Intelligent control method and system for agricultural product processing and transportation based on deep learning

Abstract

The application relates to the field of agricultural product cold chain logistics and control technology, and discloses an agricultural product processing and transportation intelligent control method and system based on deep learning, wherein the method comprises the following steps: acquiring mechanical stress data and thermal stress data of an agricultural product processing stage; inputting the stress integral vector and the carriage environment data into a deep learning prediction model to obtain a predicted respiration heat flow density sequence; when the first-order time derivative meets a preset triggering condition, performing integral operation on the predicted respiration heat flow density sequence to obtain a cooling compensation equivalent; and determining whether there is an evaporator frosting risk by comparing the dynamic dew point temperature and the estimated evaporator target pipe wall temperature. Through processing stress feedforward prediction and time sequence decoupling control, the application realizes early and accurate compensation of the surge of agricultural product respiration heat, eliminates the evaporator frosting hidden danger under high-humidity working conditions, and guarantees efficient operation of a refrigeration system and temperature stability of a carriage.

Description

Technical Field

[0001] This invention relates to the field of cold chain logistics and control technology for agricultural products, specifically to a method and system for intelligent control of agricultural product processing and transportation based on deep learning. Background Technology

[0002] In the post-harvest circulation of agricultural products, cold chain transportation is a core link in maintaining the quality of agricultural products. Existing cold chain transportation environmental control systems typically operate as independent closed-loop modules, with their control logic primarily relying on real-time data collected by temperature sensors deployed inside the transport vehicle. Before entering the transport vehicle, agricultural products undergo processing steps such as sorting and cleaning. In these upstream stages, agricultural products are subjected to physical stresses from mechanical collisions and thermal stresses from traversing different temperature zones. These stresses can cause latent physiological damage within the agricultural product tissue, leading to abnormal increases in respiratory metabolism and the release of large amounts of heat during subsequent transportation. Existing control technologies disconnect the data link between the processing and transportation stages; refrigeration equipment at the transportation end cannot predict the latent physiological states carried by agricultural products, resulting in a lack of effective feedforward intervention for environmental control.

[0003] Due to a lack of feedforward data, existing vehicle-mounted refrigeration units generally employ a passive feedback control strategy based on actual temperature deviations. When agricultural products enter their peak respiration heat release period, the dissipated heat first causes a rise in the local air temperature inside the vehicle. Only after sensors detect that the temperature exceeds the limit will the compressor be driven to increase the refrigeration load. This response mechanism is constrained by the physical hysteresis characteristics of the air-side heat exchange process and the refrigeration cycle itself, making it impossible to offset the heat dissipation in its initial stages. When the refrigeration system is fully engaged, irreversible heat accumulation and temperature fluctuations have already formed inside the vehicle, directly increasing the risk of spoilage for fresh agricultural products.

[0004] When dealing with such sudden respiratory heat loads, directly and drastically lowering the set temperature to force cooling output under conventional control logic will trigger specific thermodynamic engineering contradictions. Agricultural products release heat through respiration, accompanied by strong transpiration, maintaining a consistently high relative humidity within the enclosed space of the vehicle. Large temperature difference cooling commands force a rapid drop in evaporator wall temperature. Once this surface temperature falls significantly below the dynamic dew point and freezing point of the high-humidity air, water vapor in the air rapidly undergoes a phase change on the evaporator's metal fins, forming a dense frost layer. The accumulation of frost increases heat transfer resistance, hinders air circulation, and severely reduces the actual cooling capacity of the refrigeration unit. This situation frequently triggers the unit's defrosting program. The defrosting process not only interrupts cooling but also introduces additional heat into the vehicle interior, causing secondary fluctuations in ambient temperature. Existing control methods cannot simultaneously meet the demands of offsetting high-intensity heat loads and resolve the conflict between phase change and frost formation on the evaporator surface under high humidity conditions. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a deep learning-based intelligent control method and system for agricultural product processing and transportation. This solves the problems of existing cold chain transportation systems, which lack feedforward correlation with upstream processing stress and are unable to intervene in advance for the surge in respiratory heat of agricultural products, and whose passive response to high heat loads and direct cooling can easily lead to evaporator phase change frosting, reduced refrigeration efficiency, and secondary fluctuations in the temperature of the vehicle compartment under high humidity conditions.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The first aspect of this invention provides a deep learning-based intelligent control method for agricultural product processing and transportation, the method comprising the following steps:

[0008] S100, during the agricultural product processing stage, acquire and generate a stress integral vector based on the mechanical stress data and thermal stress data on the agricultural product processing line, and write the stress integral vector into the electronic tag of the loading box after the agricultural product processing is completed;

[0009] S200, during the transportation phase, the stress integral vector is read and the carriage environment data is collected. The stress integral vector and the carriage environment data are input into a deep learning prediction model to obtain a predicted respiratory heat flux density sequence.

[0010] S300, calculate the first time derivative of the predicted respiratory heat flux density sequence, generate a trigger signal based on the calculated first time derivative and a pre-configured positive slope threshold, and perform an integral operation on the predicted respiratory heat flux density sequence based on the trigger signal to obtain the cooling compensation equivalent.

[0011] S400: The dynamic dew point temperature is calculated based on the cabin environment data, and the estimated target evaporator pipe wall temperature is calculated based on the cooling compensation equivalent. The dynamic dew point temperature and the estimated target evaporator pipe wall temperature are compared to determine whether there is a risk of evaporator frosting.

[0012] S500: When there is a risk of frost formation on the evaporator, the opening of the electronic expansion valve and the speed of the evaporator inverter fan are adjusted upwards to reduce the sensible heat ratio in the refrigeration cycle physical process of the control system and adjust the operating state of the control system to isothermal dehumidification operation. At the same time, the dynamic dew point temperature is updated. Based on the estimated target pipe wall temperature of the evaporator and the updated dynamic dew point temperature, it is determined whether the updated dynamic dew point temperature has dropped to the preset safety margin. When the updated dynamic dew point temperature drops to the preset safety margin, the opening of the electronic expansion valve and the speed of the evaporator inverter fan are restored and adjusted to the operating state of the control system to sensible heat pre-cooling operation.

[0013] Preferably, step S100, which involves acquiring and generating a stress integral vector based on the mechanical stress data and thermal stress data on the agricultural product processing line, and writing the stress integral vector into the electronic tag of the loading box after the agricultural product processing is completed, includes:

[0014] Based on a triaxial accelerometer, mechanical collision acceleration signals of each sorting node and each transfer node in the agricultural product processing line are collected. Based on the mechanical collision acceleration signals of each sorting node and each transfer node, the effective acceleration values ​​of each sorting node and each transfer node are extracted. The extracted effective acceleration values ​​are used to generate an effective acceleration value sequence, and the effective acceleration value sequence is used as the mechanical stress data.

[0015] The temperature of the cleaning water bath and the processing environment are collected in real time by temperature sensors. The difference between the cleaning water bath temperature and the processing environment temperature at different time stamps is obtained and a difference sequence is generated. The difference sequence is used as the thermal stress data. The temperature sensors include a water bath temperature sensor and an ambient temperature sensor.

[0016] The processing duration parameter is obtained based on the start time of agricultural products entering the processing line and the end time of packaging and leaving the line. Based on the stress integral vector calculation formula, the mechanical stress data and the thermal stress data are integrated over time within the processing duration parameter to obtain the stress integral vector. The stress integral vector calculation formula is as follows: ;

[0017] In the formula, It is the stress integral vector; This is a parameter for processing duration; The effective acceleration values ​​at the same timestamp; The washing water bath temperature at the same timestamp; The processing environment temperature at the same timestamp; Representing time variables The differential element.

[0018] After agricultural product processing is completed, the stress integral vector is transmitted via an industrial control network and radio frequency writing equipment. Write it into the electronic tag of the corresponding loading box for agricultural products.

[0019] Preferably, in step S200, the steps of reading the stress integral vector and collecting the carriage environment data, and inputting the stress integral vector and the carriage environment data into a deep learning prediction model to obtain the predicted respiratory heat flux density sequence include:

[0020] The stress integral vector in the electronic tag of the loading box is read, and the environmental data of the car body within a preset time window is collected. The environmental data of the car body includes triaxial vibration acceleration sequence data, environmental dry-bulb temperature sequence data and relative humidity sequence data.

[0021] The stress integral vector is merged and spliced ​​with the triaxial vibration acceleration sequence data, environmental dry-bulb temperature sequence data and relative humidity sequence data within a preset time window to generate the model input matrix;

[0022] The model input matrix is ​​input into a deep learning prediction model, wherein the deep learning prediction model includes a cascaded architecture of a multilayer perceptron and a temporal convolutional network.

[0023] The deep learning prediction model uses a multilayer perceptron to receive the stress integral vector in the model input matrix and performs nonlinear mapping to obtain a high-dimensional hidden layer feedforward. The deep learning prediction model also uses a temporal convolutional network to receive the environmental dry-bulb temperature sequence data in the model input matrix and extract long-period dependency features from the environmental dry-bulb temperature sequence data.

[0024] Based on the deep learning prediction model, the high-dimensional hidden layer feedforward quantity and long-period dependent features are tensor concatenated and fully connected to obtain a fused feature vector. The fused feature vector is then processed through the linear output layer of the deep learning prediction model to calculate and obtain the predicted value of heat generation per unit volume at different times within the future prediction time window. Based on the predicted value of heat generation per unit volume at all times within the future prediction time window, a predicted respiratory heat flux density sequence is generated.

