Refrigerator and method for managing stored food items thereof

Abstract

The application discloses a refrigerator and a food storage management method thereof. The refrigerator is powered by a power module when a power failure occurs, and the related devices in the refrigerator are powered. The temperature of the chamber, the freshness feature and the related information are detected during the power failure. On this basis, a power failure influence vector is constructed to measure the influence of the power failure event on the food. A food attribute vector is constructed to represent the individual attribute features of the food. A freshness feature vector is constructed to represent the freshness features of the food. The power failure influence vector, the food attribute vector and the freshness feature vector are input into a network model for deep fusion. The network model directly generates decision information containing disposal suggestions, so that different foods can obtain differentiated and accurate evaluation, and users can perform targeted disposal without professional knowledge, thereby effectively improving the food storage safety and intelligent management level.

Description

Technical Field

[0001] This application relates to the field of refrigeration equipment technology, and in particular to a refrigerator and a method for managing stored food therein. Background Technology

[0002] As living standards improve, consumers are increasingly demanding safer food storage. Refrigerators, as the core equipment for household food storage, directly impact food quality through their operational stability. However, unexpected power outages can occur during refrigerator operation due to grid failures, line maintenance, or other reasons, causing the temperature in the cooling compartment to rise. During this process, food may undergo a "freezing-thawing-refreezing" process, which can damage the food's cellular structure and accelerate microbial growth.

[0003] In related technologies, after a power outage occurs and power is restored, the refrigerator only alerts the user to the power outage via an alarm. Users then have to rely on the appearance of the food to determine if it has spoiled, which poses a risk of accidental ingestion and makes it difficult to guarantee food safety.

[0004] Therefore, it is necessary to optimize the management plan for food in the refrigerator in response to power outage events. Summary of the Invention

[0005] To address the aforementioned issues, this application provides a refrigerator, comprising a cabinet, a temperature detection device, and a controller, wherein the cabinet defines a refrigeration compartment; the temperature detection device is disposed within the refrigeration compartment and is used to detect the compartment temperature of the refrigeration compartment.

[0006] The controller is electrically connected to the temperature detection device and is configured to perform the following actions: constructing a power outage impact vector, which is based on at least one of the following parameters: power outage duration, temperature rise of the refrigeration compartment during the power outage, integral of the compartment temperature over time during the power outage, highest temperature of the compartment during the power outage, and rate of change of the compartment temperature after power is restored; constructing a food attribute vector, which characterizes the individual attribute features of the food; inputting the power outage impact vector and the food attribute vector into a network model, and generating decision information through the network model, which includes at least a disposal suggestion for the food.

[0007] Thus, in the above technical solution, the impact of power outage events on food is measured by constructing a power outage impact vector, and the individual attribute vector of food is constructed to represent the individual attribute characteristics of food. The power outage impact vector and the food attribute vector are input into the network model for deep fusion. The network model is used to directly generate decision information containing disposal suggestions, so that foods with different characteristics can obtain differentiated and accurate assessments. Users can perform targeted disposal without professional knowledge, effectively improving the safety of food storage and the level of intelligent management.

[0008] In some embodiments of this application, the food attribute vector is constructed based on a first attribute and a second attribute. The first attribute includes at least one of type, storage quantity, and spatial coordinates. The second attribute includes at least one of freshness score at the time of power failure, storage date, shelf life expiration date, packaging type, freeze resistance, and thawing sensitivity.

[0009] In the above technical solution, by dividing the food attribute vector into a first attribute and a second attribute, hierarchical management of the static characteristics and dynamic freshness characteristics of food is realized. This enables the network model to conduct precise analysis based on basic information such as type and placement location, combined with biological characteristics such as freeze resistance and thawing sensitivity, further improving the accuracy of the assessment of the differentiated impact of power outage shock on different foods.

[0010] In some embodiments of this application, the refrigerator further includes a freshness detection device disposed in the refrigeration compartment for detecting the freshness information of food in the refrigeration compartment. The controller is electrically connected to the freshness detection device and configured to perform the following: constructing a freshness feature vector based on the freshness information detected by the freshness detection device, the freshness feature vector being used to characterize the freshness of the food; and inputting the power outage impact vector, the food attribute vector, and the freshness feature vector into a network model to generate decision information through the network model.

[0011] In the above technical solution, by adding a freshness detection device and constructing a freshness feature vector, the real-time status information of food is incorporated into the input dimension of the network model. This allows the decision-making information to be based not only on power outage impact and individual attributes, but also on the current freshness status of the food. This helps to improve the comprehensiveness and timeliness of risk assessment and reduce the risk of misjudgment caused by food spoiling prematurely and not being identified.

