.

[0049] The calculated value of H needs to be mapped to an integer, which can be achieved by rounding or truncating. At the same time, the value range of H is limited to a minimum of 5 and a maximum of 50. For example, when the calculated value of H is 4.6, it is rounded to 5; when the calculated value of H is 51, it is forced to be set to 50. This limitation can be implemented through programming to prevent the network complexity from being too high or too low.

[0050] In this embodiment, the adaptive optimization of the network structure is realized through the threshold determination of the MSE change rate (5% decrease rate), balancing the model convergence speed and stability. The value range of H is restricted (5 ≤ H ≤ 50) to avoid overfitting caused by too many nodes or underfitting caused by too few nodes, and improve the model generalization ability.

[0051] In another embodiment, in step S2-1, the quality change during the cooling process of livestock and poultry meat can be monitored in real time by a high-precision weight sensor, and the water loss rate DL is calculated. The weight sensor can be selected as the METTLER TOLEDO ICS489 type with a measuring range of 0.1 kg to 20 kg and an accuracy of ±0.01 g. The installation position can be set below the hook for hanging livestock and poultry meat or at the support point of the conveying equipment. The environmental humidity RH can be measured by an HMP155 type temperature and humidity sensor with a measuring range of 0% to 100% and an accuracy of ±1.5%. The range of the air supply speed U can be set to 0.5 m / s to 8 m / s, and the corresponding wind speed sensor can be installed at a position 0.3 m to 0.8 m downstream of the air supply outlet. The calculation formula of the water loss rate DL is DL = (m0 - m t ) / m0, where m0 is the initial mass, and m t is the mass at time t, and the sampling interval can be set to 1 minute to 5 minutes.

[0052] In step S2-2, the calculation of the surface evaporation loss can be realized through the improved Penman-Monteith equation, where the influence factor of the air supply speed U is applicable to the wind speed range of 0.5 m / s to 8 m / s. The detection of the internal moisture migration loss can select the Niumai Technology PQ001 type low-field nuclear magnetic resonance instrument, and the detection parameters are set as the echo time of 0.5 ms and the number of scans of 16 times. The probe of the nuclear magnetic resonance instrument can be installed outside the cooling chamber to avoid interfering with the air flow through non-invasive measurement. The range of the air supply temperature T can be set to -35°C to 10°C.

[0053] Nonlinear regression model The dynamic parameters k1 and k2 can be optimized by the particle swarm algorithm, t is the cooling time, and T can use the Kelvin temperature. The population size of the particle swarm algorithm can be set to 20 to 50, the number of iterations is 100 to 300 times, and the convergence error threshold is set to 0.05. The value range of the parameter k1 can be limited to 0.1 to 1.2, and k2 is 0.8 to 2.0.

[0054] In this embodiment, through a high-precision sensor combination, synchronous monitoring of mass, temperature, humidity, and wind speed is achieved, providing highly reliable data input for the water loss rate model. By combining nuclear magnetic resonance detection and an improved evaporation model, the water loss contributions of surface evaporation and internal migration are accurately quantified, enhancing the physical interpretability of water loss rate prediction. The parameters are dynamically optimized through the particle swarm algorithm to adapt to the thermodynamic characteristics of different types of livestock and poultry meat, reducing the model error to an acceptable range (MSE < 0.05).

[0055] In another embodiment, when establishing the feature vector F = (f1, f2, f3), it can be based on the muscle fiber density, myoglobin content, and thermal denaturation threshold of the livestock and poultry meat type. The muscle fiber density f1 can be measured by microscopic image analysis, and the sample slice thickness can be set to 10 to 20 microns, and the number of muscle fibers per square millimeter generally ranges from 50 to 200. The myoglobin content f2 can be measured using a spectrophotometer at a wavelength of 540 nm, with a measurement range of 0.1 mg / g to 5.0 mg / g. The thermal denaturation threshold f3 can be measured by a differential scanning calorimeter (DSC) with a temperature scanning rate of 2°C / min to 5°C / min and a thermal denaturation temperature range of 55°C to 75°C. The normalization process of the feature vector can be achieved through the Min - Max normalization method, mapping each parameter to the interval [0, 1].