[0025] Preferably, step S300 further includes:

[0026] The predicted heat generation power per unit volume corresponding to data points at adjacent time steps in the predicted respiratory heat flux density sequence is obtained and subjected to finite difference operation to obtain the first time derivative of the predicted respiratory heat flux density sequence in the time dimension.

[0027] The obtained first-order time derivative is compared with the pre-configured positive slope threshold. When the obtained first-order time derivative is greater than the pre-configured positive slope threshold within a consecutive preset number of time steps, it is determined that the physiological metabolic latency period of agricultural products has ended and is about to enter the respiratory heat surge stage. A trigger signal is generated and sent. The pre-configured positive slope threshold is the upper limit of the normal physiological heat release rate fluctuation set by the control system.

[0028] Based on the trigger signal, the complete predicted respiratory heat flux density sequence within the future time window is extracted and subjected to time definite integral operation to obtain the total accumulated heat of agricultural products released into the enclosed space of the vehicle within the future time window; the pre-stored physical environment parameter data is obtained, and the cooling compensation equivalent is calculated based on the physical environment parameter data and the total accumulated heat.

[0029] Preferably, step S400 includes determining the dynamic dew point temperature and determining the estimated target evaporator wall temperature, wherein the step of determining the dynamic dew point temperature is as follows:

[0030] Extract the current sampling period's dry-bulb temperature sequence data and relative humidity sequence data from the carriage environment data, and calculate and determine the intermediate conversion coefficients based on the current sampling period's dry-bulb temperature sequence data and relative humidity sequence data.

[0031] The dynamic dew point temperature is calculated based on the intermediate conversion coefficient, as well as the ambient dry-bulb temperature sequence data and relative humidity sequence data of the current sampling period. The dynamic dew point temperature represents the temperature threshold at which a gas-liquid phase change occurs under the current air humidity conditions in the carriage.

[0032] The steps for determining the estimated target tube wall temperature of the evaporator are as follows:

[0033] The target set temperature of the current carriage is read, and the estimated target evaporator tube wall temperature is determined based on the target set temperature of the current carriage, the cooling compensation equivalent, and the pre-calibrated inherent heat exchange temperature difference parameters. The estimated target evaporator tube wall temperature... In the formula, Characterizes the target set temperature of the current carriage; Characterizing the cooling compensation equivalent, Characterizes the pre-calibrated inherent heat transfer temperature difference parameter;

[0034] The dynamic dew point temperature is compared with the estimated target evaporator wall temperature. When the estimated target evaporator wall temperature is less than or equal to the dynamic dew point temperature, it is determined that the evaporator is at risk of frosting, and the determination result is output.

[0035] Preferably, in step S500, when there is a risk of frost formation on the evaporator, the step of adjusting the opening of the electronic expansion valve and the speed of the evaporator inverter fan upwards to reduce the sensible heat ratio in the refrigeration cycle physical process of the control system, and adjusting the operating state of the control system to an isothermal dehumidification operating state further includes:

[0036] When the output judgment result indicates that there is a risk of frost formation on the evaporator, the target set temperature of the current compartment is locked, and a valve reduction command is sent to the stepper motor driver of the electronic expansion valve via the bus to reduce the physical opening of the electronic expansion valve, thereby reducing the refrigerant flow into the evaporator. The actual evaporator pipe wall temperature is limited to a specific physical range by adjusting the opening of the electronic expansion valve. The upper limit of the physical range is the dynamic dew point temperature calculated in the previous steps, and the lower limit is the freezing point temperature of pure water under standard atmospheric pressure.

[0037] While adjusting the electronic expansion valve, a frequency reduction command is simultaneously sent to the evaporator variable frequency fan to force the working speed of the evaporator variable frequency fan to be reduced to the preset lower limit frequency value.

[0038] By jointly adjusting the opening of the electronic expansion valve and the speed of the evaporator variable frequency fan, the operating state of the control system is adjusted to an isothermal dehumidification operating state.

[0039] Preferably, in step S500, the dynamic dew point temperature is updated simultaneously. Based on the estimated target evaporator wall temperature and the updated dynamic dew point temperature, it is determined whether the updated dynamic dew point temperature has dropped to a preset safety margin. When the updated dynamic dew point temperature drops to the preset safety margin, the opening of the electronic expansion valve and the speed of the evaporator variable frequency fan are restored and adjusted to adjust the operating state of the control system to the sensible heat pre-cooling operating state. This step specifically includes:

[0040] When the control system is in isothermal dehumidification operation mode, it continuously collects the ambient dry-bulb temperature sequence data and relative humidity sequence data inside the carriage at fixed time steps, and updates the dynamic dew point temperature synchronously based on the collected ambient dry-bulb temperature sequence data and relative humidity sequence data.

[0041] The updated dynamic dew point temperature is compared with the estimated target evaporator pipe wall temperature. When the estimated target evaporator pipe wall temperature is greater than the sum of the updated dynamic dew point temperature and the preset safety constant, it is determined that the absolute moisture content of the air in the compartment has been significantly reduced and the dynamic dew point temperature has dropped to the preset safety margin. A recovery command is then sent to the evaporator variable frequency fan to remove the preset lower limit frequency limit of the evaporator variable frequency fan, so that the evaporator variable frequency fan returns to the dynamic speed state controlled by the temperature loop. At the same time, a command is sent to the electronic expansion valve to cancel the manual intervention of the electronic expansion valve opening and set the opening of the electronic expansion valve to the initial opening, thereby adjusting the operating state of the control system to the sensible heat pre-cooling operation state.

[0042] The preset safety constant is used to compensate for sensor measurement errors and the non-uniformity of the spatial temperature field, and the initial opening is the opening set when the electronic expansion valve is started.

[0043] A second aspect of the present invention provides a deep learning-based intelligent control system for agricultural product processing and transportation, the system comprising:

[0044] The feedforward feature acquisition module is used to acquire and generate a stress integral vector based on the mechanical stress data and thermal stress data on the agricultural product processing line during the agricultural product processing stage, and write the stress integral vector into the electronic tag of the loading box after the agricultural product processing is completed.

[0045] The multimodal time series prediction module is used to read the stress integral vector and collect the carriage environment data during the transportation phase, and input the stress integral vector and the carriage environment data into the deep learning prediction model to obtain the predicted respiratory heat flux density sequence.

[0046] The heat peak warning and compensation calculation module is used to calculate the first time derivative of the predicted respiratory heat flux density sequence, generate a trigger signal based on the calculated first time derivative and a pre-configured positive slope threshold, and perform an integral operation on the predicted respiratory heat flux density sequence based on the trigger signal to obtain the cooling compensation equivalent.

[0047] The phase change risk assessment module is used to calculate the dynamic dew point temperature based on the cabin environment data and to calculate the estimated target evaporator pipe wall temperature based on the cooling compensation equivalent. The module determines whether there is a risk of evaporator frosting by comparing the dynamic dew point temperature and the estimated target evaporator pipe wall temperature.

[0048] The timing-decoupled thermodynamic control module is used to adjust the opening of the electronic expansion valve and the speed of the evaporator variable frequency fan upward when there is a risk of frost formation on the evaporator. This reduces the sensible heat ratio in the refrigeration cycle physical process of the control system and adjusts the operating state of the control system to isothermal dehumidification operation. At the same time, it updates the dynamic dew point temperature and determines whether the updated dynamic dew point temperature has dropped to a preset safety margin based on the estimated target evaporator pipe wall temperature and the updated dynamic dew point temperature. When the updated dynamic dew point temperature drops to the preset safety margin, the opening of the electronic expansion valve and the speed of the evaporator variable frequency fan are restored and adjusted to the operating state of the control system to sensible heat precooling operation.

[0049] Preferably, the heat peak early warning and compensation calculation module specifically includes:

[0050] The derivative slope determination unit is used to calculate the first time derivative of the predicted respiratory heat flux density sequence and generate a trigger signal based on the calculated first time derivative and a pre-configured positive slope threshold.

[0051] The heat integration unit is used to, upon receiving the trigger signal, extract and perform time-definite integration on the complete predicted respiratory heat flux density sequence within the future time window, based on the trigger signal, to obtain the total accumulated heat of agricultural products released into the enclosed space of the carriage within the future time window.

[0052] The compensation equivalent conversion unit is used to acquire pre-stored physical environment parameter data, and calculate and obtain the cooling compensation equivalent based on the physical environment parameter data and the total accumulated heat.

[0053] Preferably, the phase transition risk determination module specifically includes:

[0054] The dew point calculation unit is used to extract the current sampling period's dry-bulb temperature sequence data and relative humidity sequence data from the carriage environment data, calculate and determine the intermediate conversion coefficient based on the current sampling period's dry-bulb temperature sequence data and relative humidity sequence data, and calculate the dynamic dew point temperature based on the intermediate conversion coefficient and the current sampling period's dry-bulb temperature sequence data and relative humidity sequence data.

[0055] The pipe wall temperature measurement unit is used to read the target set temperature of the current carriage and determine the estimated target pipe wall temperature of the evaporator based on the target set temperature of the current carriage, the cooling compensation equivalent, and the pre-calibrated inherent heat exchange temperature difference parameters.

[0056] The risk comparison and judgment unit is used to compare the dynamic dew point temperature with the estimated target evaporator wall temperature. When the estimated target evaporator wall temperature is less than or equal to the dynamic dew point temperature, it is determined that there is a risk of frost formation in the evaporator, and the judgment result is output.

