A fresh food delivery emergency guarantee method

Abstract

The present application provides an emergency guarantee method for fresh food delivery, including: if the temperature distribution does not meet the preset temperature uniformity threshold requirement and the temperature difference exceeds the threshold, identifying the local area where the temperature gradient exceeds the preset gradient threshold and is close to the vehicle door based on the three-dimensional temperature distribution cloud map, and using it as the target area for key adjustment of the fan and door sealing performance; using computational fluid dynamics methods to perform numerical simulation of the airflow field of the target area, the geometric structure characteristics inside the temperature control cabin, the fan layout position, different fan blade angles, and different door opening areas and angles; using the door sensor on the fresh food delivery vehicle to collect the time, angle and frequency data of the door opening and closing in real time, and combining it with a pre-established door opening and closing state judgment model to judge whether the current vehicle is in a frequent door opening and closing condition.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to an emergency guarantee method for fresh food delivery. Background Art

[0002] In the temperature-controlled cabin of fresh produce delivery vehicles, the angle and speed of the air circulation fan blades are key factors influencing uniform temperature distribution. However, due to the frequent opening and closing of vehicle doors to load and unload goods during delivery, the air flow conditions within the cabin constantly change, making it difficult for the fan to adjust in time to accommodate these dynamic changes. A fan speed that is too low can lead to large temperature differences between different areas within the cabin; a speed that is too high can exacerbate temperature fluctuations when the doors are opened and closed. Furthermore, an inappropriate fan blade angle can cause airflow short-circuits within the cabin, making it difficult to control the temperature in certain areas. Especially during peak fresh produce delivery periods, such as before holidays, when vehicles open and close doors more frequently, this can lead to significant temperature fluctuations in the area surrounding the doors, impacting the quality of fresh produce. Frequent door opening and closing causes a significant loss of cool air, causing the cabin temperature to rise sharply. After the doors are closed, the fan takes a long time to cool down, potentially compromising the quality of the goods. In addition to the frequency of door opening and closing, the compression of the door seals is also a significant factor influencing temperature distribution in the area surrounding the doors. Insufficient door seal compression can create gaps between the door and the vehicle body, allowing for rapid loss of cooling air. However, over-compression of the door seal increases the resistance to door opening, affecting loading and unloading efficiency. Therefore, dynamic adjustment of the door seal compression is necessary based on different operating conditions, minimizing cooling loss while ensuring door opening convenience. Dynamically optimizing fan blade angle and speed parameters based on the vehicle's real-time operating status, while also synergistically controlling the door seal compression, minimizes temperature disturbances caused by frequent door opening and closing, and achieves rapid cabin temperature balancing. This is a pressing technical challenge. Summary of the Invention

[0003] The present invention provides a fresh food delivery emergency guarantee method, which mainly includes:

[0004] Acquire real-time temperature data collected by multiple temperature sensors in different areas of the temperature-controlled cabin of fresh food delivery vehicles to generate a three-dimensional temperature distribution cloud map of the entire cabin. This allows users to determine whether the temperature distribution of different cargo storage areas within the cabin meets the preset temperature uniformity threshold requirements.

[0005] If the temperature distribution does not meet the preset temperature uniformity threshold requirement and the temperature difference exceeds the threshold, the system will identify the local area near the door where the temperature gradient exceeds the preset gradient threshold based on the 3D temperature distribution cloud map, and use this area as the target area for key adjustments to the fan and door sealing performance.

[0006] Computational fluid dynamics (CFD) methods were used to numerically simulate the airflow field in the target area, the geometric structural characteristics of the temperature control chamber, the fan layout, different fan blade angles, and different door opening areas and angles.

[0007] Through numerical simulation of the airflow field, the airflow streamline distribution inside the temperature-controlled cabin was obtained for different fan blade angles and door opening areas and angles. The fan angle range and door opening angle threshold that lead to air short-circuiting and outside air infiltration were identified and excluded from the fan and door adjustable ranges. The initial fan speed and the compression of the door seal were determined.

[0008] The door sensors on fresh delivery vehicles collect real-time data on the time, angle, and frequency of door opening and closing. Combined with a pre-established door opening and closing status judgment model, it determines whether the vehicle is currently in a frequent door opening and closing condition.

[0009] If the vehicle is in a condition of frequent door opening and closing, the optimal adjustment parameters corresponding to the current vehicle door opening condition are obtained based on the determined initial fan speed and door seal compression, as well as the pre-established mapping model between the fan blade angle, speed, and the temperature gradient around the door. The fan blade angle, speed, and door seal compression are then adjusted in real time.

[0010] During fan and door seal adjustment, the variance of temperature data near the door in the temperature control chamber before and after adjustment is calculated in real time. Based on the variance trend, the fan blade angle, speed, and door seal compression are optimized online to find the optimal parameter combination. This optimal parameter combination is then used to simultaneously adjust the fan and door seal performance.

[0011] A door temperature prediction model based on a long short-term memory network is used to predict the temperature change trend of the area around the door in the future based on the historical opening and closing status data of the door, the thermal conductivity of the door material, and the current temperature gradient around the door. Based on the predicted temperature change trend, the fan and door sealing strip are pre-adjusted in advance to reduce the temperature fluctuation of the cargo around the door caused by frequent opening and closing of the door.

[0012] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0013] The present invention discloses an emergency guarantee method for fresh food delivery. The method collects temperature data inside a temperature control cabin through multiple temperature sensors, generates a three-dimensional temperature distribution cloud map, and determines whether the temperature uniformity meets the requirements. If the temperature distribution is uneven, the target area where the temperature gradient exceeds the threshold is identified, and the airflow field is numerically simulated using computational fluid dynamics methods to determine the fan and door adjustment parameters. Combined with the door switch state judgment model, the fan blade angle, speed and door seal compression are adjusted in real time. The gradient descent algorithm is used to optimize the adjustment parameters online, and the temperature change trend around the door is predicted based on the long and short-term memory network, and pre-adjustment is performed in advance. The present invention effectively improves the temperature uniformity and stability of the temperature control cabin of the fresh food delivery vehicle through multi-dimensional data collection, intelligent algorithm analysis and dynamic adjustment, reduces temperature fluctuations caused by frequent door opening and closing, and ensures the quality of fresh products. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a flow chart of a fresh food delivery emergency guarantee method of the present invention.

[0015] Figure 2 This is a schematic diagram of a fresh food delivery emergency guarantee method of the present invention.

[0016] Figure 3 This is another schematic diagram of a fresh food delivery emergency guarantee method of the present invention. DETAILED DESCRIPTION

[0017] To further understand the content of the present invention, the present invention is described in detail with reference to the accompanying drawings and examples. The present application is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are intended only to illustrate the relevant invention and are not intended to limit the invention. It should also be noted that, for ease of description, only portions relevant to the invention are shown in the accompanying drawings.

