Temperature control method and system for food preservation box

Abstract

The invention discloses a food preservation box temperature control method and system, and belongs to the technical field of food preservation, in the food preservation box temperature control method and system, food is just in the best preservation state during food taking by intelligently predicting the food taking habit of a user and adjusting the temperature in advance. For example, the system can start pre-cooling or pre-heating two hours in advance according to the food taking time of a user at ordinary times, energy waste caused by the fact that a traditional preservation box needs to continuously maintain low temperature is avoided, and the problem of quality reduction caused by repeated temperature difference changes of food is also solved. Meanwhile, the temperature adjusting process can be dynamically adjusted in combination with the running state of the current equipment, for example, the temperature is rapidly reduced when the heat dissipation efficiency is high, the speed is reduced when the heat dissipation efficiency is low, the equipment runs more stably, power consumption cannot be suddenly increased, and along with the increase of the use time, the prediction model can continuously adapt to the change of the meal taking habit of a user; and the temperature control better meets the actual demand. A user does not need to frequently and manually adjust temperature setting, and daily use is more convenient and worry-saving.

Description

Technical Field

[0001] This invention belongs to the field of food preservation technology, specifically a method and system for temperature control of a food preservation box. Background Technology

[0002] Temperature control in food storage containers is a technology that extends the shelf life of food and maintains its freshness and nutritional quality by regulating the storage environment temperature. It typically employs methods such as semiconductor refrigeration, compressor refrigeration, or phase change materials to achieve precise temperature control. Its core principle lies in setting appropriate temperature ranges based on the characteristics of different foods, inhibiting microbial growth and enzyme activity, and slowing down the oxidative spoilage process. Modern smart food storage containers are also equipped with temperature sensors, PID control algorithms, and IoT functionality, enabling real-time monitoring and adjustment of the internal temperature. They offer advantages such as remote monitoring, energy efficiency, quiet operation, and multi-temperature zone adjustment, and are widely used in homes, catering, fresh food logistics, and medical sample preservation, making them essential equipment for ensuring food safety and improving quality of life.

[0003] However, most existing food storage containers lack intelligent prediction functions and can only keep food fresh by continuously maintaining a fixed low temperature. This not only easily leads to energy waste, but may also cause the food to taste worse due to prolonged low temperature. In addition, users need to manually adjust the temperature settings, and it cannot automatically optimize according to usage habits. It also cannot promptly remind users when the temperature is abnormal, which can easily lead to food spoilage. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for temperature control of food preservation containers in order to solve the problems mentioned above.

[0005] The technical solution adopted in this invention is as follows: A method for temperature control of a food preservation container, the method comprising the following steps:

[0006] S1: System initialization and multi-sensor calibration, start the self-test program of temperature, humidity and air pressure sensors, synchronously record the initial environmental parameters and transmit them to the central controller to provide reference values ​​for subsequent data acquisition and analysis;

[0007] S2: Intelligent food type identification and parameter matching. The category is determined by scanning the food packaging barcode with a camera or by image recognition technology. The pre-stored food preservation database (associated with the initial parameters of S1) is called up to automatically match the optimal temperature range and humidity threshold for the category.

[0008] S3: Dynamic temperature field modeling and zone control. Based on the parameters determined in S2, combined with the three-dimensional space heat conduction simulation inside the box, the food storage box is divided into at least 3 independent temperature control zones. Each zone is equipped with a micro PTC heating element and a semiconductor cooling element to achieve precise temperature difference control of ±0.5℃ between zones.

[0009] S4: Dual-cycle heat exchange is executed, the main temperature control loop is started to coarsely adjust the overall ambient temperature (in response to the zone command of S3), and the secondary heat pipe system is activated. The heat dissipation efficiency is adjusted by the humidity sensor data calibrated by S1 to avoid condensation.

[0010] S5: Phase change energy storage module charge and discharge management. When the cooling / heating power of S4 exceeds the preset threshold, the phase change material (PCM) energy release mode is triggered. The energy storage unit is dynamically allocated through the S3 partition temperature data, and the excess energy is temporarily stored in the phase change module and released in reverse during the power trough.

[0011] S6: Adaptive learning and predictive temperature control, real-time collection of operating data from S1 to S5, training of user habit model through LSTM neural network, such as identifying peak time for users to pick up meals every day, starting the pre-cooling / preheating program of S4 2 hours in advance, and feeding back the prediction results to S3 to update the zone temperature strategy.

[0012] S7: Cross-system collaborative error correction mechanism, continuously compares the food parameter database of S2 with the prediction model output of S6. When the actual temperature deviates from the predicted value by more than 2°C, it immediately freezes the energy storage allocation of S5 and switches to the emergency temperature control mode of S4, while sending a warning message to the user terminal.

[0013] S8: User behavior feedback and parameter iteration. Receives food preservation effect ratings input by users through the APP (related to the category identification results of S2), incorporates the rating data into the training set of the learning model in S6, generates an optimization report weekly, and automatically updates the partition control algorithm of S3 and the database parameters of S2.