[0057] This invention provides a deep learning-based intelligent control method and system for agricultural product processing and transportation. It offers the following advantages:

[0058] 1. This invention generates a stress integral vector by acquiring mechanical and thermal stress data from the agricultural product processing stage, and inputs this vector, along with the vehicle's environmental data, into a deep learning prediction model to obtain a predicted respiratory heat flux density sequence. By introducing the latent physiological stress caused by upstream processing as a feedforward variable into the environmental control during transportation, this invention overcomes the physical lag problem caused by existing cold chain systems relying solely on a single increase in ambient temperature to trigger refrigeration. It achieves early prediction and proactive response to the peak of respiratory heat generation in agricultural products, reducing the risk of spoilage due to localized temperature increases.

[0059] 2. This invention uses the first-order time derivative of the predicted respiratory heat flux density sequence as the intervention trigger condition, and performs integral calculation on the sequence when the condition is met to obtain the cooling compensation equivalent. This can transform the biological heating trend output by the deep learning model into quantitative thermodynamic load parameters, enabling the refrigeration system to accurately allocate the pre-cooling load according to the estimated total heat accumulation. This avoids the energy waste caused by using fixed high-power overload refrigeration and the problem of cold damage to agricultural products that can easily be caused, and improves the accuracy of load matching of the control system.

[0060] 3. This invention determines the risk of frosting by comparing the dynamic dew point temperature and the estimated target pipe wall temperature of the evaporator. When a risk is identified, isothermal dehumidification is performed by jointly adjusting the opening of the electronic expansion valve and the speed of the evaporator variable frequency fan to reduce the sensible heat ratio of the system. Sensible heat precooling is then performed after the dynamic dew point temperature drops to a preset safety margin. This time-decoupled thermodynamic control strategy solves the engineering contradiction that frost formation on the evaporator surface is easily triggered when refrigerated truck compartments perform large temperature difference deep cryogenic compensation, avoiding the increase in heat exchange resistance caused by frosting and ensuring the heat exchange stability and continuity of the refrigeration unit when dealing with high heat loads. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0062] Figure 2 This is a schematic diagram of the system architecture of the present invention;

[0063] Figure 3 This is a schematic diagram of the temperature fluctuation comparison curve of the carriage according to the present invention;

[0064] Figure 4 This is a schematic diagram of the time-series variation curves of the thermodynamic control parameters in the experimental group of this invention. Detailed Implementation

[0065] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] Reference Figure 1 , Figure 1 This is a flowchart of a deep learning-based intelligent control method for agricultural product processing and transportation according to an embodiment of the present invention. The present invention provides a deep learning-based intelligent control method for agricultural product processing and transportation, comprising the following steps:

[0067] S100: During the agricultural product processing stage, acquire and generate a stress integral vector based on the mechanical stress data and thermal stress data on the agricultural product processing line, and write the stress integral vector into the electronic tag of the loading box after the agricultural product processing is completed.

[0068] S200, during the transportation phase, reads the stress integral vector and collects the cabin environment data. The stress integral vector and the cabin environment data are input into a deep learning prediction model to obtain the predicted respiratory heat flux density sequence.

[0069] S300 calculates the first time derivative of the predicted respiratory heat flux density sequence, generates a trigger signal based on the calculated first time derivative and a pre-configured positive slope threshold, and performs an integral operation on the predicted respiratory heat flux density sequence based on the trigger signal to obtain the cooling compensation equivalent.

[0070] S400 calculates the dynamic dew point temperature based on the cabin environment data and the estimated target evaporator pipe wall temperature based on the cooling compensation equivalent. It then determines whether there is a risk of evaporator frosting by comparing the dynamic dew point temperature and the estimated target evaporator pipe wall temperature.

[0071] S500: When there is a risk of frost formation on the evaporator, the opening of the electronic expansion valve and the speed of the evaporator inverter fan are adjusted upwards to reduce the sensible heat ratio in the refrigeration cycle physical process of the control system and adjust the operating state of the control system to isothermal dehumidification operation. At the same time, the dynamic dew point temperature is updated. Based on the estimated target pipe wall temperature of the evaporator and the updated dynamic dew point temperature, it is determined whether the updated dynamic dew point temperature has dropped to the preset safety margin. When the updated dynamic dew point temperature drops to the preset safety margin, the opening of the electronic expansion valve and the speed of the evaporator inverter fan are restored and adjusted to the operating state of the control system to sensible heat pre-cooling operation.

[0072] Regarding the parameter settings of the deep learning prediction model, the multilayer perceptron branch contains three fully connected hidden layers, with the number of neurons in each hidden layer decreasing sequentially. All layers use a linear rectified function as the activation function to extract the nonlinear hidden layer features of the input vector. The temporal convolutional network branch consists of three stacked dilated causal convolutional layers, with the dilation rate increasing exponentially layer by layer. The kernel size is set to 3, and a linear rectified function with leakage is used for activation to capture medium- to long-term dependencies in the environmental time series. No nonlinear activation function is used after the fully connected output layer, and real-valued predictions are directly output.

[0073] Regarding the specific methods for training the deep learning prediction model, a supervised offline training approach was adopted. In the laboratory, a historical sample set covering various typical transportation conditions was collected using a high-precision respiratory calorimeter. Real mechanical and thermal stress data and carriage environment data were used as input features, and the actual heat generation power sequence per unit volume measured by the sensor was used as the real label data.

[0074] During training, mean squared error is used as the loss function to measure the difference between the model's predicted values ​​and the true labels, and an adaptive moment estimator optimizer is used to update the network weights.

[0075] In this embodiment, the control method described above is implemented across the agricultural product processing and cold chain transportation stages. First, mechanical stress and thermal stress data characterizing physical collisions and temperature fluctuations are collected at the agricultural product processing line. Based on this data, a time definite integral operation is performed within a time interval determined by the processing duration parameter. The processed result constitutes a constant stress integral vector, which is then written into the electronic tag of the physical loading box. Based on this, a data correlation is established between the upstream processing status and the downstream transportation environment.

[0076] The control system on the refrigerated transport vehicle then reads the stress integral vector from the electronic tag and continuously acquires real-time, multi-dimensional environmental data of the vehicle's interior. The control system uses both the stress integral vector and the environmental data as input variables for a deep learning prediction model. This deep learning prediction model is used to calculate and output a predicted respiratory heat flux density sequence reflecting the heat release power of agricultural products in future time periods.

[0077] Based on the input stress integral vector and the carriage environment data, the dynamic change slope of the predicted respiratory heat flux density sequence is extracted by performing derivative operations on the deep learning prediction model. When the dynamic change slope meets the preset triggering conditions, it is determined that a sudden change in the heat flux load inside the carriage is about to occur. The accumulated heat value within the future time window is calculated using time definite integral operations, and then the accumulated heat value is converted into the parameter equivalent required for the refrigeration system to perform cooling operations.

[0078] Before actually intervening in the refrigeration hardware, the air dew point temperature is calculated based on the currently collected cabin environment data, and the operating temperature of the evaporator surface is calculated by combining the aforementioned parameters. By comparing the values ​​of the air dew point temperature and the operating temperature of the evaporator surface, it is determined whether the refrigeration operation will trigger phase change frost formation on the evaporator surface.

[0079] If a risk of frosting is confirmed, the sensible heat ratio of the refrigeration cycle is altered by changing the operating frequency of the underlying hardware and the valve opening, prioritizing dehumidification to reduce the moisture content in the air. The system continuously monitors the drop in dew point temperature; once the air conditions meet safety standards, the hardware intervention restrictions are lifted, the set temperature is lowered, and cooling capacity is concentrated.

[0080] Reference Figure 2 , Figure 2 This is an architecture diagram of a deep learning-based intelligent control system for agricultural product processing and transportation according to an embodiment of the present invention. The present invention provides a deep learning-based intelligent control system for agricultural product processing and transportation, which is applied to the deep learning-based intelligent control method for agricultural product processing and transportation described in the above embodiment. This deep learning-based intelligent control system for agricultural product processing and transportation physically spans both fixed agricultural product processing workshops and mobile cold chain transport vehicles, and mainly includes a feedforward feature acquisition module, a multimodal time series prediction module, a heat peak early warning and compensation calculation module, a phase change risk determination module, and a time series decoupling thermodynamic control module.

[0081] in,

[0082] The feedforward feature acquisition module is used to acquire and generate stress integral vectors based on mechanical stress data and thermal stress data on the agricultural product processing line during the agricultural product processing stage, and write the stress integral vectors into the electronic tag of the loading box after the agricultural product processing is completed.

[0083] The multimodal time series prediction module is used to read stress integral vectors and collect cabin environment data during the transportation phase. The stress integral vectors and cabin environment data are input into a deep learning prediction model to obtain the predicted respiratory heat flux density sequence.

[0084] The heat peak warning and compensation calculation module is used to calculate the first time derivative of the predicted respiratory heat flux density sequence, generate a trigger signal based on the calculated first time derivative and the pre-configured positive slope threshold, and perform integral calculation on the predicted respiratory heat flux density sequence based on the trigger signal to obtain the cooling compensation equivalent.

[0085] The phase change risk assessment module is used to calculate the dynamic dew point temperature based on the cabin environment data and to calculate the estimated target evaporator pipe wall temperature based on the cooling compensation equivalent. The module determines whether there is a risk of evaporator frosting by comparing the dynamic dew point temperature and the estimated target evaporator pipe wall temperature.