[0018] like Figure 1-3 In this embodiment, a fresh food delivery emergency guarantee method may specifically include:

[0019] S101. Acquire temperature data collected in real time by multiple temperature sensors arranged in different areas of the temperature-controlled cabin of a fresh food delivery vehicle, obtain a three-dimensional temperature distribution cloud map inside the entire temperature-controlled cabin, and determine whether the temperature distribution of different cargo storage areas in the current temperature-controlled cabin meets a preset temperature uniformity threshold requirement.

[0020] Real-time temperature data collected by multiple temperature sensors in the temperature-controlled cabin is obtained, and a three-dimensional temperature distribution matrix inside the temperature-controlled cabin is constructed based on the temperature data; a trilinear interpolation algorithm is used to estimate the temperature values between sampling points in the three-dimensional temperature distribution matrix to obtain a continuous three-dimensional temperature distribution cloud map inside the temperature-controlled cabin; multiple temperature monitoring units are divided according to different cargo storage areas in the temperature-controlled cabin, and the average temperature value and standard deviation in the temperature monitoring units are calculated; if the temperature standard deviation of the temperature monitoring unit exceeds a preset temperature uniformity threshold, it is determined that the temperature distribution of the temperature monitoring unit is uneven; and the cooling or heating power is calculated based on the temperature deviation and change rate of the temperature monitoring unit to generate an adjustment instruction for the temperature control system.

[0021] For example, real-time temperature data collected by multiple temperature sensors within the temperature-controlled cabin is acquired. A three-dimensional temperature distribution matrix is constructed based on the sensor locations, acquisition timestamps, vehicle driving status, and ambient temperature. A trilinear interpolation algorithm is used to estimate the temperature values between sampling points, resulting in a continuous three-dimensional temperature distribution cloud map of the entire temperature-controlled cabin. The cabin is divided into multiple temperature monitoring units for different cargo storage areas. The average temperature value and standard deviation within each monitoring unit are calculated. If the temperature standard deviation of a monitoring unit exceeds a preset temperature uniformity threshold (set at 1°C), the temperature distribution in that area is considered uneven, and the location and temperature fluctuation range of the abnormal area are recorded. Based on the cargo type and storage location information, combined with the temperature distribution of each monitoring unit, the actual temperature deviation value for each cargo environment is calculated. A PID control strategy is used to calculate the cooling or heating power based on the temperature deviation and rate of change. Adjustment commands for the temperature control system are then generated to target the abnormal area with cooling or heating until the temperature uniformity of each monitoring unit meets the preset threshold. Based on the driving status of the delivery vehicle and ambient temperature fluctuations, the temperature data collection frequency and temperature control system response speed are dynamically adjusted to ensure temperature monitoring accuracy while reducing energy consumption. The LZ77 lossless compression algorithm is used to compress temperature data, reducing transmission delay and enabling real-time monitoring and precise adjustment of the temperature distribution within the temperature-controlled cabin. Twenty temperature sensors are placed within the cabin, collecting temperature data every 5 seconds. A 60x60x60 temperature distribution matrix is constructed based on sensor coordinates, acquisition time, vehicle speed, and external temperature. A trilinear interpolation algorithm is used to estimate the temperature between sampling points, generating a continuous three-dimensional temperature cloud map with a resolution of 0.1°C. The cabin is divided into 100 monitoring cells (10x10x10cm), and the average temperature and standard deviation of each cell are calculated. A temperature uniformity threshold of 1°C is set. If the standard deviation of a cell exceeds 1°C, the temperature distribution in that area is considered uneven, and the location and temperature range of the abnormality are recorded. Assume that three types of cargo are stored in the temperature-controlled cabin: Category A (0-4°C), Category B (4-8°C), and Category C (8-12°C). Based on the cargo's location and temperature distribution, the actual temperature deviation for each cargo item is calculated. A PID control strategy is employed, with proportional coefficient Kp = 0.5, integral coefficient Ki = 0.1, and differential coefficient Kd = 0.05. Based on the temperature deviation e(t) and the rate of change de(t) / dt, the cooling or heating power u(t) is calculated as Kp*e(t)+Ki*∫e(t)dt+Kd*de(t) / dt. Adjustment instructions are generated, such as "Increase the power of refrigeration unit 1 by 20%," to precisely address abnormal areas. The data collection frequency is dynamically adjusted based on vehicle speed and external temperature changes. For example, when the vehicle speed exceeds 60 km / h or the external temperature changes by more than 5°C / h, the data collection interval is shortened to 3 seconds. The LZ77 lossless compression algorithm is used to increase the temperature data compression rate to 50%, reducing transmission delays.

[0022] S102. If the temperature distribution does not meet the preset temperature uniformity threshold requirement and the temperature difference exceeds the threshold, identify the local area where the temperature gradient exceeds the preset gradient threshold and is close to the vehicle door based on the three-dimensional temperature distribution cloud map, and use it as the target area for key adjustment of the fan and door sealing performance.

[0023] Obtain a three-dimensional temperature distribution cloud map and use the central difference method to calculate the temperature gradient between adjacent grid points. Determine whether the temperature gradient exceeds a preset threshold and is located within the area surrounding the door. If so, mark the area as a local temperature anomaly. Use a clustering algorithm to group the marked local temperature anomaly areas. Calculate the average temperature, area, and center of gravity of each anomaly area. Based on the internal structure of the vehicle, identify target areas for fan and door sealing performance adjustments.