[0056] The global search parameter range of the genetic algorithm - particle swarm hybrid optimization can be set as k1 ∈ [0.1, 1.2], k2 ∈ [0.8, 2.0]. The tournament selection scale of the genetic algorithm can be set to 3 to 5, the uniform crossover probability is 0.6 to 0.8, and the adaptive mutation probability is 0.01 to 0.1. The inertia weight decreasing formula of the particle swarm algorithm is , where the total number of iterations itermax can be set to 100 to 300 times, and the convergence error threshold ≤ 0.05. The optimization process can be implemented through the GlobalOptimization Toolbox of MATLAB, and the particle swarm population size can be set to 20 to 50. The termination condition of the hybrid optimization algorithm can be set to that the change rate of the fitness function in 10 consecutive iterations is lower than 1%.

[0057] The muscle density ρ in the weight adjustment formula muscle can have a value range set to 1.05 g / cm³ to 1.1 g / cm³, and the density of water ρ water is 1.0 g / cm³. The input range of the supply air temperature T can be limited to - 35°C to 10°C. The formula The temperature correction coefficient exp(0.1⋅T) in it needs to be dynamically calculated according to the actual temperature. For experimental verification, 20 groups of samples of pork and beef can be selected, and the goodness of fit (R²≥0.9) between the adjusted parameters k1' and k2' and the measured water loss rate can be verified through linear regression analysis. Through F-test evaluation (p<0.05), there is no significant difference between the model prediction value and the measured value.

[0058] In this embodiment, a feature vector is established through multi-dimensional meat quality parameters, which improves the ability of the model to distinguish between different types of livestock and poultry meat and enhances the physical relevance of parameter optimization. By combining the global search of the genetic algorithm and the local optimization of the particle swarm, the search efficiency and accuracy are balanced to ensure that the dynamic parameters converge within a reasonable range. The introduction of muscle density and temperature correction factors improves the adaptability of the model to the actual cooling environment and reduces the water loss rate prediction error to less than 0.05.

[0059] In another embodiment, in step S3-1, the objective function can be set as

[0060] , where SF and DL are calculated in real time by the first relational model and the second relational model respectively. The value ranges of the weight coefficients α and β are both from 0 to 1, and satisfy α + β = 1. The weight allocation can be dynamically adjusted by the entropy weight method, and the calculation of the entropy weight method can be based on the coefficient of variation of tenderness and water loss rate in historical data, and the data sample size can be set to 100 to 200 groups. The multi-objective optimization algorithm can select the NSGA-II algorithm, the population size can be set to 50 to 100, the crossover probability is 0.7 to 0.9, and the mutation probability is 0.01 to 0.1. The generation of the candidate parameter set can be realized through the Global Optimization Toolbox of MATLAB, and 10 to 20 groups of (T, U) candidate parameters are generated in each iteration.

[0061] In step S3-2, the constraint range of the air supply temperature T can be set to -35°C to 10°C, the range of the air supply speed U is 0.5 m / s to 8 m / s, and the threshold of the temperature change rate |dT / dt| is set to 5°C / min. Real-time data acquisition can be realized through a PT100 temperature sensor (surface temperature Ts) and a thermocouple probe (central temperature Tc). The surface temperature sensor can be installed 1 cm to 2 cm below the surface of the meat block, and the insertion depth of the central temperature sensor is 1 / 2 to 2 / 3 of the thickness of the meat block. The measurement of the environmental humidity RH can select an HMP155 type temperature and humidity sensor, and the installation position is 1 m to 1.5 m away from the air supply outlet. The rolling optimization process can be realized through the CVXPY library of Python, the prediction time domain can be set to 5 minutes to 10 minutes, and the control time domain is 2 minutes to 5 minutes.

[0062] In this embodiment, the weights are dynamically allocated by the entropy weight method to balance the optimization objectives of tenderness and water loss rate, and to meet the quality requirements of different types of livestock and poultry meat.