[0086] The timing-decoupled thermodynamic control module is used to adjust the opening of the electronic expansion valve and the speed of the evaporator inverter fan upward when there is a risk of frost formation on the evaporator. This reduces the sensible heat ratio in the refrigeration cycle of the control system and adjusts the operating state of the control system to isothermal dehumidification. At the same time, it updates the dynamic dew point temperature and determines whether the updated dynamic dew point temperature has dropped to the preset safety margin based on the estimated target evaporator pipe wall temperature and the updated dynamic dew point temperature. When the updated dynamic dew point temperature drops to the preset safety margin, it restores the adjustment of the opening of the electronic expansion valve and the speed of the evaporator inverter fan, adjusting the operating state of the control system to sensible heat precooling.

[0087] The preset safety margin is a compensation parameter that integrates measurement error tolerance, spatial temperature field non-uniformity, and evaporator thermal inertia. Its specific value range is typically set between 0.5℃ and 2℃. The specific calculation method for this safety margin is the sum of three error estimates. The first is the sensor accuracy compensation constant, directly taken from the factory maximum nominal measurement error value of the temperature and humidity sensor inside the carriage. The second is the spatial thermal field offset, determined by pre-deploying a multi-point temperature probe array in an empty carriage to test the airflow distribution and extracting the average maximum temperature difference between the temperature at the return air vent and the temperature at the coldest point in the center of the carriage. The third is the hardware thermal inertia delay constant, representing the dynamic temperature lag required for the evaporator surface temperature to naturally recover after the frequency reduction command is stopped.

[0088] The final preset safety margin value is obtained by adding the sensor accuracy compensation constant, the spatial thermodynamic field offset, and the hardware thermal inertia delay constant.

[0089] In this embodiment, the heat peak early warning and compensation calculation module specifically includes:

[0090] The derivative slope determination unit is used to calculate the first time derivative of the predicted respiratory heat flux density sequence and generate a trigger signal based on the calculated first time derivative and the pre-configured positive slope threshold.

[0091] The heat integration unit is used to extract and perform time-definite integration on the complete predicted respiratory heat flux density sequence within the future time window after receiving the trigger signal, so as to obtain the total heat accumulated by agricultural products in the closed space of the carriage within the future time window.

[0092] The compensation equivalent conversion unit is used to acquire pre-stored physical environment parameter data and calculate and obtain the cooling compensation equivalent based on the physical environment parameter data and the total accumulated heat.

[0093] In this embodiment, the phase transition risk determination module specifically includes:

[0094] The dew point calculation unit is used to extract the current sampling period's dry-bulb temperature sequence data and relative humidity sequence data from the carriage environment data, calculate and determine the intermediate conversion coefficient based on the current sampling period's dry-bulb temperature sequence data and relative humidity sequence data, and calculate the dynamic dew point temperature based on the intermediate conversion coefficient and the current sampling period's dry-bulb temperature sequence data and relative humidity sequence data.

[0095] The pipe wall temperature measurement unit is used to read the target set temperature of the current carriage and determine the estimated target pipe wall temperature of the evaporator based on the target set temperature of the current carriage, the cooling compensation equivalent, and the pre-calibrated inherent heat exchange temperature difference parameters.

[0096] The risk comparison and judgment unit is used to compare the dynamic dew point temperature with the estimated target evaporator wall temperature. When the estimated target evaporator wall temperature is less than or equal to the dynamic dew point temperature, it is determined that there is a risk of frost formation in the evaporator, and the judgment result is output.

[0097] To enable those skilled in the art to better understand the intelligent control system for agricultural product processing and transportation based on deep learning provided by this invention, further supplementary descriptions of the system are now provided.

[0098] The feedforward feature acquisition module is deployed at the end of the agricultural product processing production line. This module includes a data acquisition card integrated into the edge computing node of the production line rack. The data acquisition card is connected via signal cables to temperature sensors and triaxial accelerometers distributed at the water bath and sorting stations. The feedforward feature acquisition module is equipped with a microprocessor that receives analog signals from the sensors, performs analog-to-digital conversion, and executes time definite integral operations to generate a stress integral vector. The feedforward feature acquisition module also includes a radio frequency (RF) communication antenna, through which the microprocessor writes the stress integral vector into an electronic tag in a passive electronic tag register on the surface of the physical loading box.

[0099] The multimodal time-series prediction module is integrated into the onboard main controller of the cold chain transport vehicle. This module includes an RFID reader terminal and an array of environmental sensors distributed throughout the vehicle compartment. After loading, the RFID reader terminal scans the loading box and extracts the stress integral vector. The environmental sensor array uploads real-time environmental data from the vehicle compartment to the onboard main controller via a fieldbus. The multimodal time-series prediction module is internally equipped with an embedded processor featuring a neural network acceleration engine. This embedded processor combines the stress integral vector with the vehicle compartment environmental data into a multidimensional tensor, inputs it into a deep learning prediction model stored in internal flash memory, and outputs a predicted respiratory heat flux density sequence reflecting future changes in heat load.

[0100] The heat peak warning and compensation calculation module resides in the logic operation unit of the vehicle's main controller as software firmware. Specifically, it includes a derivative slope determination unit, a heat integration unit, and a compensation equivalent conversion unit. The derivative slope determination unit acquires the predicted respiratory heat flux density sequence, performs a differential operation to extract the first-order time derivative, and compares it with a positive slope threshold stored in read-only memory for consecutive time steps. When the condition is met, a trigger signal is sent to the system bus. Upon receiving the trigger signal, the heat integration unit extracts and performs a time definite integral operation on the complete predicted respiratory heat flux density sequence within the future time window, obtaining the total accumulated heat released by agricultural products into the enclosed space of the vehicle compartment within the future time window. The compensation equivalent conversion unit queries the control system database through a data interface, extracts the air specific heat capacity, air density, and net volume of the vehicle compartment, and performs a division operation to convert the accumulated heat into a cooling compensation equivalent in the temperature dimension.

[0101] The phase change risk assessment module is also integrated into the vehicle's main controller, interacting with the underlying thermodynamic control system. This module specifically includes a dew point calculation unit, a pipe wall temperature measurement unit, and a risk comparison and assessment unit. The dew point calculation unit extracts the ambient dry-bulb temperature and relative humidity uploaded by the environmental sensor array, and uses the built-in floating-point arithmetic unit to execute an empirical logarithmic formula to calculate the dynamic dew point temperature of the current cabin air. The pipe wall temperature measurement unit reads the inherent heat exchange temperature difference parameter from the system configuration parameters and the current target set temperature of the cabin, combines it with the cooling compensation equivalent to perform calculations, and outputs the estimated target evaporator pipe wall temperature. The risk comparison and assessment unit includes digital comparator logic, receives the dynamic dew point temperature and the estimated target evaporator pipe wall temperature, and outputs a judgment result indicating a risk of evaporator frosting when the estimated pipe wall temperature is lower than or equal to the dynamic dew point temperature.

[0102] The timing-decoupled thermodynamic control module is connected to the underlying actuator of the vehicle-mounted refrigeration unit via an electrical isolation interface. Upon receiving a frosting risk status signal, this module first sends a regulating pulse to the stepper motor drive controller of the electronic expansion valve to limit the refrigerant throttling opening. Simultaneously, it sends a low-frequency operation command to the evaporator inverter to limit the variable frequency fan speed, maintaining isothermal dehumidification operation. The module contains an internal cyclic monitoring timer that periodically triggers the dew point calculation unit to update its values. When logical judgment confirms that the dynamic dew point temperature has dropped to the preset safety margin, the module sends a command to restore normal operation to the evaporator inverter and the electronic expansion valve drive controller via the communication bus. It also lowers the target set temperature in the main control program and outputs a control analog quantity to the compressor inverter, driving the compressor to increase its speed for sensible heat pre-cooling.

[0103] Reference Figures 1-2In this embodiment of the invention, the specific steps for acquiring and generating a stress integral vector based on mechanical stress data and thermal stress data on the agricultural product processing line during the agricultural product processing stage, and writing the stress integral vector into the electronic tag of the loading box after the agricultural product processing is completed are as follows:

[0104] Triaxial accelerometers are deployed at each sorting and transfer node in the agricultural product processing line. These sensors collect mechanical collision acceleration signals in real time at each node. The components of the mechanical collision acceleration signals along the three orthogonal axes in space are extracted, and the composite acceleration is calculated. The root mean square (RMS) calculation is performed on the composite acceleration at each sorting and transfer node according to a preset sampling frequency to extract the effective acceleration values ​​for each node. A sequence of effective acceleration values ​​is generated from all extracted values ​​and used as mechanical stress data to reflect the transient impact intensity experienced by the agricultural products during physical transport.

[0105] Water bath temperature sensors are deployed at the water bath cleaning station of the processing line, and ambient temperature sensors are deployed in the workshop space where the line is located. Real-time water bath temperature is collected by the water bath temperature sensors, and real-time processing ambient temperature is collected by the ambient temperature sensors. The water bath temperature and the processing ambient temperature at the same time point are subtracted, and the absolute value of the result is extracted as the difference between the two. A difference sequence is generated from all the differences between the water bath temperature and the processing ambient temperature acquired at various time points during the agricultural product processing stage. This difference sequence constitutes thermal stress data, used to quantify the physical quantity of thermal shock experienced by agricultural products when crossing different temperature zones.

[0106] Record the start time of the current batch of agricultural products entering the processing line and the end time of completion of packing and removal from the line. Extract the time span from the start time to the end time as the processing duration parameter. Based on the stress integral vector calculation formula, perform time definite integral operations on the previously calculated effective acceleration value sequence and difference sequence within the time interval determined by the processing duration parameter. The stress integral vector calculation formula is as follows:

[0107] ;

[0108] In the formula, It is the stress integral vector; This is a parameter for processing duration; The effective acceleration values ​​at the same timestamp; The washing water bath temperature at the same timestamp; The processing environment temperature at the same timestamp; Representing time variables The differential element.