[0014] In a preferred embodiment, during step S1, a three-level self-test mechanism is activated during system initialization. The temperature sensor (PT1000 type) is calibrated by comparing with a built-in high-precision constant temperature bath (temperature control accuracy ±0.05℃). The calibration cycle is set to once every 24 hours. If a signal drift exceeding 0.2℃ is detected during calibration, the sensor reset procedure is automatically triggered. The humidity sensor (capacitive type) uses a two-point calibration method, introducing standard humidity gases of 30%RH and 70%RH respectively. The measurement error is controlled within ±2%RH by adjusting the internal compensation resistor. The barometric pressure sensor (MEMS type) collects the current ambient atmospheric pressure in real time and converts it into an altitude correction coefficient. This coefficient will serve as the basic parameter for temperature field modeling in S3, ensuring the accuracy of heat conduction simulation at different altitudes. All sensor data is transmitted to the central controller via an RS485 bus at a transmission rate of 115200bps. The data frame format uses a 16-bit CRC checksum to avoid transmission errors.

[0015] In a preferred embodiment, in step S2, the image recognition module uses a lightweight YOLOv5s model to capture real-time images of the food packaging (resolution 1280×720, frame rate 15fps), and generates category labels by extracting shape features and text areas on the packaging (OCR recognition accuracy ≥98%). The barcode scanning module supports EAN-13 and UPC-A codes, with a scanning distance set to 5–15 cm. If three consecutive scans fail, it automatically switches to image recognition mode. The food parameter database uses a distributed storage architecture, locally caching the optimal temperature control parameters for commonly used categories (approximately 500 types). The cloud database is incrementally updated daily at 2 AM, including temperature sensitivity coefficients for newly added categories (e.g., the temperature fluctuation tolerance value for tropical fruits is ±1.5℃) and humidity-coordinated thresholds (e.g., bread requires humidity and temperature linkage control; for every 10% decrease in humidity, the temperature needs to be lowered by 0.8℃).

[0016] In a preferred embodiment, in step S3, the partitioning is performed using the K-means clustering algorithm. Based on the user's historical meal pickup coordinate data (sampling density 1 time / minute), the internal space of the food storage container is divided into a core area (40%), a transition area (35%), and an edge area (25%). The core area is equipped with a dual-redundant temperature control unit (PTC heating element power 5W, semiconductor cooling element maximum cooling capacity 8W). Temperature field modeling is performed using the finite element method, with a mesh generation accuracy of 5mm×5mm×5mm, by solving the heat conduction equation. Steady-state temperature distribution was calculated, with the thermal diffusivity α dynamically adjusted based on the air pressure data of S1 (α value increased by 2.3% for every 1000m increase in altitude). Zonal control employed a model predictive control (MPC) algorithm, with a rolling optimization time domain of 120 seconds. Control variables included heating / cooling power duty cycle and fan speed. The objective function was to minimize the sum of squared temperature deviations and the weighted average of energy consumption.

[0017] In a preferred embodiment, in step S4, the main temperature control circuit adopts a vapor compression refrigeration system with a compressor displacement of 2.5cc / rev and an evaporation temperature setting range of -5℃ to 15℃. The refrigerant flow is adjusted via an electronic expansion valve (500 pulse steps) to achieve a coarse adjustment accuracy of ±1℃. The secondary heat pipe system uses gravity-type heat pipes (working fluid is R134a). The number of heat pipes is configured according to the partition area of ​​S3 (4 in the core area, 2 in the transition area, and 1 in the edge area). The heat dissipation end is equipped with aluminum fins (fin spacing 2mm) and a PWM-controlled fan (speed range 800-2500RPM). The humidity-heat dissipation linkage mechanism is implemented through fuzzy control: when the humidity sensor reading in S1 exceeds 65%RH, the fan speed increases to 80% of the rated value, and simultaneously, the micro heating wires (1.2W power) on the fin surface are activated to prevent frost formation. The heating wire operating time is positively correlated with the humidity value (for every 5% increase in humidity RH, the heating time increases by 15 seconds).

[0018] In a preferred embodiment, in step S5, the phase change energy storage module consists of three independent units, each containing 500g of paraffin-based phase change material with a phase change temperature of 5°C and a latent heat of 205kJ / kg. These units are coupled to the evaporator / condenser of S4 via a copper heat-conducting plate. The charge / discharge thresholds are set as follows: when the compressor power of S4 exceeds 35W for 30 seconds (cooling mode) or the PTC heating element power exceeds 25W (heating mode), energy storage release is triggered; when the power is below 15W for 1 minute, energy storage charging is initiated. Energy allocation uses a greedy algorithm, prioritizing the energy storage needs of the core area (weight 0.5), followed by the transition area (0.3) and the edge area (0.2). The charge / discharge status of each unit is monitored in real-time by an NTC temperature sensor (sampling interval 10 seconds). When the phase change material is detected to be completely solidified (temperature ≤3°C) or completely melted (temperature ≥7°C), the adjacent unit is automatically switched to operation.