[0063] In another embodiment, during the process of correcting the model parameters in step S3, the actual tenderness SF' at the current moment can be measured in real time by a TA.XT Plus texture analyzer, and a residual analysis can be performed with the predicted value SF of the first relationship model. The residual threshold can be set to 0.5 N. When it exceeds this threshold, online fine-tuning of the BP neural network is triggered. The fine-tuning process can adopt an incremental learning algorithm, the learning rate can be set to 0.001 to 0.01, and the fine-tuning data window can select historical data in the most recent 30 minutes to 60 minutes. Data acquisition can be achieved through a NI-9213 data acquisition card, and the sampling frequency is 1 Hz to 5 Hz. The texture analyzer can be installed at the end of the conveying track of the cooling device, 1.5 meters to 2 meters away from the air supply outlet, to avoid air flow interfering with the measurement results.

[0064] The measured water loss rate DL' can be obtained through a Mettler-Toledo ICS489 weight sensor, and the sampling interval is 5 minutes to 10 minutes. The update of the dynamic parameters k1 and k2 of the second relationship model can be achieved through an adaptive Kalman filter. The filter parameters can be set as the process noise covariance Q = 0.01 to 0.1, and the observation noise covariance R = 0.1 to 0.5. The parameter update period Δt can be set to 15 minutes to 30 minutes, synchronized with the sampling frequency of the weight sensor. For the verification experiment, 20 groups of beef samples can be selected. By comparing the model prediction errors (such as RMSE) before and after the update, it is required that the decline amplitude of RMSE after the update is ≥10%, and the significance is verified through a paired t-test (p < 0.05).

[0065] In this embodiment, by setting a clear residual threshold (0.5 N), real-time monitoring and automatic correction of the model prediction error are achieved, improving the long-term prediction accuracy of the first relationship model. Combining the weight sensor data with the adaptive filtering algorithm ensures the dynamic optimization of the parameters of the second relationship model with the change of the environment. Using the air supply temperature and air supply speed determined in the above embodiment to process beef, after testing, the deviation between the actual tenderness and the target value is controlled within ±0.5 N, and the deviation between the actual water loss rate and the target value is ≤0.3%.

[0066] See Figure 2 and Figure 3 , the embodiments of the present application also provide an intelligent air supply cooling control device, including a refrigeration module and a control module:

[0067] The refrigeration module may include a refrigeration unit and an air supply duct 1. The refrigeration unit can be selected from BITZER ECOLINE series, with a refrigeration capacity range of 10 kW to 50 kW. The air supply duct 1 can be made of galvanized steel, with a thickness of 0.8 mm to 1.2 mm. Multiple circular air outlets 101 can be provided on the air supply duct 1, with a diameter of the air outlet 101 ranging from 10 cm to 20 cm and a spacing between adjacent air outlets 101 of 0.5 m to 1.5 m. The control range of the air supply temperature is -35°C to 10°C, and the adjustment range of the air supply speed is 0.5 m / s to 8 m / s. The compressor of the refrigeration unit can be installed outside the cooling chamber and connected to the evaporator through a copper refrigerant pipeline. The evaporator is located at the inlet of the air supply duct 1, with a distance of 1.5 m to 2 m from the air outlet 101. The installation position of the temperature sensor can be set at a place 0.3 m to 0.8 m downstream of the air outlet 101, and the vertical height from the surface of the livestock and poultry meat 4 is 0.5 m to 1 m.

[0068] The control module can preset the first relationship model and the second relationship model of the foregoing embodiments, and the hardware can be selected as a Siemens S7-1200 type PLC. The model operation can be run by embedding the C code generated by MATLAB Coder into the PLC, and the sampling period is set to 5 seconds to 10 seconds. The control accuracy requirement of the air supply temperature is ±0.5°C, and the control error of the air supply speed is ≤0.1 m / s. The input interface of the control module can be connected to a PT100 temperature sensor and an HMP155 type humidity sensor, and the output interface communicates with the variable frequency fan and the refrigeration unit through the Modbus protocol. The installation position of the PLC can be inside the control cabinet.