[0109] In the specific calculation process using the stress integral vector calculation formula, the results of the two definite integrals in the mechanical and thermal dimensions are concatenated to generate a fixed-length two-dimensional constant vector. This two-dimensional constant vector is the stress integral vector. Its value represents the total latent physiological stress accumulated during the processing of this batch of agricultural products.

[0110] Generating a two-dimensional constant vector does not involve directly adding the mechanical and thermal integral values, which have different dimensions, to scalar values. Instead, it involves preprocessing the dimensional feature matrix used as input to the deep learning model. Specifically, after calculating the definite integrals of the mechanical and thermal dimensions, standard reference constants for each dimension are introduced to perform max-min normalization on the two integral results. This eliminates the differences in physical units and dimensions in the original data and maps them to the dimensionless standard interval between 0 and 1.

[0111] The two normalized dimensionless eigenvalues ​​are concatenated at the data structure level to form a stress integral vector in the form of an array containing two elements. This stress integral vector is essentially a feature expression representing the intensity of the combined environmental stimuli during the processing stage. Since it eliminates the dimension barrier, when it is subsequently input into the multilayer perceptron, the hidden layer weight matrix of the deep neural network can adaptively learn and mine the cross-coupling weights between the two different stressors. Then, the neural network calculates the overall latent physiological stress level accumulated in the latent space of the batch of agricultural products.

[0112] After agricultural product processing is completed, the stress integral vector is transmitted through an industrial control network. The data is transmitted to an RF writing device located at the end of the packing station. The RF writing device uses electromagnetic induction to convert the stress integral vector... The electronic tags are written into the corresponding loading boxes of agricultural products, and these tags are stored in chips. Once written, the electronic tags are transported along with the loading boxes to the transport vehicles, establishing a physical data carrier between the processing and transportation ends.

[0113] Reference Figures 1-2 In this embodiment of the invention, during the transportation phase, the stress integral vector is read and the carriage environment data is collected. The stress integral vector and the carriage environment data are then input into a deep learning prediction model to obtain the predicted respiratory heat flux density sequence. The specific steps are as follows:

[0114] After the transport vehicle finishes loading agricultural products, the control system obtains the two-dimensional constant vector, i.e., the stress integral vector, stored in the electronic tag inside the loading box through a radio frequency identification (RFID) reader. At the same time, the control system activates vibration sensors and integrated temperature and humidity sensors distributed inside the refrigerated compartment.

[0115] Using a vibration sensor and an integrated temperature and humidity sensor, environmental data of the refrigerated truck compartment is collected within a preset time window according to a set sampling period. This environmental data includes triaxial vibration acceleration sequence data, ambient dry-bulb temperature sequence data, and relative humidity sequence data. The preset time window length is [not specified]. The data buffer queue is used, and the stress integral vector of a single value is constantly padded and expanded in the time dimension to ensure that the expanded stress integral vector is aligned with the time steps within a preset time window. The expanded stress integral vector is then merged and concatenated with the triaxial vibration acceleration sequence data, environmental dry-bulb temperature sequence data, and relative humidity sequence data in the feature channel dimension to construct a unified model input matrix. The specific formula for constructing the model input matrix is ​​as follows:

[0116] ;

[0117] In the formula, Indicates the current time The model input matrix; Represents the stress integral vector; Indicates the length of the time window Triaxial vibration acceleration sequence data within; Indicates the length of the time window Indoor dry-bulb temperature sequence data; Indicates the length of the time window The relative humidity sequence data within.

[0118] Input the completed model into the matrix The input is fed into a pre-trained deep learning prediction model deployed in the control system. This deep learning prediction model has a cascaded architecture of multilayer perceptrons and temporal convolutional networks. The feature processing path within the deep learning prediction model is divided into static and dynamic branches. The stress integral vector in the model input matrix is ​​received through the multilayer perceptron, and a nonlinear mapping is performed to output a high-dimensional hidden layer feedforward, which represents the physiological damage baseline of agricultural products affected by upstream processing.

[0119] Temporal convolutional networks receive environmental dry-bulb temperature sequence data from the model input matrix through dilated causal convolutional layers. They utilize dilated convolutions to expand the receptive field and extract long-period dependency features from the dry-bulb temperature sequence data. The deep learning prediction model then performs tensor concatenation and fully connected computation on the high-dimensional hidden layer feedforward and long-period dependency features at the fusion layer to obtain the fused feature vector.

[0120] The fused feature vectors are processed by a linear output layer to calculate the predicted heat generation power per unit volume at different times within the future prediction time window. Based on the predicted heat generation power per unit volume at all times within the future prediction time window, a predicted respiratory heat flux density sequence is generated. This predicted respiratory heat flux density sequence reflects the dynamic trend of heat flux density released from agricultural products into the vehicle compartment due to metabolism within the preset future time window. The expression for the predicted respiratory heat flux density sequence is as follows:

[0121] ;

[0122] In the formula, Indicates a future moment The predicted respiratory heat flux density value; Indicates the current moment; Indicates the length of the future forecast time window; This represents an independent time variable within a future forecast time window; The set represents a symbol, which restricts independent time variables. The specific value must be within the set time interval. Within the range.

[0123] Reference Figures 1-2 The specific implementation process of calculating the first-order time derivative of the predicted respiratory heat flux density sequence and performing integral calculation to obtain the cooling compensation equivalent when the triggering condition is met is as follows in this embodiment of the invention.

[0124] The control system receives the predicted respiratory heat flux density sequence output by the deep learning prediction model. The control system performs finite difference operations on the predicted heat power per unit volume corresponding to data points at adjacent time steps in the predicted respiratory heat flux density sequence to obtain the first-order time derivative of the predicted respiratory heat flux density sequence in the time dimension. This first-order time derivative reflects the rate of change of the heat power released by agricultural products in the future.

[0125] The calculated first-order time derivative is extracted within each calculation cycle and compared numerically with a pre-configured positive slope threshold. The pre-configured positive slope threshold is the upper limit of the normal physiological heat release rate fluctuation set by the control system.

[0126] The pre-set positive slope threshold was determined through standard baseline testing and prior calibration experiments before loading. The samples of healthy agricultural products of the same type to be tested, which were in a state of no mechanical damage and heat shock, were placed in a standard constant temperature and humidity breathing chamber, and the heat release power sequence generated by their natural basal metabolism at different reference ambient temperatures was continuously recorded using a microcalorimeter.

[0127] The first time derivative of the obtained stationary heat release power time series is calculated to extract the maximum absolute value of the change in heat release rate of this type of agricultural product under normal physiological rhythm. In order to cope with measurement noise and individual differences, the calculated maximum absolute value of the change in heat release rate is multiplied by an empirical tolerance coefficient (which is usually between 1.2 and 1.5). The final product is set as the positive slope threshold corresponding to this type of agricultural product.

[0128] A sliding observation window is set to record the historical derivative status. When the calculated first-order time derivative is greater than a pre-configured positive slope threshold for a consecutive preset number of time steps, it is determined that the physiological metabolic latency period of the agricultural product has ended and it is about to enter the respiratory heat surge stage. At this time, a trigger signal is generated to start the subsequent thermodynamic intervention program. The judgment logic of the preset trigger condition is as follows:

[0129] when At that time, a trigger signal is generated;

[0130] In the formula, Indicates a future moment The predicted respiratory heat flux density value; This represents a pre-configured positive slope threshold; Representing time variables The differential element; Represent the differential operator symbol. The inequality must be valid in continuous equations. The condition is always true within a time step to filter out high-frequency noise interference, where n is an integer greater than 1.

[0131] Upon receiving the trigger signal, a complete predicted respiratory heat flux density sequence within a future time window is extracted. Using the current warning time as the lower limit and the end time of the future time window as the upper limit, a time definite integral is performed on the extracted complete predicted respiratory heat flux density sequence within the future time window to obtain the total accumulated heat. This time definite integral of the total accumulated heat represents the total heat released by the agricultural product population into the enclosed space of the vehicle within a preset future time period. This total accumulated heat is used as the basic parameter for matching the cooling load.

[0132] The system retrieves physical environment parameter data stored in the control system database, specifically including air specific heat capacity, air density, and the net volume of the current transport vehicle's cargo compartment. Based on the calculation formula for the cooling compensation equivalent, the control system uses the previously calculated total accumulated heat as the dividend and the product of the air specific heat capacity, air density, and cargo compartment net volume as the divisor for division. This calculation process converts the heat value in the energy dimension into a control parameter in the temperature dimension. The control system uses the result of this division as the cooling compensation equivalent. The cooling compensation equivalent provides the parameter basis for the refrigeration unit to determine the precise temperature offset command required to eliminate the estimated heat load. The calculation formula for the cooling compensation equivalent is as follows:

[0133] ;

[0134] In the formula, Indicates the cooling compensation equivalent; Indicates a future moment The predicted respiratory heat flux density value; Indicates the current moment; Indicates the length of the time window for future forecasting; For the current moment To the future Total accumulated heat; This represents the specific heat capacity of air. This represents the numerical value of air density; This indicates the net volume of the carriage. Representing time variables The differential element. (Refer to...) Figures 1-2 In this embodiment of the invention, the dynamic dew point temperature is calculated based on the cabin environment data, and the estimated target evaporator pipe wall temperature is calculated in combination with the cooling compensation equivalent. The specific steps for determining the risk of evaporator frosting by comparing the dynamic dew point temperature and the estimated target evaporator pipe wall temperature are as follows:

[0135] After calculating the cooling compensation equivalent, the control system simultaneously extracts the ambient dry-bulb temperature sequence and relative humidity sequence data for the current sampling period from multi-dimensional cabin environmental data. These ambient dry-bulb temperature sequence and relative humidity sequence data constitute the real-time state coordinates of the humid air inside the cabin. The control system then substitutes the extracted ambient dry-bulb temperature sequence and relative humidity sequence data for the current sampling period into a built-in empirical logarithmic formula calculation matrix.