[0024] For example, based on the acquired three-dimensional temperature distribution cloud map, the central difference method is used to calculate the temperature gradient between adjacent grid points. If the temperature gradient exceeds a preset threshold and is located within the door perimeter, the area is marked as a local temperature anomaly. The area surrounding the door perimeter is dynamically determined based on the vehicle cabin dimensions, typically extending inward from the door edge to 10% of the cabin length. For these marked local temperature anomaly areas, the K-means clustering algorithm is used to group the outliers and calculate the average temperature, area, and center of gravity of each anomaly area. Combined with the vehicle cabin's internal structural information, target areas for fan and door sealing performance adjustments are identified. Based on the temperature anomaly characteristics of the target areas, a fan adjustment model based on a decision tree algorithm is invoked. This model is trained using historical temperature data and adjustment results. Parameters such as the temperature gradient, the area and location of the anomaly area are input, and adjustment instructions such as fan speed and angle are output. A preliminary adjustment recommendation for the door seal pressure is also generated. The fan adjustment instructions are executed by the temperature control system, which continuously monitors temperature changes in the target areas. If the temperature gradient still exceeds the threshold, the door sealing performance optimization process is triggered. The program first detects the pressure distribution of the sealing strip and then adjusts the pressure of each section of the sealing strip based on the size and direction of the temperature gradient. The specific adjustment method is to increase the sealing strip pressure in areas with larger temperature gradients and reduce the sealing strip pressure in areas with smaller temperature gradients. This process is repeated until the temperature gradient in the target area drops below the threshold or reaches the preset upper limit of the adjustment number. 50 temperature sensors are arranged in a refrigerated compartment that is 10 meters long, 3 meters wide, and 2.5 meters high. Temperature data is collected every 2 seconds to construct a 100x30x25 three-dimensional temperature distribution cloud map. The temperature gradient is calculated using the central difference method, and the gradient threshold is set to 2°C / m. The area around the door is defined as within 1 meter of the door. If the temperature gradient of a point exceeds 2°C / m and is within 1 meter of the door, it is marked as an outlier. The K-means clustering algorithm is used, where K = 3, to group the outliers and calculate the average temperature, area, and center of gravity of each group. For example, three abnormal areas were identified: Zone A (4.5°C, 0.5 m2, 0.3 m from the door), Zone B (6.2°C, 0.3 m2, 0.7 m from the door), and Zone C (3.8°C, 0.4 m2, 0.5 m from the door). The decision tree-based fan adjustment model inputs these parameters and outputs adjustment instructions: 1500 rpm, 30° fan speed in Zone A; 1800 rpm, 45° fan speed in Zone B; and 1200 rpm, 20° fan speed in Zone C. It also generates sealing strip pressure adjustment recommendations: 20% increase in Zone A, 15% increase in Zone B, and 10% increase in Zone C. After fan adjustment, temperature changes are continuously monitored. Five minutes later, the temperature gradients in Zones A and C fall below the threshold, while Zone B still exceeds the standard, triggering the sealing performance optimization program. The sealing strip pressure in Zone B is detected to be 0.2 MPa, below the standard of 0.3 MPa. The program gradually increases the sealing strip pressure in Zone B to 0.35 MPa, increasing it by 0.05 MPa at a time and waiting 30 seconds to observe the effect.After two adjustments, the temperature gradient in zone B dropped to 1.8°C / m, which was lower than the threshold, and the optimization was completed.

[0025] S103. Use computational fluid dynamics methods to perform numerical simulation of the airflow field for the target area, the geometric structural characteristics of the temperature control chamber, the fan layout, different fan blade angles, and different door opening areas and angles.

[0026] The internal geometric structure characteristics of the temperature control cabin, the fan layout position, and the door opening state parameters are obtained. A digital model of the temperature control cabin is constructed based on these parameters, and local mesh refinement is performed on the target area identified in the digital model. The computational domain of the digital model is discretized. If the computational domain contains complex geometric shapes, the complex geometric shapes are adaptively processed. Boundary conditions of the computational domain are set, including inlet air velocity, outlet pressure, and a no-slip condition on the wall. The governing equations are discretized using the finite volume method to solve the velocity field and pressure field. The pressure field is solved by solving the Poisson equation. The airflow field distribution under different operating conditions is simulated based on the set fan blade angle and door opening state parameters. The convergence of the simulation results is determined. If the residual of the simulation results is less than a preset threshold, the simulation results are determined to be converged. The converged simulation results are post-processed to generate airflow velocity vector diagrams, temperature value line diagrams, streamline diagrams, and pressure contour diagrams. The airflow velocity, temperature distribution, and pressure gradient data of the target area are extracted from the post-processed results. The extracted data are subjected to dimensionality reduction to determine the main factors affecting the airflow field distribution in the target area.

[0027] For example, a digital model of the temperature control cabin was constructed using AutoCAD software based on parameters such as the cabin's internal geometry, fan placement, and door opening status. Local mesh refinement was performed on identified target areas. A mixed tetrahedral and hexahedral mesh was used to discretize the entire computational domain to accommodate complex geometries. Boundary conditions for the computational domain, including inlet velocity, outlet pressure, and no-slip conditions, were set. The κ-ε turbulence model and the k-ω SST model were selected to describe the airflow characteristics in the high and low Reynolds number regions, respectively. The governing equations were discretized using the finite volume method, and the SIMPLE algorithm was used to solve the velocity and pressure fields. The pressure correction was implemented by solving the Poisson equation. Using parameterized settings, the airflow distribution was simulated for various fan blade angles ranging from 0° to 90° with 10° intervals, as well as for door openings ranging from 0% to 100% with 10% intervals, and for angles from 0° to 90° with 10° intervals. Each operating condition was iterated until convergence, with the convergence criterion set to a residual error of less than 10^-4. The simulation results were post-processed using ParaView software to generate visualizations such as airflow velocity vector diagrams, temperature contour plots, streamlines, and pressure contours. Data mining techniques were used to extract key data such as airflow velocity, temperature distribution, and pressure gradient in the target area. Principal component analysis was used to reduce the dimensionality of multiple simulation results and identify the main factors influencing the airflow distribution in the target area. An airflow characteristic database was generated, providing a quantitative basis for subsequent fan control strategy optimization and door opening management. A precise three-dimensional digital model of a refrigerated compartment measuring 12 meters long, 2.5 meters wide, and 3 meters high was constructed using AutoCAD software, including six fans and one double door. The mesh size was set to 5 mm for the target area within 1 meter of the door and 20 mm for other areas, resulting in approximately 5 million mesh elements. The inlet velocity was set to 5 m / s, the outlet pressure was standard atmospheric pressure, and a no-slip condition was applied to the walls. The k-ω SST model was used near the target area, while the κ-ε model was used for other areas. The relaxation factor for the pressure correction equation in the SIMPLE algorithm was set to 0.3. The simulations included 110 operating conditions, including 10 fan blade angles ranging from 0° to 90° in 10° increments and 11 door openings ranging from 0% to 100% in 10% increments. Each condition was iterated 2000 times or terminated when the residual error was less than 10^-4. ParaView software was used to generate visualizations, such as a velocity vector diagram on a 100×100×100 uniform grid of points in the target area. Key parameters were extracted through data mining, including an average wind speed range of 0.5–2.5 m / s, a temperature uniformity standard deviation of <1°C, and a pressure gradient of <10 Pa / m. Principal component analysis was used to reduce the dimensionality of the 110 simulation results to three principal components, which explained 95% of the total variance. A database of airflow field characteristics containing 11,000 data points was generated, each containing 15 key parameters related to fan angle, door opening, and the target area.

[0028] S104. Through numerical simulation of the airflow field, obtain the airflow streamline distribution diagram inside the temperature-controlled cabin under different fan blade angles and door opening areas and angles, identify the fan angle range and door opening angle threshold that cause airflow short-circuiting and external air infiltration, exclude them from the adjustable range of the fan and door, and determine the initial fan speed and the compression amount of the door sealing strip.