[0019] In a preferred embodiment, in step S6, a multi-source time-series dataset is first constructed. The sensor data calibrated in S1, the partition temperature control command in S3, the heat exchange power fluctuation value in S4, and the phase change energy storage module charging and discharging status in S5 are aligned by timestamps to form an input matrix of dimension .

[0020] The LSTM neural network adopts a two-layer architecture. The first hidden layer extracts short-term dependency features through a gating mechanism, the second layer captures long-term trends, and the output layer is activated by softmax to obtain the probability distribution of meal pick-up for each hour in the next 24 hours.

[0021] The model training uses a sliding window method, and the parameters are automatically updated at 3:00 AM every day. At the same time, user ratings of S8 are introduced as a supervision signal. When the prediction accuracy is below 85% for three consecutive days in a certain period, the feature importance is reordered, and the weights of key variables such as heat exchange efficiency of S4 and energy storage release speed of S5 are strengthened.

[0022] During the predictive temperature control execution phase, the system marks the top-2 high-probability periods in the probability distribution output by the LSTM as the "predicted food retrieval window." Combined with the food parameter database in S2, it calculates the pre-cooling / preheating start time. For example, if the model predicts that users have the highest probability of retrieval between 12:00 and 13:00, and the optimal temperature matched by S2 is 5℃, then a "zone temperature pre-adjustment instruction" is sent to S3: the target temperature of frequently used areas is gradually reduced from the current 7℃ to 5℃, with the cooling rate dynamically adjusted based on the current heat dissipation efficiency in S4. Simultaneously, the prediction results update the zone priority matrix in S3 through the feedback channel, improving the temperature control accuracy of high-frequency food retrieval areas from ±0.5℃ to ±0.3℃, and reducing the energy consumption allocation weight of low-usage areas.

[0023] The formula for predicting the probability of meal pickup is:

[0024]

[0025] In the formula:

[0026] P(t) represents the probability of picking up the meal in the t-th hour in the future;

[0027] h t This represents the hidden layer output vector of the LSTM network at time step t;

[0028] W o Representation: Output layer weight matrix;

[0029] b o Indicates the output layer bias term;

[0030] U t This represents the user habit factor, calculated from S8's historical ratings: Where si is the i-th rating, ti is the corresponding meal pick-up time, and δ(·) is the Dirac function.

[0031] α: Habitual factor weight;

[0032] The formula for calculating the precooling / preheating start-up time is:

[0033]

[0034] In the formula:

[0035] T start Indicates the pre-adjusted start time;

[0036] t peak Indicates the predicted start time of the peak meal pickup period;

[0037] T target This indicates the optimal temperature for S2 matching;

[0038] T current This indicates the real-time temperature of the current area;

[0039] r max This indicates the maximum cooling / heating rate of the S4 heat exchange system;

[0040] Δt min This indicates the minimum pre-adjustment duration.

[0041] In a preferred embodiment, in step S7, deviation monitoring employs a sliding window comparison method with a window length of 5 minutes. The root mean square error (RMSE) between the actual temperature sequence and the predicted sequence in S6 is calculated. When RMSE > 2°C, the error correction process is immediately activated. In emergency temperature control mode, the system cuts off the energy storage module output of S5 and switches the main circuit of S4 to "full-speed operation" (compressor speed increased to 110% of rated value, heating element duty cycle set to 100%). Simultaneously, the secondary heat pipe system fan is locked at maximum speed. Warning information is transmitted to the user terminal via the LoRaWAN protocol. The information includes the deviation value, the current temperature of each zone, and the operating status code of S4. If no user confirmation command is received within 15 minutes, the backup power supply is automatically activated to maintain the core area temperature.

[0042] In a preferred embodiment, in step S8, the user submits a rating on a 5-point scale via the APP interface. The system stores the rating data according to three dimensions: "preservation effect - category - environmental parameters," where 1 point corresponds to "severe spoilage" and 5 points correspond to "completely fresh." The rating weights use a time decay function: 0.6 for ratings within the last 7 days, 0.3 for ratings from 8 to 30 days, and 0.1 for ratings over 30 days. The weight values ​​are used in real-time for training the LSTM model in S6. Parameter iteration is performed every Sunday at 3 AM, updating the partition control algorithm in S3 and the database parameters in S2 using gradient descent. If the model loss function is detected to increase by more than 5% during iteration, the system automatically rolls back to the previous version of parameters and records the anomaly log.

[0043] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0044] 1. In this invention, the system intelligently predicts users' meal-collecting habits and adjusts the temperature in advance to ensure that food is in its optimal freshness state when it is picked up. For example, the system can start pre-cooling or preheating two hours in advance based on the user's usual meal-collecting time, avoiding the energy waste caused by traditional food storage containers needing to continuously maintain low temperatures, and also reducing the quality degradation caused by repeated temperature changes in food. At the same time, the temperature adjustment process is dynamically adjusted based on the current operating status of the equipment. For example, when the heat dissipation efficiency is high, the temperature is cooled down quickly, and when the efficiency is low, the speed is slowed down, making the equipment operate more stably and preventing sudden increases in power consumption.