[0136] The dynamic dew point temperature of the cabin air is calculated using a formula based on the dynamic dew point temperature. First, an intermediate conversion coefficient is calculated based on the ambient dry-bulb temperature and relative humidity. Then, based on this intermediate conversion coefficient, the critical temperature point at which water vapor in the air reaches saturation, i.e., the dynamic dew point temperature, is derived. The dynamic dew point temperature reflects the temperature threshold for a gas-liquid phase transition under the current humidity conditions of the cabin air. The formula for calculating the dynamic dew point temperature is as follows:

[0137] ;

[0138] In the formula, ; Indicates the ambient dry-bulb temperature; Indicates relative humidity; Indicates intermediate conversion coefficients; Represents the dynamic dew point temperature; the constant term is the empirical fitting coefficient. The operator represents the natural logarithm operator.

[0139] While solving for the dynamic dew point temperature, the inherent heat transfer temperature difference parameter, pre-calibrated by the control system, is obtained. This inherent heat transfer temperature difference parameter represents the fixed physical temperature gradient required for the refrigeration unit to drive heat transfer from the air side to the refrigerant side within the evaporator tubes at rated airflow. The target set temperature of the current compartment is read; this target set temperature is the baseline maintenance temperature specified for cold chain transportation tasks.

[0140] The operating boundary of the underlying refrigeration hardware is calculated based on the formula for estimating the target evaporator wall temperature. The theoretical air temperature required to offset the heat load is obtained by subtracting the cooling compensation equivalent calculated in the previous steps from the current target set temperature of the passenger compartment. Further subtracting the inherent heat exchange temperature difference parameter from this result, the final calculation result is used as the estimated target evaporator wall temperature. This parameter represents the actual physical temperature that the evaporator's metal outer surface will reach if the cooling compensation command is directly executed. The formula for estimating the target evaporator wall temperature is as follows:

[0141] ;

[0142] In the formula, This indicates the estimated target tube wall temperature of the evaporator; This indicates the target set temperature for the current carriage; Indicates the cooling compensation equivalent; This represents the inherent heat exchange temperature difference parameter.

[0143] The dynamic dew point temperature is compared with the estimated target evaporator wall temperature. When the estimated target evaporator wall temperature is less than or equal to the dynamic dew point temperature, it is determined that the operating temperature of the evaporator surface is below the critical point of air phase change, and condensation will occur on the surface, condensing into frost under cryogenic conditions. At this time, the determination result is output, confirming the risk of evaporator frosting under the current operating condition, and triggering the subsequent timing decoupling control sequence.

[0144] Reference Figures 1-2 In this embodiment of the invention, when there is a risk of frost formation on the evaporator, the opening of the electronic expansion valve and the speed of the evaporator inverter fan are adjusted upwards to reduce the sensible heat ratio in the refrigeration cycle physical process of the control system and adjust the operating state of the control system to isothermal dehumidification operation. Simultaneously, the dynamic dew point temperature is updated. Based on the estimated target evaporator pipe wall temperature and the updated dynamic dew point temperature, it is determined whether the updated dynamic dew point temperature has dropped to a preset safety margin. When the updated dynamic dew point temperature drops to the preset safety margin, the opening of the electronic expansion valve and the speed of the evaporator inverter fan are restored to normal operation. The steps to adjust the operating state of the control system to sensible heat pre-cooling operation are as follows:

[0145] When the control system detects a risk of evaporator frosting, it locks the target set temperature for the current compartment in the controller's internal memory. It then pauses the response to conventional proportional-integral-derivative (PID) control logic based on temperature deviation, preventing the compressor from directly increasing its operating frequency.

[0146] A valve reduction command is sent via bus to the stepper motor driver of the electronic expansion valve, decreasing the physical opening of the electronic expansion valve and reducing the refrigerant flow into the evaporator. This adjustment of the electronic expansion valve's opening limits the actual evaporator wall temperature within a specific physical range. The upper limit of this range is the dynamic dew point temperature calculated in the previous steps, and the lower limit is the freezing point of pure water at standard atmospheric pressure. The logical relationship for limiting the evaporator wall temperature within this specific physical range is as follows:

[0147] ;

[0148] In the formula, This indicates the freezing point temperature of pure water under standard atmospheric pressure. This indicates the temperature of the evaporator tube wall; This indicates the dynamic dew point temperature. Setting the dynamic dew point temperature forces the humid air flowing through the evaporator to condense and release moisture, forming condensate, while preventing the condensate from undergoing a solid-phase change on the tube wall surface.

[0149] While adjusting the electronic expansion valve, a frequency reduction command is simultaneously sent to the evaporator inverter fan. The operating speed of the evaporator inverter fan is forcibly reduced to a preset lower frequency limit. The decrease in fan speed reduces the volumetric airflow across the evaporator fins, increasing the contact time between the air and the pipe wall. Through the combined adjustment of the electronic expansion valve opening and the evaporator inverter fan speed, the total cooling capacity distribution ratio of the refrigeration unit is altered, reducing the sensible heat ratio in the refrigeration cycle, thereby adjusting the control system to an isothermal dehumidification operation state.

[0150] The formula for calculating the sensible heat ratio of the control system is: ;

[0151] In the formula, This represents the sensible heat ratio of the control system during the refrigeration cycle physical process; This represents the sensible heat and cold energy output by the control system. This represents the latent heat output by the control system.

[0152] During the isothermal dehumidification process, i.e., while the control system is in isothermal dehumidification operation mode, the environmental monitoring program is cyclically invoked. The control system continuously collects ambient dry-bulb temperature and relative humidity data inside the carriage at fixed time steps, and synchronously updates the calculated value of the dynamic dew point temperature. The control system retrieves the initially calculated estimated target evaporator wall temperature from the memory register and compares it with the difference between each updated dynamic dew point temperature.

[0153] The control system determines whether the estimated target evaporator wall temperature is greater than the sum of the updated dynamic dew point temperature and the preset safety constant. The preset safety constant is used to compensate for sensor measurement errors and the non-uniformity of the spatial temperature field. The expression for the estimated target evaporator wall temperature being greater than the sum of the updated dynamic dew point temperature and the preset safety constant is as follows:

[0154] ;

[0155] In the formula, This indicates the estimated target tube wall temperature of the evaporator; This indicates the updated dynamic dew point temperature; This represents the preset safety constant.

[0156] when At that time, the control system determined that the absolute humidity of the air inside the carriage had decreased significantly and the dynamic dew point temperature had dropped to the preset safety margin.

[0157] The control system sends a recovery command to the evaporator inverter fan, releasing the preset lower limit frequency limit and restoring the fan to its dynamic speed controlled by the temperature loop. Simultaneously, the control system sends a command to the electronic expansion valve, canceling manual intervention in its opening and setting it to its initial opening. This adjusts the control system's operation to sensible heat pre-cooling mode. A preset safety constant is used to compensate for sensor measurement errors and the non-uniformity of the ambient temperature field; the initial opening is the opening set when the electronic expansion valve is started.

[0158] The control system subtracts the previously calculated cooling compensation equivalent from the locked target set temperature in the main controller to obtain a new temperature value. This new temperature value is then used to overwrite the original target set temperature, resulting in an updated target set temperature. Based on the updated target set temperature, the control system calculates the new temperature deviation and drives the compressor inverter to increase its operating frequency. Through this timing decoupling operation, the control system performs sensible heat precooling while eliminating frost, thus offsetting the impending heat surge load from the agricultural products.

[0159] To enable those skilled in the art to better understand the above-mentioned technical solutions provided by the present invention, further supplementary explanations are given below using examples: Specific application embodiments:

[0160] During the agricultural product processing stage, the peaches undergo a washing and sorting production line. The control system collected an effective value of 1.5 m / s² for the synthetic mechanical collision acceleration of this batch of peaches. 2 The cleaning water bath temperature is 10℃, and the processing ambient temperature is 25℃, with an absolute temperature difference of 15℃. The processing duration is 45 minutes. The control system performs a definite time integral on the above data, generates a stress integral vector, and writes it into the electronic tag of the loading box.

[0161] During the cold chain transportation phase, the net volume of the refrigerated vehicle compartment is 30 cubic meters, and the air density is taken as 1.2 kg / m³. 3 The specific heat capacity of air is taken as 1005 J / (kg·℃). The initial target temperature is set at 4.0℃. The control system reads the stress integral vector and continuously collects real-time vibration data of the carriage, ambient dry-bulb temperature (currently 5.0℃), and relative humidity (currently 90%). These parameters are input into the deep learning prediction model. The model outputs a predicted respiratory heat flux density sequence, which indicates that the cargo will experience a respiratory heat surge within the next 2 hours.

[0162] The control system calculates the first-order time derivative of the predicted respiratory heat flux density sequence and determines whether it exceeds the positive slope threshold of 2.0 W / (m²) within five consecutive time steps. 3The system calculates the total accumulated heat by performing a time-definite integral operation on the complete predicted respiratory heat flux density sequence within the captured future time window (in minutes), thereby triggering a heat peak warning. Then, using the formula for calculating the cooling compensation equivalent, the system divides the calculated total accumulated heat as the divisor by the product of the air specific heat capacity, air density, and the net volume of the carriage, resulting in a cooling compensation equivalent of 3.5℃.