[0012] In some embodiments of this application, the freshness information includes at least one of visual information, odor information, and spectral information; the freshness detection device includes at least one of a visual module, a gas sensing module, and a spectral detection module. The visual module is used to acquire image data of the food to extract the visual information, the gas sensing module is used to detect the concentration of a target gas to obtain the odor information, the target gas includes at least one of ethylene, ammonia, hydrogen sulfide, and volatile basic nitrogen, and the spectral detection module is used to acquire the spectral information of the food.

[0013] The above technical solution employs at least one of a visual module, a gas sensing module, and a spectral detection module to collect multimodal freshness information, overcoming the limitations of a single detection method. This enables the network model to perform cross-validation by integrating image features, target gas concentrations, and spectral information, effectively improving the comprehensiveness and reliability of freshness assessment. Simultaneously, by detecting specific target gases such as ethylene, ammonia, hydrogen sulfide, and volatile basic nitrogen, it achieves accurate identification of different spoilage mechanisms, such as protein decay and vegetable aging.

[0014] In some embodiments of this application, generating decision information through the network model includes: determining the impact weight of a power outage event on different foods based on the food attribute vector and the power outage impact vector; weighting and fusing the power outage impact vector, the food attribute vector, and the freshness feature vector based on the impact weight to obtain a comprehensive feature vector; and generating decision information based on the comprehensive feature vector.

[0015] In the above technical solution, differentiated impact weights are determined based on food attribute vectors and power outage impact vectors, and the modal features are weighted and fused to achieve a deep coupling analysis of historical power outage impacts and individual food differences. This enables the network model to automatically focus on foods that are significantly affected by power outages, and even small temperature rises can obtain higher risk assessment weights, significantly improving the accuracy of decision-making information.

[0016] In some embodiments of this application, the decision information further includes a remaining shelf life confidence level and / or a risk level, wherein the remaining shelf life confidence level represents the probability that the food is safe to eat, and the risk level represents the risk of food spoilage; generating decision information through the network model includes: generating a remaining shelf life confidence level through the network model; and generating the risk level and the disposal recommendation based on the remaining shelf life confidence level.

[0017] In the above technical solution, by generating confidence levels and risk levels for remaining shelf life, the abstract model output is transformed into probabilistic indicators and hierarchical classifications that users can intuitively understand. This enables users to quickly grasp the food safety status and provides a quantitative basis for multi-level decision-making, such as prioritizing the handling of high-risk foods, thereby improving the user interaction experience.

[0018] In some embodiments of this application, generating the remaining shelf life confidence score through the network model includes: obtaining the historical remaining shelf life confidence score generated by the network model for the previous power outage event; generating the remaining shelf life confidence score corresponding to the current power outage event based on the historical remaining shelf life confidence score and the received target vector, wherein the target vector includes at least the power outage impact vector and the food attribute vector.

[0019] In the above technical solution, by introducing the confidence level of the historical remaining shelf life into the generation of the current confidence level, the correlation analysis of the cumulative effect of multiple power outage events on food quality is realized, making the risk assessment more in line with the actual spoilage pattern of food and helping to avoid safety hazards caused by the one-sidedness of a single assessment.

[0020] In some embodiments of this application, the refrigerator includes a display, and the controller is electrically connected to the display and configured to: after generating decision information through the network model, use the display to identify the placement locations of food items with different risk levels using different colors, and in response to the user's operation on the food icons, display a decision analysis report corresponding to the food item, wherein the decision analysis report records the basis for generating the decision information.

[0021] In the above technical solution, the risk level is marked by different colors on the display and users can view the decision analysis report. The complex model reasoning process is transformed into an intuitive visual interactive interface, which enables users to quickly locate high-risk foods and understand the basis for decision-making, thus helping to enhance users' trust and acceptance of the decision.

[0022] In some embodiments of this application, the controller is configured to: record user feedback on the decision information and adjust the network model based on the feedback information; the controller is also configured to: upload the network model to the cloud and adjust the network model through a federated learning mechanism.

[0023] In the above technical solution, by recording user feedback to adjust the network model or by using a federated learning mechanism to aggregate the experience of multiple users, the model achieves continuous optimization and self-evolution. It can adapt to the usage habits of different users and the performance differences of different refrigerators, becoming more accurate with use. At the same time, the federated learning mechanism improves the generalization ability of the model while protecting user privacy.