[0045] 2. In this invention, as usage time increases, the predictive model continuously adapts to changes in users' meal-taking habits, making temperature control more aligned with actual needs. When a temperature deviation is detected, the system automatically triggers an error correction mechanism and sends a warning message to the user, reducing the risk of food spoilage due to equipment malfunction. Users do not need to frequently manually adjust temperature settings; the equipment can continuously iterate and optimize temperature control parameters based on feedback data, making daily use more convenient and worry-free. Attached Figure Description

[0046] Figure 1 This is a schematic diagram illustrating the process principle of the present invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0048] Example:

[0049] Reference Figure 1 ,

[0050] A method for temperature control of a food storage container, comprising the following steps:

[0051] S1: System initialization and multi-sensor calibration, start the self-test program of temperature, humidity and air pressure sensors, synchronously record the initial environmental parameters and transmit them to the central controller to provide reference values ​​for subsequent data acquisition and analysis;

[0052] S2: Intelligent food type identification and parameter matching. The category is determined by scanning the food packaging barcode with a camera or by image recognition technology. The pre-stored food preservation database (associated with the initial parameters of S1) is called up to automatically match the optimal temperature range and humidity threshold for the category.

[0053] S3: Dynamic temperature field modeling and zone control. Based on the parameters determined in S2, combined with the three-dimensional space heat conduction simulation inside the box, the food storage box is divided into at least 3 independent temperature control zones. Each zone is equipped with a micro PTC heating element and a semiconductor cooling element to achieve precise temperature difference control of ±0.5℃ between zones.

[0054] S4: Dual-cycle heat exchange is executed, the main temperature control loop is started to coarsely adjust the overall ambient temperature (in response to the zone command of S3), and the secondary heat pipe system is activated. The heat dissipation efficiency is adjusted by the humidity sensor data calibrated by S1 to avoid condensation.

[0055] S5: Phase change energy storage module charge and discharge management. When the cooling / heating power of S4 exceeds the preset threshold, the phase change material (PCM) energy release mode is triggered. The energy storage unit is dynamically allocated through the S3 partition temperature data, and the excess energy is temporarily stored in the phase change module and released in reverse during the power trough.

[0056] S6: Adaptive learning and predictive temperature control, real-time collection of operating data from S1 to S5, training of user habit model through LSTM neural network, such as identifying peak time for users to pick up meals every day, starting the pre-cooling / preheating program of S4 2 hours in advance, and feeding back the prediction results to S3 to update the zone temperature strategy.

[0057] S7: Cross-system collaborative error correction mechanism, continuously compares the food parameter database of S2 with the prediction model output of S6. When the actual temperature deviates from the predicted value by more than 2°C, it immediately freezes the energy storage allocation of S5 and switches to the emergency temperature control mode of S4, while sending a warning message to the user terminal.

[0058] S8: User behavior feedback and parameter iteration. Receives food preservation effect ratings input by users through the APP (related to the category identification results of S2), incorporates the rating data into the training set of the learning model in S6, generates an optimization report weekly, and automatically updates the partition control algorithm of S3 and the database parameters of S2.

[0059] In step S1, a three-level self-test mechanism needs to be activated during system initialization. The temperature sensor (PT1000 type) is calibrated by comparing with the built-in high-precision constant temperature bath (temperature control accuracy ±0.05℃). The calibration cycle is set to once every 24 hours. If a signal drift exceeding 0.2℃ is detected during calibration, the sensor reset procedure is automatically triggered. The humidity sensor (capacitive type) uses a two-point calibration method, introducing standard humidity gases of 30%RH and 70%RH respectively. The measurement error is controlled within ±2%RH by adjusting the internal compensation resistor. The barometric pressure sensor (MEMS type) collects the current ambient atmospheric pressure in real time and converts it into an altitude correction coefficient. This coefficient will serve as the basic parameter for S3 temperature field modeling, ensuring the accuracy of heat conduction simulation at different altitudes. All sensor data is transmitted to the central controller via an RS485 bus at a transmission rate of 115200bps. The data frame format uses a 16-bit CRC checksum to avoid transmission errors.

[0060] In step S2, the image recognition module uses a lightweight YOLOv5s model to capture real-time images of food packaging (1280×720 resolution, 15fps). It generates category labels by extracting shape features and text areas from the packaging (OCR recognition accuracy ≥98%). The barcode scanning module supports EAN-13 and UPC-A codes, with a scanning distance set to 5–15cm. If three consecutive scans fail, it automatically switches to image recognition mode. The food parameter database uses a distributed storage architecture, locally caching the optimal temperature control parameters for approximately 500 commonly used categories. The cloud database is incrementally updated daily at 2 AM, including temperature sensitivity coefficients for newly added categories (e.g., the temperature fluctuation tolerance of tropical fruits is ±1.5℃) and humidity-coordinated thresholds (e.g., bread requires linked humidity and temperature control; for every 10% decrease in humidity, the temperature needs to be lowered by 0.8℃).