[0069] The conveying track 2 can be made of 304 stainless steel, with a track width of 60 cm to 80 cm, and the surface is anti-slip treated. The conveying equipment 3 can be selected as a food-grade PP material hanging fixture, with a fixture spacing of 30 cm to 50 cm and a single load capacity of 20 kg to 50 kg. The air supply duct 1 can be arranged in parallel directly above the conveying track 2 at a height of 1.2 m to 1.8 m, and the air outlet 101 is vertically downward and aligned with the livestock and poultry meat 4 on the fixture. The conveying speed can be adjusted to 0.1 m / s to 0.3 m / s.

[0070] In this embodiment, through modular design and precise parameter control, the stability of the air supply temperature and speed is ensured, and the energy consumption fluctuation during the cooling process is reduced. With the help of high-precision sensors and embedded model algorithms, the real-time dynamic adjustment of the air supply parameters is realized, and the control response speed and accuracy are improved. Through the standardized conveying track 2 and adjustable fixture design, the continuity and uniformity of the cooling operation are improved, and the need for manual intervention is reduced.

[0071] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and embodiments. It can be fully applied to various fields suitable for the present invention. For those skilled in the art, additional modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and the illustrated examples described herein.

Claims

1. An intelligent air supply cooling control method for controlling the air supply cooling process of livestock and poultry meat, characterized in that, Including: S1: Construct a first relationship model among the tenderness, air supply temperature, and air supply speed of livestock and poultry meat; S2: Construct a second relationship model among the water loss rate, air supply temperature, and air supply speed of the livestock and poultry meat; S3: According to the target tenderness and target water loss rate, use the first relationship model and the second relationship model to control the air supply temperature and air supply speed, and cool the livestock and poultry meat; The specific method for constructing the first relationship model in S1 includes: S1-1: During the cooling process of livestock and poultry meat, temperature data is collected in real time through multi-point temperature sensors arranged on the surface and inside of the livestock and poultry meat, and the air supply temperature T and air supply speed U are recorded synchronously; S1-2: Obtain the tenderness value SF of livestock and poultry meat under different air supply temperatures T and air supply speeds U through a tenderness meter; S1-3: Based on the thermodynamic model, calculate the temperature difference ΔT between the center temperature and the surface temperature of the livestock and poultry meat. Combining the air supply temperature T and the air supply speed U, use the BP neural network to construct a mapping relationship: SF = f(T, U, ΔT); The specific method for constructing the second relationship model in S2 includes: S2-1: Through a high-precision weight sensor, monitor the mass change during the cooling process of livestock and poultry meat in real time, calculate the water loss rate DL, and synchronously record the air supply temperature T, air supply speed U, and ambient humidity RH; S2-2: Based on the thermodynamics-moisture migration coupling model, the water loss rate DL is decomposed into surface evaporation loss and internal moisture migration loss. Taking the supply air temperature T, supply air velocity U, and ambient humidity RH as input variables and the water loss rate DL as the output variable, a non-linear regression model is constructed: , where , t is the cooling time, and T is in Kelvin temperature; where k1 and k2 are dynamic parameters, which are determined by hybrid optimization of genetic algorithm - particle swarm algorithm according to the type of livestock and poultry meat.

2. The intelligent air supply cooling control method according to claim 1, characterized in that, The specific method for constructing the mapping relationship in S1-3 includes: S1-3-1: Based on the thermodynamic model, calculate the temperature difference ΔT between the center temperature Tc and the surface temperature Ts of the livestock and poultry meat through the Fourier heat conduction equation, where the thermal conductivity k is dynamically adjusted according to the type of livestock and poultry meat; S1-3-2: Take the air supply temperature T, air supply speed U, and temperature difference ΔT as input variables, and the tenderness value SF as the output variable to construct a BP neural network model; S1-3-3: Train the BP neural network using the Levenberg-Marquardt algorithm, where the number of hidden layer nodes H is dynamically determined according to the number of input variables n: ; where m is the number of output variables, a is a constant, and the value range is 1 - 10; S1-3-4: Optimize the neural network parameters through the cross-validation method, and use the mean square error MSE as the loss function to ensure that the model prediction accuracy MSE < 0.