[0024] This application also provides a method for managing stored food in a refrigerator, wherein a temperature detection device is installed in the refrigeration compartment of the refrigerator to detect the compartment temperature; the method includes: constructing a power outage impact vector, the power outage impact vector being constructed based on at least one of the following parameters: power outage duration, temperature rise of the refrigeration compartment during the power outage, integral of the compartment temperature over time during the power outage, maximum value of the compartment temperature during the power outage, and rate of change of the compartment temperature after power is restored; constructing a food attribute vector, the food attribute vector being used to characterize the individual attribute features of the food; inputting the power outage impact vector and the food attribute vector into a network model, and generating decision information through the network model, the decision information including at least a disposal suggestion for the food.

[0025] The above technical solution integrates multi-dimensional information such as the power outage impact vector and food attribute vector, and uses a network model to generate end-to-end disposal suggestions, thereby realizing intelligent food management in the refrigerator. It eliminates the need for manual intervention and reduces the user's decision-making cost. At the same time, it avoids extreme situations such as excessive discarding or accidental consumption of spoiled food through accurate assessment.

[0026] It should be understood that the above general description and the following detailed description are merely exemplary and do not limit this application. Attached Figure Description

[0027] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the specification, serve to explain the principles of this application.

[0028] Figure 1 A schematic diagram of a refrigerator according to one embodiment of this application is shown.

[0029] Figure 2 A partial structural block diagram of a refrigerator according to an embodiment of this application is shown.

[0030] Figure 3 A flowchart of a food storage management method for a refrigerator according to an embodiment of this application is shown.

[0031] Figure 4 It shows Figure 3 The flowchart showing a detailed embodiment of step S340 is shown.

[0032] Figure 5 It shows Figure 3 A detailed flowchart of another embodiment of step S340 is shown.

[0033] Figure 6 It shows Figure 5 The flowchart showing a detailed embodiment of step S510 is shown.

[0034] Figure 7 A flowchart of a method for managing stored food in a refrigerator, according to another embodiment of this application, is shown.

[0035] The annotations in the attached figures are explained as follows: 10. Cabinet; 11. Refrigeration compartment; 20. Temperature detection device; 30. Freshness detection device; 31. Vision module; 32. Gas sensing module; 40. Power supply; 50. Controller; 60. Display. Detailed Implementation

[0036] To make the objectives, implementation methods and advantages of this application clearer, the exemplary implementation methods of this application will be clearly and completely described below with reference to the accompanying drawings of the exemplary embodiments of this application. Obviously, the described exemplary embodiments are only some embodiments of this application, and not all embodiments.

[0037] It should be noted that the brief descriptions of terms in this application are only for the convenience of understanding the embodiments described below, and are not intended to limit the embodiments of this application. Unless otherwise stated, these terms should be understood in their ordinary and common meaning.

[0038] In the description of this application, it should be understood that the terms "inner", "outer", "top", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.

[0039] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0040] In the description of this application, it should be noted that, unless otherwise explicitly specified and limited, the terms "set" and "connection" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0041] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0042] There are significant bottlenecks in the technology for handling refrigerator power outages, mainly in the following three aspects: First, the recording of physical parameters is incomplete. Although some refrigerators are equipped with power outage recording functions that can record the time and duration of power outages, or issue alarms through indicator lights and buzzers, they only output raw physical parameters such as "temperature-time" or simple alarm signals. They cannot perform differentiated analysis based on dimensions such as temperature fluctuation range and recovery rate, and they do not consider the different tolerances of different foods to temperature changes. As a result, users still need to judge the state of food themselves based solely on the "power outage" prompt, which poses a risk of accidental ingestion or excessive disposal. Second, the decision-making value is insufficient. The relevant solutions output intermediate data such as "temperature rise of X degrees" or "freshness score Y points," which users need to interpret and judge again, failing to provide direct and actionable handling suggestions. Third, the cumulative effect is lacking. For multiple minor power outage events occurring in a short period of time, the relevant technologies lack the ability to correlate and cumulatively analyze them, and cannot assess the cumulative damage to food quality caused by gradual thermal shock, which can easily lead users to underestimate the risk of spoilage.

[0043] In view of this, this application designs a management scheme for optimizing food in a refrigerator in response to power outage events. It measures the impact of power outage events on food by constructing a power outage impact vector and constructs a food attribute vector to represent the individual attribute characteristics of food. The power outage impact vector and the food attribute vector are input into a network model for deep fusion. The network model directly generates decision information containing disposal suggestions, enabling foods with different characteristics to obtain differentiated and accurate assessments. Users can perform targeted disposal without professional knowledge, effectively improving food storage safety and intelligent management level.