[0061] In step S3, the K-means clustering algorithm is used for partitioning. Based on the user's historical food collection coordinate data (sampling density 1 time / minute), the internal space of the food storage container is divided into a core area (40%), a transition area (35%), and an edge area (25%). The core area is equipped with a dual-redundant temperature control unit (PTC heating element power 5W, semiconductor cooling element maximum cooling capacity 8W). The temperature field modeling adopts the finite element analysis method, with a mesh generation accuracy of 5mm×5mm×5mm, and the heat conduction equation is solved. Steady-state temperature distribution was calculated, with the thermal diffusivity α dynamically adjusted based on the air pressure data of S1 (α value increased by 2.3% for every 1000m increase in altitude). Zonal control employed a model predictive control (MPC) algorithm, with a rolling optimization time domain of 120 seconds. Control variables included heating / cooling power duty cycle and fan speed. The objective function was to minimize the sum of squared temperature deviations and the weighted average of energy consumption.

[0062] In step S4, the main temperature control circuit adopts a vapor compression refrigeration system with a compressor displacement of 2.5cc / rev and an evaporation temperature setting range of -5℃ to 15℃. The refrigerant flow is adjusted via an electronic expansion valve (500 pulse steps) to achieve a coarse adjustment accuracy of ±1℃. The secondary heat pipe system uses gravity-type heat pipes (using R134a as the working fluid). The number of heat pipes is configured according to the partition area of ​​S3 (4 in the core area, 2 in the transition area, and 1 in the edge area). The heat dissipation end is equipped with aluminum fins (fin spacing 2mm) and a PWM-controlled fan (speed range 800-2500RPM). The humidity-heat dissipation linkage mechanism is implemented through fuzzy control: when the humidity sensor reading in S1 exceeds 65% RH, the fan speed increases to 80% of the rated value, and simultaneously, the miniature heating wire (1.2W power) on the fin surface is activated to prevent frost formation. The heating wire's operating time is positively correlated with the humidity value (for every 5% increase in humidity RH, the heating time increases by 15 seconds).

[0063] In step S5, the phase change energy storage module consists of three independent units, each containing 500g of paraffin-based phase change material with a phase change temperature of 5℃ and a latent heat of 205kJ / kg. These units are coupled to the evaporator / condenser of S4 via a copper heat-conducting plate. The charge / discharge thresholds are set as follows: when the compressor power of S4 exceeds 35W for 30 seconds (cooling mode) or the PTC heating element power exceeds 25W (heating mode), energy storage release is triggered; when the power is below 15W for 1 minute, energy storage charging is initiated. Energy allocation uses a greedy algorithm, prioritizing the energy storage needs of the core area (weight 0.5), followed by the transition area (0.3) and the edge area (0.2). The charge / discharge status of each unit is monitored in real-time by an NTC temperature sensor (sampling interval 10 seconds). When the phase change material is detected to be completely solidified (temperature ≤3℃) or completely melted (temperature ≥7℃), the adjacent unit is automatically switched to operation.

[0064] In step S6, a multi-source time-series dataset is first constructed. The sensor data calibrated in S1, the partition temperature control command in S3, the heat exchange power fluctuation value in S4, and the charging and discharging status of the phase change energy storage module in S5 are aligned by timestamps to form an input matrix of dimension .

[0065] The LSTM neural network adopts a two-layer architecture. The first hidden layer extracts short-term dependency features through a gating mechanism, the second layer captures long-term trends, and the output layer is activated by softmax to obtain the probability distribution of meal pick-up for each hour in the next 24 hours.

[0066] The model training uses a sliding window method, and the parameters are automatically updated at 3:00 AM every day. At the same time, user ratings of S8 are introduced as a supervision signal. When the prediction accuracy is below 85% for three consecutive days in a certain period, the feature importance is reordered, and the weights of key variables such as heat exchange efficiency of S4 and energy storage release speed of S5 are strengthened.

[0067] During the predictive temperature control execution phase, the system marks the top-2 high-probability periods in the probability distribution output by the LSTM as the "predicted food retrieval window." Combined with the food parameter database in S2, it calculates the pre-cooling / preheating start time. For example, if the model predicts that users have the highest probability of retrieval between 12:00 and 13:00, and the optimal temperature matched by S2 is 5℃, then a "zone temperature pre-adjustment instruction" is sent to S3: the target temperature of frequently used areas is gradually reduced from the current 7℃ to 5℃, with the cooling rate dynamically adjusted based on the current heat dissipation efficiency in S4. Simultaneously, the prediction results update the zone priority matrix in S3 through the feedback channel, improving the temperature control accuracy of high-frequency food retrieval areas from ±0.5℃ to ±0.3℃, and reducing the energy consumption allocation weight of low-usage areas.