[0029] The system receives numerical simulation results of an airflow field including fan blade angles and door opening states, performs streamline tracing based on the numerical simulation results, and obtains an airflow streamline distribution map. A Canny edge detection algorithm is used on the airflow streamline distribution map to obtain streamline feature data including streamline density, curvature, and length. Based on the streamline feature data, a clustering algorithm is used to determine characteristic patterns of airflow short-circuiting and external air infiltration. The airflow short-circuiting determination criteria are defined as a streamline closure degree greater than a preset closure threshold and a duration exceeding a preset time threshold. External air infiltration is defined as a sudden change in streamline density and a reversal of direction near the door. Based on the characteristic patterns of airflow short-circuiting and external air infiltration, a multivariate linear regression method is used to determine a mathematical model for the constraint relationship between the fan angle and the door opening angle. Based on the constraint relationship mathematical model, a random forest algorithm is used to determine a prediction model for the fan initial speed and the door seal compression. Input parameters of the prediction model include the temperature difference between the inside and outside of the temperature-controlled cabin, the cargo type and load, and the internal structural characteristics of the vehicle compartment. The internal structural characteristics of the vehicle compartment include the location of obstacles and the distribution of cargo.

[0030] For example, based on numerical simulation results of the airflow field, the Runge-Kutta method was used to trace streamlines, generating streamline distribution maps for different fan blade angles and door opening conditions. Streamline features, including streamline density, curvature, and length, were extracted using the Canny edge detection algorithm. The K-means clustering algorithm was used to classify streamline features and identify characteristic patterns of airflow short-circuiting and external air infiltration. The criteria for airflow short-circuiting were set as a streamline closure greater than 80% and lasting for more than 5 seconds, while the criteria for external air infiltration were set as a sudden change in streamline density and a reversal of direction near the door. Through traversal analysis, the fan angle range that causes airflow short-circuiting and the door opening angle threshold that causes external air infiltration were determined. A multivariate linear regression method was used to establish a mathematical model for the constraint relationship between the fan and door opening angles. This constraint relationship was applied to the optimization of the adjustable range, generating a list of optimized fan and door opening angle combinations. Based on the optimized adjustable range, a random forest algorithm was used to train a prediction model for the initial fan speed and door seal compression. Input parameters included the temperature difference between the interior and exterior of the temperature-controlled compartment, the cargo type and load, and internal structural features such as obstacle locations and cargo distribution. The model outputs the initial fan speed and door seal compression, achieving the initial optimal configuration of the temperature-controlled compartment's internal environment. Streamline tracing was performed in a refrigerated compartment measuring 12 meters long, 2.5 meters wide, and 3 meters high using the Runge-Kutta fourth-order method with a time step of 0.01 seconds and 10,000 calculation steps. A streamline distribution map was generated on a 100×100×100 grid. Streamline features were extracted using the Canny edge detection algorithm with a low threshold of 50 and a high threshold of 150. The K-means clustering algorithm was used with k=5 and a maximum number of iterations of 100 to classify the streamline features into five categories. Airflow short-circuiting is defined as the return of more than 80% of streamlines to their starting point within 5 seconds, with a distance less than 0.5 meters. External air infiltration is defined as a 50% increase in streamline density within 1 meter of the door, with the direction opposite to the main flow. By traversing fan angles from 0° to 90° with a step size of 5° and door opening angles from 0° to 60° with a step size of 5°, the constraint relationship is Y = 0.8X + 15, where Y is the maximum allowable door opening angle and X is the fan angle. The optimized adjustable range is 15° to 75° for the fan angle and 0° to 45° for the door opening angle. The random forest algorithm uses 100 decision trees with a maximum depth of 10. It inputs eight features, including the internal and external temperature difference, three cargo types, loading rate, two obstacle location parameters, and cargo distribution uniformity. It outputs an initial fan speed range of 500-2000 rpm and a sealing strip compression range of 2-10 mm. The model was trained on 1000 sets of historical data, achieving a mean absolute error of less than 50 rpm and 0.5 mm.

[0031] S105. The door sensors on the fresh delivery vehicles collect the time, angle, and frequency data of door opening and closing in real time, and combine the pre-established door opening and closing state judgment model to determine whether the current vehicle is in a frequent door opening and closing condition.

[0032] The door opening and closing time and angle data collected by the door sensor are obtained, and the collection frequency of the door sensor is a preset value; the Kalman filter algorithm is applied to the collected data, and the measurement noise covariance of the Kalman filter algorithm is set to a preset value, and the process noise covariance is set to a preset value; the number of door openings, the average opening time and the average opening angle are determined, and the number of door openings, the average opening time and the average opening angle are calculated within a time window dynamically adjusted according to the vehicle operation status; the door opening and closing state is judged, and the judgment is obtained by inputting the number of door openings, the average opening time and the average opening angle into a pre-trained support vector machine model, and the support vector machine model adopts a radial basis function kernel; if the probability value output by the support vector machine model exceeds a preset threshold for a preset number of time windows after exponential moving average smoothing, it is determined that the vehicle is in a frequent door opening and closing condition.

[0033] For example, a door sensor collects real-time door opening and closing time and angle data at a data acquisition frequency of 10 times per second. The raw data is processed using a Kalman filter, with the measurement noise covariance R set to 0.1 and the process noise covariance Q set to 0.01, to produce a smoothed door state change curve. Based on this processed door state change curve, the number of door openings, average opening duration, and average opening angle per unit time are calculated. The time window length is dynamically adjusted based on the vehicle's operating state: 3 minutes for parking, 5 minutes for low-speed driving, and 10 minutes for high-speed driving, sliding every 30 seconds. This computed feature vector is then input into a pre-trained door opening and closing state judgment model. This model utilizes a support vector machine algorithm with a radial basis function kernel, a kernel parameter γ set to 0.1, and a penalty parameter C set to 1. The model outputs the probability of the vehicle experiencing frequent door opening and closing within the current time window. The probability value output by the support vector machine is smoothed by applying exponential moving average, with the smoothing coefficient α set to 0.2, and the judgment threshold for frequent door opening and closing conditions is set to 0.8. If the smoothed probability value exceeds the threshold for three consecutive time windows, it is determined that the current vehicle is in a frequent door opening and closing condition, and the corresponding temperature control strategy adjustment mechanism is triggered. A high-precision angle sensor is installed on a delivery refrigerated truck, and the acquisition frequency is set to 10Hz to record the door opening and closing status. Assume that in a 30-minute delivery process, the original data shows that the door is opened and closed 35 times, with an average opening angle of 75 degrees. The Kalman filter algorithm is used to process the data, and the measurement noise covariance R is set to 0.1, and the process noise covariance Q is set to 0.01. After filtering, the effective number of switches is reduced to 32 times, and the average opening angle is corrected to 72 degrees. The vehicle operation status is judged based on the GPS data, and it is found that the delivery process includes 5 minutes of parking and 15 minutes of low-speed driving.