01.

3. The intelligent air supply cooling control method according to claim 2, wherein, The dynamic determination method of the number of hidden layer nodes H in S1-3-3 further includes: During each iteration training process, the value of a is dynamically adjusted according to the mean square error (MSE) of the current model: if the MSE decline rate for three consecutive iterations is lower than 5%, then let , otherwise, if the MSE increases, then let ; map the adjusted number of hidden layer nodes H to an integer and limit its minimum value to 5 and maximum value to 50 to control the network complexity.

4. The intelligent air supply cooling control method according to claim 1, characterized in that, Establish a feature vector F = (f1, f2, f3) based on the muscle fiber density, myoglobin content, and thermal denaturation threshold of the type of livestock and poultry meat, and map F to the parameter space (k1, k2) through hybrid optimization of genetic algorithm - particle swarm algorithm, where: The genetic algorithm adopts tournament selection, uniform crossover and adaptive mutation operations, globally searches the parameter ranges k1 ∈ [0.1, 1.2] and k2 ∈ [0.8, 2.0], and the particle swarm algorithm is based on the inertia weight decreasing strategy. For local optimization, the convergence error ≤ 0.05, iter represents the current number of iterations, and itermax represents the maximum number of iterations for the algorithm to run. The weight adjustment formula is as follows: , where ρ muscle is muscle density, ρ wate is water density, T is the air supply temperature, k i is obtained by hybrid optimization of genetic algorithm - particle swarm algorithm, k i ′ is the final dynamic parameter, i = 1 or 2.

5. The intelligent air supply cooling control method according to claim 4, wherein S3 includes: S3-1: According to the target tenderness and the target water loss rate , a candidate parameter set of the air supply temperature T and the air supply speed U is generated by a multi-objective optimization algorithm, where the optimization objective function is: In the formula, SF and DL are respectively calculated in real time by the first relational model and the second relational model, α and β are weight coefficients, which are dynamically allocated by the entropy weight method according to the type of livestock and poultry meat, and α + β = 1; S3-2: Based on the surface temperature Ts, center temperature Tc of the livestock and poultry meat and ambient humidity RH collected in real time, perform rolling optimization on the candidate parameter set through model predictive control, and screen the optimal solution that meets the following constraints: T min T ≤ T ≤ max , U min U ≤ U ≤ max , |dT / dt| ≤ 5 °C / min; Among them, T min and T max are set to -35°C to 10°C, and U min and U max are set to 0.5 m / s to 8 m / s.

6. The intelligent air supply cooling control method according to claim 5, wherein S3 also includes: Modify the model parameters at intervals of Δt time, and the specific method is: Perform a residual analysis on the actual tenderness SF' at the current moment and the model prediction value SF. If the residual is met, then trigger the online fine-tuning of the BP neural network; According to the deviation between the measured water loss rate DL' and the model predicted value DL, update the dynamic parameters k1 and k2 in the second relationship model through an adaptive Kalman filter.

7. An apparatus for implementing the intelligent air supply cooling control method according to any one of claims 1-6, characterized in that, Including: A refrigeration module, which includes a refrigeration unit and a air supply duct, and a plurality of air outlets are arranged on the air supply duct for respectively supplying air to a plurality of livestock and poultry meats; A control module, which pre-sets a first relationship model and a second relationship model, and is used to control the air supply temperature and air supply speed of the air outlets according to the target tenderness and target water loss rate, and cool the livestock and poultry meats by using the first relationship model and the second relationship model.

8. The device according to claim 7, characterized in that, It further includes a conveying track and a conveying device connected to the conveying track. The conveying device is used to load or clamp livestock and poultry meats. The air supply duct is arranged above the conveying track, and the air outlets are arranged facing the conveying device.

Images