[0068] The formula for predicting the probability of meal pickup is:

[0069]

[0070] In the formula:

[0071] P(t) represents the probability of picking up the meal in the t-th hour in the future;

[0072] h t This represents the hidden layer output vector of the LSTM network at time step t;

[0073] W o Representation: Output layer weight matrix;

[0074] b o Indicates the output layer bias term;

[0075] U t This represents the user habit factor, calculated from S8's historical ratings: Where si is the i-th rating, ti is the corresponding meal pick-up time, and δ(·) is the Dirac function.

[0076] α: Habitual factor weight;

[0077] The formula for calculating the precooling / preheating start-up time is:

[0078]

[0079] In the formula:

[0080] T start Indicates the pre-adjusted start time;

[0081] t peak Indicates the predicted start time of the peak meal pickup period;

[0082] T target This indicates the optimal temperature for S2 matching;

[0083] T current This indicates the real-time temperature of the current area;

[0084] r max This indicates the maximum cooling / heating rate of the S4 heat exchange system;

[0085] Δt min This indicates the minimum pre-adjustment duration.

[0086] In step S7, deviation monitoring uses a sliding window comparison method with a window length of 5 minutes. The root mean square error (RMSE) between the actual temperature sequence and the predicted sequence in S6 is calculated. When RMSE > 2℃, the error correction process is activated immediately. In emergency temperature control mode, the system cuts off the energy storage module output of S5 and switches the main circuit of S4 to "full speed operation" (compressor speed increased to 110% of the rated value, heating element duty cycle set to 100%). At the same time, the secondary heat pipe system fan is locked at maximum speed. Warning information is transmitted to the user terminal via the LoRaWAN protocol. The information includes the deviation value, the current temperature of each zone, and the operating status code of S4. If no user confirmation command is received within 15 minutes, the backup power supply is automatically activated to maintain the core area temperature.

[0087] In step S8, users submit their ratings on a 5-point scale via the app interface. The system stores the rating data categorized into three dimensions: "preservation effect - category - environmental parameters," where 1 point corresponds to "severe spoilage" and 5 points correspond to "completely fresh." The rating weights use a time decay function: 0.6 for ratings within the last 7 days, 0.3 for ratings from 8 to 30 days, and 0.1 for ratings older than 30 days. These weights are used in real-time for training the LSTM model in S6. Parameter iteration is performed every Sunday at 3 AM, updating the partition control algorithm in S3 and the database parameters in S2 using gradient descent. If the model loss function increases by more than 5% during iteration, the system automatically rolls back to the previous version of parameters and records the anomaly.

[0088] A temperature control system for a food preservation container, which is applied to the aforementioned temperature control method for the food preservation container.

[0089] From the above, we can conclude that:

[0090] In this invention, the system intelligently predicts users' meal-collecting habits and adjusts the temperature in advance to ensure that food is in its optimal freshness state when it is picked up. For example, the system can start pre-cooling or preheating two hours in advance based on the user's usual meal-collecting time, avoiding the energy waste caused by traditional food storage containers needing to maintain a low temperature continuously, and also reducing the quality degradation caused by repeated temperature changes in food. At the same time, the temperature adjustment process is dynamically adjusted based on the current operating status of the equipment. For example, it cools down quickly when the heat dissipation efficiency is high and slows down when the efficiency is low, making the equipment operate more stably and preventing sudden increases in power consumption.

[0091] In this invention, as usage time increases, the predictive model continuously adapts to changes in users' meal-taking habits, making temperature control more aligned with actual needs. When a temperature deviation is detected, the system automatically triggers an error correction mechanism and sends a warning message to the user, reducing the risk of food spoilage due to equipment malfunction. Users do not need to frequently manually adjust temperature settings; the equipment can continuously iterate and optimize temperature control parameters based on feedback data, making daily use more convenient and worry-free.