[0034] <30km / h and 10 minutes of high-speed driving ≥30km / h. Accordingly, three time windows are set as 3 minutes for parking, 5 minutes for low speed, and 10 minutes for high speed. Calculate the feature vector, which includes the number of door openings per minute, the average opening time, and the average opening angle. Input the feature vector into the pre-trained support vector machine model, which uses a radial basis function kernel with a kernel parameter γ of 0.1 and a penalty parameter C of 1. The model output shows that the probability of frequent door opening and closing is 0.9 during parking, 0.7 when driving at low speed, and 0.3 when driving at high speed. Apply the exponential moving average to these probability values with a smoothing coefficient α of 0.2 to obtain a smoothed probability sequence. Set the judgment threshold to 0.8 and find that the probability values exceed the threshold in three consecutive time windows during parking, totaling 9 minutes. It is determined that the vehicle is in a frequent door opening and closing condition, triggering the adjustment of the temperature control strategy.

[0035] S106. If the vehicle is in a condition of frequent door opening and closing, the optimal adjustment parameters corresponding to the current vehicle door opening condition are obtained based on the determined initial fan speed and the compression of the door sealing strip, as well as the pre-established mapping model of the fan blade angle, speed and the temperature gradient around the door, and the fan blade angle, speed and the compression of the door sealing strip are adjusted in real time.

[0036] Acquire characteristic parameters of a vehicle under frequent door opening and closing conditions, including door opening and closing frequency, door opening duration, and cabin temperature change rate; collect fan speed, blade angle, and door sealing strip compression, and use temperature sensors distributed around the door to collect temperature data; calculate a temperature gradient based on the characteristic parameters, fan speed, blade angle, sealing strip compression, and temperature data; input the characteristic parameters, fan speed, blade angle, sealing strip compression, and temperature gradient into a pre-trained multi-layer perceptron neural network; output the optimal fan blade angle, optimal fan speed, and optimal door sealing strip compression under the current vehicle door opening condition from the neural network; adjust the fan blade angle based on the optimal fan blade angle, adjust the fan speed based on the optimal fan speed, and adjust the door sealing strip compression based on the optimal door sealing strip compression.

[0037] For example, characteristic parameters of the current vehicle under frequent door opening and closing conditions are obtained, including the door opening and closing frequency, door opening duration, and cabin temperature change rate, as basic data for subsequent adjustments. The current fan speed, blade angle, and door sealing strip compression are collected. At the same time, temperature sensors distributed around the door are used to collect temperature data once per second, and the temperature gradient around the door is calculated using the central difference method. The collected characteristic parameters, fan parameters, sealing strip parameters, and temperature gradient data are used as input variables and input into a pre-trained multi-layer perceptron neural network. The network includes an input layer (10 nodes), two hidden layers (20 nodes per layer), and an output layer (3 nodes). Based on the input parameters and the mapping relationship between the fan blade angle, speed, and temperature gradient around the door obtained through historical data training, the neural network outputs the optimal fan blade angle, speed, and door sealing strip compression under the current working conditions. Based on the optimal parameters output by the neural network, actuators adjust the fan blade angle, speed, and door seal compression to 0.1mm in real time with an accuracy of 0.1 degrees. The fan blade angle, speed, and door seal compression are all adjusted to an accuracy of 10 rpm. The actuator response time is no more than 100ms. After adjustment, ambient door temperature data is re-collected at a 2Hz frequency to calculate the adjusted temperature gradient. This new data is fed back to the neural network, forming a closed-loop control loop that continuously optimizes the parameters. During a refrigerated delivery truck's operation, the onboard system detected 25 door openings and closings within 30 minutes, with an average door opening time of 45 seconds and a cabin temperature fluctuation of 3°C. This system identified frequent door opening and closing conditions. The system immediately recorded the current fan speed of 1200 rpm, blade angle of 30°, and door seal compression of 5mm. Eight temperature sensors distributed around the door simultaneously collected data, and using the central difference method, the temperature gradient was calculated to be 0.5°C / cm. This data was input into a pretrained multilayer perceptron neural network consisting of 10 input nodes, two hidden layers of 20 nodes each, and three output nodes, using the Reluctant Unit (ReLU) activation function. The network was trained for 500 epochs on 1,000 sets of historical data, with a learning rate of 0.001. The network outputted the optimal fan blade angle of 35°, speed of 1500 rpm, and sealing strip compression of 6 mm. The actuator then began adjusting the blade angle at a rate of 0.1° / s over 50 seconds, increasing the speed at 60 rpm / s over 5 seconds, and increasing the sealing strip compression at a rate of 0.2 mm / s over 5 seconds. After the adjustments were completed, the temperature sensor re-collected data at a frequency of 2 Hz for 15 seconds, calculating a new temperature gradient of 0.3°C / cm. This new data was fed back into the neural network, which outputted fine-tuning recommendations: maintain the blade angle, reduce the speed to 1450 rpm, and increase the sealing strip compression to 6.2 mm.

[0038] S107. During the fan and door seal adjustment process, the variance of the temperature data in the area near the door in the temperature control chamber before and after adjustment is calculated in real time. Based on the variance change trend, the fan blade angle, speed, and door seal compression are optimized online to find the optimal combination of adjustment parameters. The optimal combination of adjustment parameters is then used to simultaneously adjust the fan and door sealing performance.

[0039] Acquire temperature data collected by a temperature sensor, and use the temperature data to calculate temperature variance; construct an objective function using the temperature variance and an energy consumption factor, wherein the objective function includes a temperature variance term and an energy consumption term; determine the fan blade angle, speed, and sealing strip compression as optimization variables, and use the optimization variables to adjust the objective function f(x) = Σ(Ti - Tavg) 2 / n+λE, where Ti is the temperature at each point, Tavg is the average temperature, n is the number of sensors, E is the energy consumption factor, and λ is the weight coefficient; the optimization variables are optimized online using a gradient descent algorithm, which uses an adaptive step size for iteration; if the change in the temperature variance is less than a preset threshold or reaches a maximum number of iterations, the optimization process is stopped; a determination is made as to whether the objective function value is minimized; if so, the current parameter combination is determined to be the global optimal solution; and the global optimal solution is applied to the adjustment of the fan blade angle, speed, and sealing strip compression.