[0163] The control system calculates the dynamic dew point temperature under the current ambient dry-bulb temperature of 5.0℃ and relative humidity of 90%, resulting in a value of 3.4℃. Combining the cooling compensation equivalent of 3.5℃ and the control system's inherent heat transfer temperature difference parameter of 5.0℃, the control system calculates the predicted target evaporator wall temperature as 4.0℃ minus 3.5℃ and then minus 5.0℃, resulting in -4.5℃. By comparison, the control system determines that the predicted target evaporator wall temperature (-4.5℃) is lower than the dynamic dew point temperature (3.4℃), confirming a risk of evaporator frosting.

[0164] The control system locks the target set temperature at 4.0℃ in the controller, reduces the opening of the electronic expansion valve, and decreases the speed of the variable frequency fan. The control system maintains the actual evaporator wall temperature at 1.0℃, which is 3.4℃ below the dew point and above the freezing point of pure water (0℃), thus performing isothermal dehumidification. After 40 minutes of operation, the control system detects a significant decrease in the relative humidity of the compartment, and the dynamic dew point temperature is updated to -6.0℃. The control system determines that the estimated target evaporator wall temperature (-4.5℃) is greater than the sum of the updated dynamic dew point temperature (-6.0℃) and the preset safety constant (1.0℃). The control system then releases the restrictions on the fan and electronic expansion valve, lowers the target set temperature by 0.5℃ (4.0℃ minus 3.5℃), and drives the compressor to perform sensible heat pre-cooling, thereby offsetting the surge in actual respiratory heat generated by the peaches.

[0165] Reference Figure 3 and Figure 4 , Figure 3 and Figure 4 These are all schematic diagrams comparing the experimental and control group's operational data according to an embodiment of the present invention. Figure 3 This diagram illustrates the temperature fluctuations within the train carriage. The horizontal axis represents the transportation time parameter, and the vertical axis represents the actual temperature parameter within the carriage. Solid lines in the diagram represent the temperature change curves of the experimental group, dashed lines represent the temperature change curves of the control group, and dotted lines represent the initially set target maintenance temperature.

[0166] Figure 4This is a schematic diagram of the time-series variation curves of the thermodynamic control parameters for the experimental group of this invention. The horizontal axis represents the transportation time parameter, and the vertical axis represents the temperature value of the control parameter. In the figure, the dotted line represents the dynamic dew point temperature of the air in the carriage, the short dashed line represents the estimated target evaporator tube wall temperature calculated by the control system, and the solid line represents the evaporator tube wall temperature actually controlled by the control system.

[0167] To verify the practical effectiveness of the technical solution of this invention, a 48-hour comparative transportation experiment was conducted using two identical refrigerated transport vehicles loaded with the same batch of processed peaches. The control group vehicle was equipped with a traditional proportional-integral-derivative control system based on temperature deviation, while the experimental group vehicle was equipped with the intelligent control system of this invention. Test indicators included the maximum temperature fluctuation in the cargo compartment, evaporator frosting, defrosting energy consumption ratio, and final spoilage rate of the goods.

[0168] like Figure 3 and Figure 4 The operational data records showed that the control group responded slowly when the goods experienced a surge in respiratory heat, and the refrigeration unit failed to allocate cooling capacity in advance, causing the actual highest temperature inside the compartment to climb to 8.5℃, with a maximum temperature fluctuation of 4.5℃. Because the control group directly applied full-load cooling under high temperature differences, severe solid-phase frost formed on the evaporator surface, triggering the electric defrosting procedure three times throughout the transportation process. Defrosting energy consumption accounted for 18% of the total energy consumption of the control system, and the final tested goods spoilage rate was 7.5%.

[0169] The experimental group acquired front-end stress data and performed multimodal time-series prediction to pre-calculate the cooling compensation equivalent. Time-series decoupling intervention was implemented, prioritizing dehumidification followed by pre-cooling. The maximum temperature fluctuation in the experimental group's cargo compartment was controlled within 1.2℃. Due to the implementation of the isothermal dehumidification strategy, no frost formation occurred on the evaporator throughout the entire transportation process, the defrosting energy consumption was reduced to 0%, and the final cargo spoilage rate was only 1.2%. Comparative data shows that the method of this invention has engineering effectiveness in improving the accuracy of cargo compartment temperature control, eliminating frost formation on heat exchangers, and reducing agricultural product spoilage.

Claims

1. A deep learning-based intelligent control method for agricultural product processing and transportation, characterized in that, The method includes the following steps: S100, during the agricultural product processing stage, acquire and generate a stress integral vector based on the mechanical stress data and thermal stress data on the agricultural product processing line, and write the stress integral vector into the electronic tag of the loading box after the agricultural product processing is completed; S200, during the transportation phase, the stress integral vector is read and the carriage environment data is collected. The stress integral vector and the carriage environment data are input into a deep learning prediction model to obtain a predicted respiratory heat flux density sequence. S300, calculate the first time derivative of the predicted respiratory heat flux density sequence, generate a trigger signal based on the calculated first time derivative and a pre-configured positive slope threshold, and perform an integral operation on the predicted respiratory heat flux density sequence based on the trigger signal to obtain the cooling compensation equivalent. S400: The dynamic dew point temperature is calculated based on the cabin environment data, and the estimated target evaporator pipe wall temperature is calculated based on the cooling compensation equivalent. The dynamic dew point temperature and the estimated target evaporator pipe wall temperature are compared to determine whether there is a risk of evaporator frosting. S500: When there is a risk of frost formation on the evaporator, the opening of the electronic expansion valve and the speed of the evaporator inverter fan are adjusted upwards to reduce the sensible heat ratio in the refrigeration cycle physical process of the control system and adjust the operating state of the control system to isothermal dehumidification operation. At the same time, the dynamic dew point temperature is updated. Based on the estimated target pipe wall temperature of the evaporator and the updated dynamic dew point temperature, it is determined whether the updated dynamic dew point temperature has dropped to the preset safety margin. When the updated dynamic dew point temperature drops to the preset safety margin, the opening of the electronic expansion valve and the speed of the evaporator inverter fan are restored and adjusted to the operating state of the control system to sensible heat pre-cooling operation. The steps of S300 further include: The predicted heat generation power per unit volume corresponding to data points at adjacent time steps in the predicted respiratory heat flux density sequence is obtained and subjected to finite difference operation to obtain the first time derivative of the predicted respiratory heat flux density sequence in the time dimension. The obtained first-order time derivative is compared with the pre-configured positive slope threshold. When the obtained first-order time derivative is greater than the pre-configured positive slope threshold within a consecutive preset number of time steps, it is determined that the physiological metabolic latency period of agricultural products has ended and is about to enter the respiratory heat surge stage. A trigger signal is generated and sent. The pre-configured positive slope threshold is the upper limit of the normal physiological heat release rate fluctuation set by the control system. Based on the trigger signal, the complete predicted respiratory heat flux density sequence within the future time window is extracted and subjected to time definite integral operation to obtain the total accumulated heat of agricultural products released into the enclosed space of the vehicle within the future time window; the pre-stored physical environment parameter data is obtained, and the cooling compensation equivalent is calculated based on the physical environment parameter data and the total accumulated heat. The steps in S400 include determining the dynamic dew point temperature and determining the estimated target evaporator wall temperature. The step of determining the dynamic dew point temperature is as follows: Extract the current sampling period's dry-bulb temperature sequence data and relative humidity sequence data from the carriage environment data, and calculate and determine the intermediate conversion coefficients based on the current sampling period's dry-bulb temperature sequence data and relative humidity sequence data. The dynamic dew point temperature is calculated based on the intermediate conversion coefficient, as well as the ambient dry-bulb temperature sequence data and relative humidity sequence data of the current sampling period. The dynamic dew point temperature represents the temperature threshold at which a gas-liquid phase change occurs under the current air humidity conditions in the carriage. The steps for determining the estimated target tube wall temperature of the evaporator are as follows: The target set temperature of the current carriage is read, and the estimated target evaporator tube wall temperature is determined based on the target set temperature of the current carriage, the cooling compensation equivalent, and the pre-calibrated inherent heat exchange temperature difference parameters. The estimated target evaporator tube wall temperature... In the formula, This represents the target set temperature of the current carriage. Characterizing the cooling compensation equivalent, Characterizes the pre-calibrated inherent heat transfer temperature difference parameter; The dynamic dew point temperature is compared with the estimated target evaporator wall temperature. When the estimated target evaporator wall temperature is less than or equal to the dynamic dew point temperature, it is determined that the evaporator is at risk of frosting, and the determination result is output.

2. The intelligent control method for agricultural product processing and transportation based on deep learning according to claim 1, characterized in that, Step S100, which involves acquiring and generating a stress integral vector based on the mechanical stress data and thermal stress data on the agricultural product processing line, and then writing the stress integral vector into the electronic tag of the loading box after the agricultural product processing is completed, includes the following steps: Based on a triaxial accelerometer, mechanical collision acceleration signals of each sorting node and each transfer node in the agricultural product processing line are collected. Based on the mechanical collision acceleration signals of each sorting node and each transfer node, the effective acceleration values ​​of each sorting node and each transfer node are extracted. The extracted effective acceleration values ​​are used to generate an effective acceleration value sequence, and the effective acceleration value sequence is used as the mechanical stress data. The temperature of the cleaning water bath and the processing environment are collected in real time by temperature sensors. The difference between the cleaning water bath temperature and the processing environment temperature at different time stamps is obtained and a difference sequence is generated. The difference sequence is used as the thermal stress data. The temperature sensors include a water bath temperature sensor and an ambient temperature sensor. The processing duration parameter is obtained based on the start time of agricultural products entering the processing line and the end time of packaging and leaving the line. Based on the stress integral vector calculation formula, the mechanical stress data and the thermal stress data are integrated over time within the processing duration parameter to obtain the stress integral vector. The stress integral vector calculation formula is as follows: ; In the formula, It is the stress integral vector; This is a parameter for processing duration; The effective acceleration values ​​at the same timestamp; The washing water bath temperature at the same timestamp; The processing environment temperature at the same timestamp; Representing time variables The differential element; After agricultural product processing is completed, the stress integral vector is transmitted via an industrial control network and radio frequency writing equipment. Write it into the electronic tag of the corresponding loading box for agricultural products.