[0092] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the term "comprising" or any other variations thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0093] The foregoing description enables those skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for temperature control of a food preservation container, characterized in that: The method includes the following steps: S1: System initialization and multi-sensor calibration, start the self-test program of temperature, humidity and air pressure sensors, synchronously record the initial environmental parameters and transmit them to the central controller to provide reference values ​​for subsequent data acquisition and analysis; S2: Intelligent food type recognition and parameter matching. By scanning the food packaging barcode with a camera or using image recognition technology, the category is determined, and the pre-stored food preservation database is called to automatically match the optimal temperature range and humidity threshold for that category. S3: Dynamic temperature field modeling and zone control. Based on the parameters determined in S2, combined with the three-dimensional space heat conduction simulation inside the box, the food storage box is divided into at least 3 independent temperature control zones. Each zone is equipped with a micro PTC heating element and a semiconductor cooling element to achieve precise temperature difference control of ±0.5℃ between zones. S4: Dual-cycle heat exchange is executed, starting the main temperature control loop to coarsely adjust the overall ambient temperature, while activating the secondary heat pipe system. The heat dissipation efficiency is adjusted based on the humidity sensor data calibrated by S1 to prevent condensation. S5: Phase change energy storage module charge and discharge management. When the cooling / heating power of S4 exceeds the preset threshold, the phase change material release mode is triggered. The energy storage unit is dynamically allocated through the temperature data of S3 partition, and the excess energy is temporarily stored in the phase change module and released in reverse during the power trough. S6: Adaptive learning and predictive temperature control, real-time collection of operating data from S1 to S5, training of user habit model through LSTM neural network, such as identifying peak time for users to pick up meals every day, starting the pre-cooling / preheating program of S4 2 hours in advance, and feeding back the prediction results to S3 to update the zone temperature strategy. S7: Cross-system collaborative error correction mechanism, continuously compares the food parameter database of S2 with the prediction model output of S6. When the actual temperature deviates from the predicted value by more than 2°C, it immediately freezes the energy storage allocation of S5 and switches to the emergency temperature control mode of S4, while sending a warning message to the user terminal. S8: User behavior feedback and parameter iteration. Receives food preservation effect ratings input by users through the APP, incorporates the rating data into the training set of the learning model in S6, generates an optimization report weekly, and automatically updates the partition control algorithm in S3 and the database parameters in S2.

2. The temperature control method for a food preservation box as described in claim 1, characterized in that: In step S1, a three-level self-test mechanism needs to be activated during the system initialization phase. The temperature sensor is calibrated by comparing with the built-in high-precision constant temperature bath. The calibration cycle is set to once every 24 hours. If a signal drift exceeding 0.2℃ is detected during the calibration process, the sensor reset program is automatically triggered. The humidity sensor adopts a two-point calibration method, introducing standard humidity gases of 30%RH and 70%RH respectively. The measurement error is controlled within ±2%RH by adjusting the internal compensation resistor. The barometric pressure sensor collects the current ambient atmospheric pressure in real time and converts it into an altitude correction coefficient. This coefficient will serve as the basic parameter for the temperature field modeling in S3, ensuring the accuracy of heat conduction simulation in different altitude regions. All sensor data is transmitted to the central controller via an RS485 bus. The transmission rate is set to 115200bps, and the data frame format uses 16-bit CRC checksum to avoid transmission errors.

3. The temperature control method for a food preservation box as described in claim 1, characterized in that: In step S2, the image recognition module uses the YOLOv5s lightweight model to take real-time photos of the food packaging and generate category labels by extracting shape features and text areas on the packaging; the barcode scanning module supports EAN-13 and UPC-A codes, and the scanning distance is set to 5-15cm. If three consecutive scans fail, it automatically switches to image recognition mode; the food parameter database adopts a distributed storage architecture, with the best temperature control parameters for commonly used categories cached locally, and the cloud database is incrementally updated every day at 2:00 AM, including the temperature sensitivity coefficient and humidity co-sensitivity threshold for newly added categories.

4. The temperature control method for a food preservation box as described in claim 1, characterized in that: In step S3, the partitioning adopts the K-means clustering algorithm. Based on the user's historical food collection coordinate data, the internal space of the food storage box is divided into a core area, a transition area, and an edge area. The core area is equipped with a dual redundant temperature control unit. The temperature field modeling adopts the finite element analysis method, with a mesh division accuracy of 5mm×5mm×5mm. The steady-state temperature distribution is calculated by solving the heat conduction equation, wherein the thermal diffusivity α is dynamically corrected according to the air pressure data in S1. The partition control adopts the model predictive control algorithm, with the rolling optimization time domain set to 120 seconds. The control variables include the heating / cooling power duty cycle and the fan speed. The objective function is the minimum value of the sum of squared temperature deviations and the weighted value of energy consumption.

5. The temperature control method for a food preservation box as described in claim 1, characterized in that: In step S4, the main temperature control circuit adopts a vapor compression refrigeration system with a compressor displacement of 2.5cc / rev and an evaporation temperature setting range of -5℃ to 15℃. The refrigerant flow is adjusted by an electronic expansion valve to achieve a coarse adjustment accuracy of ±1℃. The secondary heat pipe system adopts gravity-type heat pipes, and the number of heat pipes is configured according to the partition area of ​​S3. The heat dissipation end is equipped with aluminum fins and a PWM-controlled fan. The humidity-heat dissipation linkage mechanism is realized through fuzzy control: when the humidity sensor reading of S1 exceeds 65%RH, the fan speed is increased to 80% of the rated value, and the micro heating wires on the fin surface are turned on to prevent frost formation. The working time of the heating wires is positively correlated with the humidity value.