[0040] For example, during the fan and door seal adjustment process, 20 temperature sensors are evenly distributed within a 1-meter radius around the door to collect real-time temperature data from the area near the door in the temperature control chamber. The sampling frequency is set to 10 times per second, and the acquisition time is 30 seconds. The variance of the temperature data before and after adjustment is calculated to obtain the temperature uniformity index. The variance calculation formula is Σ(Ti-Tavg) 2 / n, where Ti is the temperature at each point, Tavg is the average temperature, and n is the number of sensors. Based on the calculated temperature variance trend, the objective function f(x) = Σ(Ti-Tavg) is constructed. 2 / n+λE, where E is the energy consumption factor and λ is the weight coefficient. Temperature variance and energy consumption are used as optimization objectives, and fan blade angle, speed, and door seal compression are used as optimization variables. These three parameters are optimized online using a gradient descent algorithm. During the optimization process, the fan blade angle is adjusted within a range of 0° to 90°, the speed range is 500 to 3000 rpm, and the door seal compression range is 2 to 10 mm. An adaptive step size is used, with initial step sizes of 1°, 100 rpm, and 0.5 mm, respectively, decreasing by 1 / sqrt(k) with the number of iterations k. The single adjustment amplitude is set to no more than 10% of the maximum range to avoid frequent large adjustments. Through multiple iterations, the optimization process is terminated when the change in temperature variance falls below a preset threshold of 0.1°C² or when the maximum number of iterations, 50, is reached. To avoid falling into a local optimum, three independent rounds of optimization are performed using multiple random parameter initializations. The parameter combination with the minimum objective function value is selected as the global optimal solution. Finally, this optimal parameter combination is simultaneously applied to the adjustment of the fan blade angle, speed, and door seal compression. Temperature optimization control is initiated while a refrigerated delivery truck is performing its mission. Twenty high-precision temperature sensors with an accuracy of ±0.1°C are evenly arranged within a 1-meter radius around the door, forming a 4x5 matrix. Data is collected every 0.1 seconds for 30 seconds, resulting in 6,000 temperature data points. Under the initial state, the calculated temperature variance is 2.5°C², and the average temperature is 4°C. The objective function f(x) = Σ(Ti-4) is constructed. 2 / 20+0.01E, where E is the normalized energy consumption factor. The initial optimization variables were set to a fan blade angle of 45°, a speed of 1500 rpm, and a sealing strip compression of 5 mm. A gradient descent algorithm was used, with an initial learning rate of 0.1, which decayed by 10% every 10 iterations. In the first round of iterations, the partial derivatives of the objective function with respect to the three variables were calculated, resulting in a gradient vector of [-0.5, 20, 0.2]. Based on this gradient, the parameters were updated: the blade angle was adjusted to 44.95°, the speed was increased to 1502 rpm, and the compression was increased to 5.01 mm. No single adjustment exceeded 10% of the maximum range. After 30 iterations, the temperature variance decreased to 1.2°C², but the change became more gradual. A second round of optimization was then initiated, with randomly initialized parameters of a blade angle of 60°, a speed of 2000 rpm, and a compression of 7 mm. After 40 iterations in the second round, the temperature variance further decreased to 0.8°C². After the third round of optimization, the optimal parameter combination was a blade angle of 52°, a rotational speed of 1750 rpm, and a compression of 6.5 mm, resulting in a temperature variance of 0.6°C². This set of parameters was used for final adjustments to achieve the optimal balance between temperature uniformity and energy consumption.

[0041] S108. Use a door ambient temperature prediction model based on a long short-term memory network to predict the temperature change trend of the area around the door over a period of time in the future based on the historical door opening and closing status data, the thermal conductivity of the door material, and the current temperature gradient around the door. Pre-adjust the fan and door sealing strip in advance based on the predicted temperature change trend to reduce the temperature fluctuation of the cargo around the door caused by frequent opening and closing of the door.

[0042] Receive door opening and closing status data and temperature data, the data being collected by door sensors; construct an input data set containing time series features based on the data, the input data set including rolling storage data; use a long short-term memory network to construct a temperature prediction model, the prediction model including preset LSTM layers, preset units in each layer, and a dropout layer added in the middle; input the input data set into the prediction model to obtain the future temperature change trend of the area around the door; calculate the interval within which a preset percentage of temperature changes in historical data fall, and define the change beyond the interval as a temperature change threshold; determine whether the temperature change trend exceeds the temperature change threshold, and if so, use a gradient descent algorithm to optimize the objective function; calculate the optimal pre-adjustment parameters for the fan blade angle, speed, and door sealing strip compression amount based on the objective function; perform a pre-adjustment operation before the temperature change occurs, the pre-adjustment operation being based on the optimal pre-adjustment parameters; receive actual temperature change data, and calculate the deviation between the actual temperature change and the predicted value; dynamically adjust the prediction model parameters and pre-adjustment strategy based on the deviation.

[0043] For example, historical door opening and closing status data, including the number of openings and closings, opening and closing times, and opening angles, is collected and stored. The thermal conductivity of the door material, the real-time temperature gradient around the door, and ambient temperature data are also recorded. Data collection is set to once per minute, with the last 24 hours of data stored in a rolling manner. This constructs an input dataset containing time series features. A temperature prediction model is constructed using a long short-term memory network. The input layer has the same number of nodes as the number of features, consisting of three LSTM layers, each with 128 units, interposed with a dropout layer with a dropout rate of 0.2. The output layer is a fully connected layer, outputting minute-by-minute temperature predictions for the next 30 minutes. Mean squared error is used as the loss function, and the Adam optimizer is used with a learning rate of 0.001 for 500 epochs. The constructed input dataset is fed into the prediction model to determine the temperature trend around the door over the next 30 minutes. The normal range is calculated by calculating the interval within which 95% of the temperature changes in the historical data fall. Changes outside this range are defined as the temperature change threshold to determine whether pre-conditioning is necessary. If the predicted temperature change exceeds a threshold, a gradient descent algorithm is used to optimize the objective function. This objective function considers temperature uniformity and energy consumption, and calculates the optimal pre-conditioning parameters for fan blade angle, speed, and door seal compression. Pre-conditioning is performed 10 minutes before the temperature change occurs to mitigate temperature fluctuations. After pre-conditioning, the deviation between the actual temperature change and the predicted value is compared and fed back into the prediction model, dynamically adjusting model parameters and pre-conditioning strategies to continuously optimize prediction accuracy and control effectiveness. During a refrigerated delivery truck's mission, the onboard system collected data every minute, recording an average of 10 door openings per hour, an average opening and closing time of 15 seconds per door, an opening angle between 60° and 90°, a door material thermal conductivity of 0.023 W / (m·K), an average door temperature gradient of 0.5°C / cm, and an ambient temperature range of -5°C to 35°C. The last 24 hours of data are stored in a rolling manner, forming an input matrix of 14,408 elements. The Long Short-Term Memory Network model consists of three LSTM layers, each with 128 units, interleaved with a dropout layer and a learning rate of 0.2. The model was trained using historical data from the past three months, totaling 130,000 samples. It used a mean squared error loss function and the Adam optimizer with a learning rate of 0.001. After 500 training rounds, the mean absolute error on the validation set dropped to 0.3°C. The model predicted the temperature around the car door within the next 30 minutes and found that the temperature could rise from 4°C to 7°C, exceeding the 95% confidence interval of the historical data.