3. The intelligent control method for agricultural product processing and transportation based on deep learning according to claim 1, characterized in that, In step S200, the steps of reading the stress integral vector and collecting the carriage environment data, and inputting the stress integral vector and the carriage environment data into the deep learning prediction model to obtain the predicted respiratory heat flux density sequence include: The stress integral vector in the electronic tag of the loading box is read, and the environmental data of the car body within a preset time window is collected. The environmental data of the car body includes triaxial vibration acceleration sequence data, environmental dry-bulb temperature sequence data and relative humidity sequence data. The stress integral vector is merged and spliced ​​with the triaxial vibration acceleration sequence data, environmental dry-bulb temperature sequence data and relative humidity sequence data within a preset time window to generate the model input matrix; The model input matrix is ​​input into a deep learning prediction model, wherein the deep learning prediction model includes a cascaded architecture of a multilayer perceptron and a temporal convolutional network. The deep learning prediction model uses a multilayer perceptron to receive the stress integral vector in the model input matrix and performs nonlinear mapping to obtain a high-dimensional hidden layer feedforward. The deep learning prediction model also uses a temporal convolutional network to receive the environmental dry-bulb temperature sequence data in the model input matrix and extract long-period dependency features from the environmental dry-bulb temperature sequence data. Based on the deep learning prediction model, the high-dimensional hidden layer feedforward quantity and long-period dependent features are tensor concatenated and fully connected to obtain a fused feature vector. The fused feature vector is then processed through the linear output layer of the deep learning prediction model to calculate and obtain the predicted value of heat generation per unit volume at different times within the future prediction time window. Based on the predicted value of heat generation per unit volume at all times within the future prediction time window, a predicted respiratory heat flux density sequence is generated.

4. The intelligent control method for agricultural product processing and transportation based on deep learning according to claim 1, characterized in that, In step S500, when there is a risk of frost formation on the evaporator, the steps of adjusting the opening of the electronic expansion valve and the speed of the evaporator inverter fan upwards to reduce the sensible heat ratio in the refrigeration cycle physical process of the control system and adjusting the operating state of the control system to isothermal dehumidification operation further include: When the output judgment result indicates that there is a risk of frost formation on the evaporator, the target set temperature of the current compartment is locked, and a valve reduction command is sent to the stepper motor driver of the electronic expansion valve via the bus to reduce the physical opening of the electronic expansion valve, thereby reducing the refrigerant flow into the evaporator. The actual evaporator pipe wall temperature is limited to a specific physical range by adjusting the opening of the electronic expansion valve. The upper limit of the physical range is the dynamic dew point temperature calculated in the previous steps, and the lower limit is the freezing point temperature of pure water under standard atmospheric pressure. While adjusting the electronic expansion valve, a frequency reduction command is simultaneously sent to the evaporator variable frequency fan to force the working speed of the evaporator variable frequency fan to be reduced to the preset lower limit frequency value. By jointly adjusting the opening of the electronic expansion valve and the speed of the evaporator variable frequency fan, the operating state of the control system is adjusted to an isothermal dehumidification operating state.

5. The intelligent control method for agricultural product processing and transportation based on deep learning according to claim 1, characterized in that, In step S500, the dynamic dew point temperature is updated simultaneously. Based on the estimated target evaporator wall temperature and the updated dynamic dew point temperature, it is determined whether the updated dynamic dew point temperature has dropped to a preset safety margin. When the updated dynamic dew point temperature drops to the preset safety margin, the opening of the electronic expansion valve and the speed of the evaporator variable frequency fan are restored and adjusted to adjust the operating state of the control system to the sensible heat pre-cooling operating state. The specific steps include: When the control system is in isothermal dehumidification operation mode, it continuously collects the ambient dry-bulb temperature sequence data and relative humidity sequence data inside the carriage at fixed time steps, and updates the dynamic dew point temperature synchronously based on the collected ambient dry-bulb temperature sequence data and relative humidity sequence data. The updated dynamic dew point temperature is compared with the estimated target evaporator pipe wall temperature. When the estimated target evaporator pipe wall temperature is greater than the sum of the updated dynamic dew point temperature and the preset safety constant, it is determined that the absolute moisture content of the air in the compartment has been significantly reduced and the dynamic dew point temperature has dropped to the preset safety margin. A recovery command is then sent to the evaporator variable frequency fan to remove the preset lower limit frequency limit of the evaporator variable frequency fan, so that the evaporator variable frequency fan returns to the dynamic speed state controlled by the temperature loop. At the same time, a command is sent to the electronic expansion valve to cancel the manual intervention of the electronic expansion valve opening and set the opening of the electronic expansion valve to the initial opening, thereby adjusting the operating state of the control system to the sensible heat pre-cooling operation state. The preset safety constant is used to compensate for sensor measurement errors and the non-uniformity of the spatial temperature field, and the initial opening is the opening set when the electronic expansion valve is started.

6. A deep learning-based intelligent control system for agricultural product processing and transportation, applied to the deep learning-based intelligent control method for agricultural product processing and transportation as described in any one of claims 1-5, characterized in that, The system includes: The feedforward feature acquisition module is used to acquire and generate a stress integral vector based on the mechanical stress data and thermal stress data on the agricultural product processing line during the agricultural product processing stage, and write the stress integral vector into the electronic tag of the loading box after the agricultural product processing is completed. The multimodal time series prediction module is used to read the stress integral vector and collect the carriage environment data during the transportation phase, and input the stress integral vector and the carriage environment data into the deep learning prediction model to obtain the predicted respiratory heat flux density sequence. The heat peak warning and compensation calculation module is used to calculate the first time derivative of the predicted respiratory heat flux density sequence, generate a trigger signal based on the calculated first time derivative and a pre-configured positive slope threshold, and perform an integral operation on the predicted respiratory heat flux density sequence based on the trigger signal to obtain the cooling compensation equivalent. The phase change risk assessment module is used to calculate the dynamic dew point temperature based on the cabin environment data and to calculate the estimated target evaporator pipe wall temperature based on the cooling compensation equivalent. The module determines whether there is a risk of evaporator frosting by comparing the dynamic dew point temperature and the estimated target evaporator pipe wall temperature. The timing-decoupled thermodynamic control module is used to adjust the opening of the electronic expansion valve and the speed of the evaporator variable frequency fan upward when there is a risk of frost formation on the evaporator. This reduces the sensible heat ratio in the refrigeration cycle physical process of the control system and adjusts the operating state of the control system to isothermal dehumidification operation. At the same time, it updates the dynamic dew point temperature and determines whether the updated dynamic dew point temperature has dropped to a preset safety margin based on the estimated target evaporator pipe wall temperature and the updated dynamic dew point temperature. When the updated dynamic dew point temperature drops to the preset safety margin, the opening of the electronic expansion valve and the speed of the evaporator variable frequency fan are restored and adjusted to the operating state of the control system to sensible heat precooling operation.

7. The intelligent control system for agricultural product processing and transportation based on deep learning according to claim 6, characterized in that, The heat peak early warning and compensation calculation module specifically includes: The derivative slope determination unit is used to calculate the first time derivative of the predicted respiratory heat flux density sequence and generate a trigger signal based on the calculated first time derivative and a pre-configured positive slope threshold. The heat integration unit is used to, upon receiving the trigger signal, extract and perform time-definite integration on the complete predicted respiratory heat flux density sequence within the future time window, based on the trigger signal, to obtain the total accumulated heat of agricultural products released into the enclosed space of the carriage within the future time window. The compensation equivalent conversion unit is used to acquire pre-stored physical environment parameter data, and calculate and obtain the cooling compensation equivalent based on the physical environment parameter data and the total accumulated heat.

8. The intelligent control system for agricultural product processing and transportation based on deep learning according to claim 6, characterized in that, The phase transition risk assessment module specifically includes: The dew point calculation unit is used to extract the current sampling period's dry-bulb temperature sequence data and relative humidity sequence data from the carriage environment data, calculate and determine the intermediate conversion coefficient based on the current sampling period's dry-bulb temperature sequence data and relative humidity sequence data, and calculate the dynamic dew point temperature based on the intermediate conversion coefficient and the current sampling period's dry-bulb temperature sequence data and relative humidity sequence data. The pipe wall temperature measurement unit is used to read the target set temperature of the current carriage and determine the estimated target pipe wall temperature of the evaporator based on the target set temperature of the current carriage, the cooling compensation equivalent, and the pre-calibrated inherent heat exchange temperature difference parameters. The risk comparison and judgment unit is used to compare the dynamic dew point temperature with the estimated target evaporator wall temperature. When the estimated target evaporator wall temperature is less than or equal to the dynamic dew point temperature, it is determined that there is a risk of frost formation in the evaporator, and the judgment result is output.

Images