6. The temperature control method for a food preservation box as described in claim 1, characterized in that: In step S5, the phase change energy storage module consists of three independent units, each containing 500g of paraffin-based phase change material with a phase change temperature of 5°C and a latent heat of 205kJ / kg. These units are coupled to the evaporator / condenser of S4 via a copper heat-conducting plate. The charge / discharge thresholds are set as follows: when the compressor power of S4 exceeds 35W for 30 seconds or the PTC heating element power exceeds 25W, energy storage release is triggered; when the power is below 15W for 1 minute, energy storage charging is initiated. Energy allocation uses a greedy algorithm, prioritizing the energy storage needs of the core area, followed by the transition and edge areas. The charge / discharge status of each unit is monitored in real-time by an NTC temperature sensor. When the phase change material is detected to be completely solidified or completely melted, the adjacent unit is automatically switched to operation.

7. The temperature control method for a food preservation box as described in claim 1, characterized in that: In step S6, a multi-source time-series dataset is first constructed. The sensor data calibrated in S1, the partition temperature control command in S3, the heat exchange power fluctuation value in S4, and the phase change energy storage module charge and discharge status data in S5 are aligned by timestamps to form an input matrix of dimension . The LSTM neural network adopts a two-layer architecture. The first hidden layer extracts short-term dependency features through a gating mechanism, the second layer captures long-term trends, and the output layer is activated by softmax to obtain the probability distribution of meal pick-up for each hour in the next 24 hours. The model training uses a sliding window method, and the parameters are automatically updated at 3:00 AM every day. At the same time, user ratings of S8 are introduced as a supervision signal. When the prediction accuracy is below 85% for three consecutive days in a certain period, the feature importance is reordered to strengthen the weights of key variables such as heat exchange efficiency of S4 and energy storage release speed of S5. During the predictive temperature control execution phase, the system marks the top-2 high-probability periods in the probability distribution output by the LSTM as the "predicted meal pickup window." Combining this with the food parameter database in S2, it calculates the pre-cooling / preheating start time. If the model predicts that users have the highest probability of picking up their meals between 12:00 and 13:00, and the optimal temperature matched by S2 is 5℃, then a "zone temperature pre-adjustment instruction" is sent to S3: the target temperature of frequently used areas is gradually reduced from the current 7℃ to 5℃, and the cooling rate is dynamically adjusted according to the current heat dissipation efficiency of S4. At the same time, the prediction results update the zone priority matrix of S3 through the feedback channel, improving the temperature control accuracy of high-frequency meal pickup areas from ±0.5℃ to ±0.3℃, and reducing the energy consumption allocation weight of low-usage areas. The formula for predicting the probability of meal pickup is: In the formula: P(t) represents the probability of picking up the meal in the t-th hour in the future; h t This represents the hidden layer output vector of the LSTM network at time step t; W o Representation: Output layer weight matrix; b o Indicates the output layer bias term; U t This represents the user habit factor, calculated from S8's historical ratings: Where si is the i-th rating, ti is the corresponding meal pick-up time, and δ(·) is the Dirac function. α: Habitual factor weight; The formula for calculating the precooling / preheating start-up time is: In the formula: T start Indicates the pre-adjusted start time; t peak Indicates the predicted start time of the peak meal pickup period; T target This indicates the optimal temperature for S2 matching; T current This indicates the real-time temperature of the current area; r max This indicates the maximum cooling / heating rate of the S4 heat exchange system; Δt min This indicates the minimum pre-adjustment duration.

8. The temperature control method for a food preservation box as described in claim 1, characterized in that: In step S7, deviation monitoring uses a sliding window comparison method with a window length of 5 minutes. The root mean square error (RMSE) between the actual temperature sequence and the predicted sequence in S6 is calculated. When RMSE > 2℃, the error correction process is activated immediately. In emergency temperature control mode, the system cuts off the output of the energy storage module in S5 and switches the main circuit of S4 to "full speed operation": the compressor speed is increased to 110% of the rated value, the heating element duty cycle is set to 100%, and the secondary heat pipe system fan is locked at its maximum speed. The warning information is transmitted to the user terminal via the LoRaWAN protocol. The information includes the deviation value, the current temperature of each zone, and the operating status code of S4. If no user confirmation instruction is received within 15 minutes, the backup power supply is automatically started to maintain the core area temperature.

9. The temperature control method for a food preservation box as described in claim 1, characterized in that: In step S8, users submit their ratings on a 5-point scale via the APP interface. The system stores the rating data according to three dimensions: "preservation effect - category - environmental parameters," where 1 point corresponds to "severe spoilage" and 5 points correspond to "completely fresh." The rating weights use a time decay function: 0.6 for ratings within the last 7 days, 0.3 for ratings from 8 to 30 days, and 0.1 for ratings over 30 days. The weight values ​​are used in real time to train the LSTM model in S6. Parameter iteration is performed every Sunday at 3:00 AM, updating the partition control algorithm in S3 and the database parameters in S2 using gradient descent. If the model loss function is detected to increase by more than 5% during the iteration process, it will automatically roll back to the previous version of the parameters and record the anomaly log.

10. A temperature control system for a food preservation container, characterized in that: The system is applied to the temperature control method for food preservation boxes as described in any one of claims 1 to 9.

Images