[0044] ±1.5°C. Preconditioning calculations were immediately initiated, using a gradient descent algorithm to optimize the objective function f(x) = w1 T + w2 * E, where T represents temperature uniformity and E represents energy consumption, with weights w1 = 0.7 and w2 = 0.3. After 100 iterations, the optimal preconditioning parameters were obtained: the fan blade angle was adjusted from 25° to 35°, the speed was increased from 1500 rpm to 1800 rpm, and the sealing strip compression was increased from 6 mm to 8 mm. Preconditioning was performed eight minutes before the predicted temperature rise. After preconditioning, the actual temperature only rose to 5.5°C, 1.5°C below the predicted value. This deviation was fed back to the prediction model, and the model parameters were fine-tuned through online learning, reducing the mean absolute error of the next prediction to 0.25°C.

[0045] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A method for emergency guarantee of fresh food delivery, characterized in that: The method comprises: Acquire real-time temperature data collected by multiple temperature sensors in different areas of the temperature-controlled cabin of fresh food delivery vehicles to generate a three-dimensional temperature distribution cloud map of the entire cabin. This allows users to determine whether the temperature distribution of different cargo storage areas within the cabin meets the preset temperature uniformity threshold requirements. If the temperature distribution does not meet the preset temperature uniformity threshold requirement and the temperature difference exceeds the threshold, the system will identify the local area near the door where the temperature gradient exceeds the preset gradient threshold based on the 3D temperature distribution cloud map, and use this area as the target area for key adjustments to the fan and door sealing performance. Computational fluid dynamics (CFD) methods were used to numerically simulate the airflow field in the target area, the geometric structural characteristics of the temperature control chamber, the fan layout, different fan blade angles, and different door opening areas and angles. Through numerical simulation of the airflow field, the airflow streamline distribution inside the temperature-controlled cabin was obtained for different fan blade angles and door opening areas and angles. The fan angle range and door opening angle threshold that lead to air short-circuiting and outside air infiltration were identified and excluded from the fan and door adjustable ranges. The initial fan speed and the compression of the door seal were determined. The door sensors on fresh delivery vehicles collect real-time data on the time, angle, and frequency of door opening and closing. Combined with a pre-established door opening and closing status judgment model, it determines whether the vehicle is currently in a frequent door opening and closing condition. If the vehicle is in a frequent door opening and closing condition, the optimal adjustment parameters corresponding to the current vehicle door opening condition are obtained based on the determined initial fan speed and door seal compression, as well as a pre-established mapping model between the fan blade angle, speed, and the temperature gradient around the door. The fan blade angle, speed, and door seal compression are adjusted in real time. The characteristic parameters of the vehicle under the frequent door opening and closing condition are obtained, including the door opening and closing frequency, door opening duration, and cabin temperature change rate. The fan speed, blade angle, and door seal compression are collected, and temperature data is collected by temperature sensors distributed around the door. Calculate the temperature gradient based on characteristic parameters, fan speed, blade angle, seal strip compression and temperature data; The characteristic parameters, fan speed, blade angle, sealing strip compression and temperature gradient are input into a pre-trained multi-layer perceptron neural network; The neural network outputs the optimal fan blade angle, optimal fan speed, and optimal door seal compression under the current vehicle door opening condition; Adjust the fan blade angle according to the optimal fan blade angle, adjust the fan speed according to the optimal fan speed, and adjust the door seal compression according to the optimal door seal compression; During fan and door seal adjustment, the variance of temperature data near the door in the temperature control chamber before and after adjustment is calculated in real time. Based on the variance trend, the fan blade angle, speed, and door seal compression are optimized online to find the optimal parameter combination. This optimal parameter combination is then used to simultaneously adjust the fan and door seal performance. A door ambient temperature prediction model based on a long short-term memory network is used to predict the temperature change trend of the area around the door over a period of time based on the historical door opening and closing status data, the thermal conductivity of the door material, and the current temperature gradient around the door. Based on the predicted temperature change trend, the fan and door seal are pre-adjusted in advance to reduce the temperature fluctuation of the cargo around the door caused by frequent door opening and closing. The system receives door opening and closing status data and temperature data, which are collected by door sensors. constructing an input data set containing time series features based on the data, wherein the input data set includes rolling storage data; A temperature prediction model is constructed using a long short-term memory network. The prediction model includes preset LSTM layers, each layer has preset units, and a dropout layer is added in the middle. Inputting the input data set into a prediction model to obtain a future temperature change trend of an area around the vehicle door; Calculating an interval within which a preset percentage of temperature changes in historical data fall, and defining changes outside the interval as a temperature change threshold; Determine whether the temperature change trend exceeds a temperature change threshold, and if so, optimize the objective function using a gradient descent algorithm; Calculating optimal pre-adjustment parameters for fan blade angle, rotational speed, and door seal compression based on the objective function; performing a pre-adjustment operation before the temperature change occurs, the pre-adjustment operation being performed based on the optimal pre-adjustment parameters; receiving actual temperature change data, and calculating a deviation between the actual temperature change and a predicted value; The prediction model parameters and pre-adjustment strategies are dynamically adjusted according to the deviation.

2. The method according to claim 1, characterized in that The method includes obtaining temperature data collected in real time by multiple temperature sensors arranged in different areas of the temperature-controlled cabin of the fresh food delivery vehicle, obtaining a three-dimensional temperature distribution cloud map inside the entire temperature-controlled cabin, and determining whether the temperature distribution of different cargo storage areas in the current temperature-controlled cabin meets a preset temperature uniformity threshold requirement, including: Acquiring real-time temperature data collected by multiple temperature sensors in the temperature-controlled chamber, and constructing a three-dimensional temperature distribution matrix inside the temperature-controlled chamber based on the temperature data; Using a trilinear interpolation algorithm to estimate the temperature values between the sampling points in the three-dimensional temperature distribution matrix, to obtain a continuous three-dimensional temperature distribution cloud map inside the temperature control cabin; For different cargo storage areas in the temperature-controlled cabin, multiple temperature monitoring units are divided, and the average temperature value and standard deviation in the temperature monitoring units are calculated; If the temperature standard deviation of the temperature monitoring unit exceeds a preset temperature uniformity threshold, it is determined that the temperature distribution of the temperature monitoring unit is uneven; The cooling or heating power is calculated according to the temperature deviation and the rate of change of the temperature monitoring unit, and an adjustment instruction of the temperature control system is generated.

3. The method according to claim 1, characterized in that If the temperature distribution does not meet the preset temperature uniformity threshold requirement and the temperature difference exceeds the threshold, local areas near the vehicle door where the temperature gradient exceeds the preset gradient threshold are identified based on the three-dimensional temperature distribution cloud map, and the areas are selected as target areas for key adjustments to the fan and door sealing performance, including: Obtain a three-dimensional temperature distribution cloud map and use the central difference method to calculate the temperature gradient between adjacent grid points; Determining whether the temperature gradient exceeds a preset threshold and is located in the area around the door; If the conditions are met, the area is marked as a local temperature anomaly area; The marked local temperature anomaly areas are grouped using a clustering algorithm; Calculate the average temperature, area and center of gravity of each abnormal area; Based on the internal structural information of the vehicle cabin, determine the target areas for key adjustments to the fan and door sealing performance.

4. The method according to claim 1, wherein The computational fluid dynamics method is used to perform numerical simulation of the airflow field for the target area, the geometric structure characteristics of the temperature control cabin, the fan layout, different fan blade angles, and different door opening areas and angles, including: Obtaining the internal geometric structure characteristics of the temperature control cabin, the fan layout position, and the door opening state parameters, constructing a digital model of the temperature control cabin based on the parameters, and performing local mesh encryption on the target area identified in the digital model; discretizing the computational domain of the digital model, and if there are complex geometric shapes in the computational domain, performing adaptive processing on the complex geometric shapes; Setting boundary conditions of the computational domain, wherein the boundary conditions include inlet wind speed, outlet pressure, and a wall no-slip condition; The control equation is discretized using the finite volume method to solve the velocity field and the pressure field. The pressure field is solved by solving the Poisson equation. According to the set fan blade angle and door opening state parameters, simulate the airflow field distribution under different working conditions; Determining whether the simulation result has converged, and if the residual of the simulation result is less than a preset threshold, determining that the simulation result has converged; Post-processing the converged simulation results to generate an airflow velocity vector diagram, a temperature value line diagram, a streamline diagram, and a pressure cloud diagram; Extracting airflow velocity, temperature distribution and pressure gradient data of the target area from the post-processing results; The extracted data is processed by dimensionality reduction to obtain the main factors affecting the airflow field distribution in the target area.

5. The method according to claim 1, wherein The numerical simulation of the airflow field is used to obtain the airflow streamline distribution diagram inside the temperature-controlled cabin under different fan blade angles and door opening areas and angles, identify the fan angle range and door opening angle threshold that cause airflow short-circuiting and external air infiltration, and exclude these from the adjustable range of the fan and door, and determine the initial fan speed and the compression amount of the door sealing strip, including: receiving numerical simulation results of an airflow field including a fan blade angle and a door opening state, and performing streamline tracing according to the numerical simulation results to obtain an airflow streamline distribution map; Using the Canny edge detection algorithm on the airflow streamline distribution diagram to obtain streamline feature data including streamline density, curvature and length; Based on the streamline feature data, a clustering algorithm is used to determine characteristic patterns of airflow short-circuiting and external air infiltration, wherein the airflow short-circuiting determination criterion is that the degree of streamline closure is greater than a preset closure threshold and the duration exceeds a preset time threshold, and the external air infiltration determination criterion is that the streamline density near the vehicle door suddenly changes and the direction reverses; Based on the characteristic patterns of airflow short-circuiting and external air infiltration, a multivariate linear regression method was used to determine the mathematical model of the constraint relationship between the fan angle and the door opening angle. Based on the constraint relationship mathematical model, a random forest algorithm is used to determine a prediction model for the initial fan speed and the compression amount of the door sealing strip. The input parameters of the prediction model include the temperature difference between the inside and outside of the temperature-controlled cabin, the cargo type and load, and the internal structural characteristics of the vehicle compartment. The internal structural characteristics of the vehicle compartment include the location of obstacles and the distribution of cargo.

6. The method according to claim 1, characterized in that The door sensors on the fresh delivery vehicles collect the time, angle, and frequency data of door opening and closing in real time, and combine them with a pre-established door opening and closing state judgment model to determine whether the current vehicle is in a frequent door opening and closing condition, including: Obtaining door opening and closing time and angle data collected by a door sensor, wherein the collection frequency of the door sensor is a preset value; Applying a Kalman filter algorithm to the collected data, wherein the measurement noise covariance of the Kalman filter algorithm is set to a preset value, and the process noise covariance is set to a preset value; Determining the number of door openings, average door opening duration, and average door opening angle, wherein the number of door openings, average door opening duration, and average door opening angle are calculated within a time window dynamically adjusted according to the vehicle operating state; Determining the door opening and closing state by inputting the door opening times, average opening duration, and average opening angle into a pre-trained support vector machine model using a radial basis function kernel; If the probability value output by the support vector machine model exceeds a preset threshold value for a preset number of consecutive time windows after being smoothed by an exponential moving average, it is determined that the vehicle is in a frequent door opening and closing condition.

7. The method according to claim 1, characterized in that During the fan and door seal adjustment process, the variance of the temperature data in the area near the door in the temperature control chamber before and after adjustment is calculated in real time. Based on the variance change trend, the fan blade angle, speed, and door seal compression are optimized online to find the optimal adjustment parameter combination. The optimal adjustment parameter combination is then used to simultaneously adjust the fan and door sealing performance, including: Acquire temperature data collected by a temperature sensor, wherein the temperature data is used to calculate temperature variance; Constructing an objective function using the temperature variance and energy consumption factor, wherein the objective function includes a temperature variance term and an energy consumption term; Determine the fan blade angle, speed and sealing strip compression as optimization variables, which are used to adjust the objective function f(x)=Σ(Ti-Tavg) 2 / n+λE, where Ti is the temperature of each point, Tavg is the average temperature, n is the number of sensors, E is the energy consumption factor, and λ is the weight coefficient; Performing online optimization of the optimization variables using a gradient descent algorithm, wherein the gradient descent algorithm uses an adaptive step size for iteration; If the change in temperature variance is less than the preset threshold or the maximum number of iterations is reached, the optimization process is stopped; Determine whether the objective function value is minimum, and if the objective function value is minimum, determine that the current parameter combination is the global optimal solution; The global optimal solution is applied to adjust the fan blade angle, rotation speed and sealing strip compression.

Images