[0044] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0045] Figure 1 A schematic diagram of a refrigerator according to an embodiment of this application is shown. Figure 2 A partial structural block diagram of a refrigerator according to an embodiment of this application is shown.

[0046] like Figure 1 As shown, the refrigerator of this embodiment includes a cabinet 10, which serves as the supporting structure of the refrigerator and has an internal accommodating space (not shown in the figure). The accommodating space of the cabinet 10 may house other components of the refrigerator, such as a refrigeration system and a circuit structure. The external shape of the cabinet 10 can be designed as needed, for example, it can be a hollow cuboid shape.

[0047] The interior of the cabinet 10 defines a refrigeration compartment 11, which provides space for storing items. The refrigeration compartment 11 can be a freezer, a refrigerator, or a variable temperature compartment.

[0048] There can be multiple refrigeration compartments 11, which may include freezer compartments and refrigerator compartments, and may further include variable temperature compartments. Multiple refrigeration compartments 11 allow for partitioned storage of items within the refrigerator, improving the user experience. The multiple refrigeration compartments 11 can be arranged along the height of the refrigerator body 10. The multiple refrigeration compartments 11 can also be arranged along the width of the refrigerator body 10.

[0049] The refrigerator of this embodiment also includes a refrigeration system for providing cold air to the refrigeration compartment 11. The refrigeration system includes a compressor, a condenser, a throttling device, an evaporator, and refrigeration piping. The compressor, condenser, throttling device, and evaporator are connected via the refrigeration piping. When the compressor operates, low-temperature, low-pressure refrigerant is drawn into the compressor and compressed into high-temperature, high-pressure superheated gas within the compressor cylinder before being discharged into the condenser. The high-temperature, high-pressure refrigerant gas dissipates heat through the condenser, its temperature continuously decreasing until it is gradually cooled into room-temperature, high-pressure saturated vapor, and further cooled into a saturated liquid. The pressure of the refrigerant remains almost constant throughout the condensation process. The throttling device throttles the high-pressure room-temperature liquid into low-temperature, low-pressure wet vapor, creating conditions for efficient heat absorption in the evaporator. The low-temperature, low-pressure refrigerant returns to the compressor after passing through the evaporator. This process is repeated, allowing the evaporator to continuously provide cold air to the refrigeration compartment 11, thereby maintaining the refrigeration compartment 11 at a set temperature.

[0050] In some embodiments, such as Figure 1 As shown, the refrigerator also includes a temperature detection device 20, which is installed in the refrigeration compartment 11 and is used to detect the temperature of the refrigeration compartment 11, that is, the compartment temperature.

[0051] The temperature detection device 20 may include multiple temperature sensors. Multiple temperature sensors may be installed in a refrigeration chamber 11, distributed at different locations within the refrigeration chamber 11, to achieve high-precision, multi-point temperature acquisition of the chamber. When multiple temperature sensors are installed in a refrigeration chamber 11, the chamber temperature of the refrigeration chamber 11 is obtained based on the detection results of the multiple temperature sensors. For example, the average value of the detection results of the multiple temperature sensors may be taken as the chamber temperature, or the median value of the detection results of the multiple temperature sensors may be taken as the chamber temperature.

[0052] Each refrigeration chamber 11 may also be equipped with only one temperature sensor. When only one temperature sensor is installed in a refrigeration chamber 11, the chamber temperature of the refrigeration chamber 11 is obtained based on the detection result of the one temperature sensor. That is, the detection result of the one temperature sensor is used as the chamber temperature of the refrigeration chamber 11.

[0053] In some embodiments, such as Figure 2 As shown, the refrigerator also includes a freshness detection device 30, which is installed in the refrigeration compartment 11 and is used to detect the freshness information of the food in the refrigeration compartment 11.

[0054] The freshness detection device 30 includes at least one of a vision module, a gas sensing module, and a spectral detection module. Correspondingly, the freshness information includes at least one of visual information, odor information, and spectral information. The vision module is used to acquire image data of the food to extract visual information, the gas sensing module is used to detect the concentration of the target gas to obtain odor information, and the spectral detection module is used to acquire the spectral information of the food.

[0055] In some embodiments, such as Figure 1 As shown, the freshness detection device 30 includes a vision module 31 and a gas sensing module 32.

[0056] The vision module 31 can be located in the top area of ​​the refrigeration chamber 11, such as the center of the top wall or the inner top corner, to obtain a wide-angle field of view covering the entire refrigeration chamber 11. The vision module 31 is configured to acquire real-time image data of the food and extract multi-dimensional visual information from it using image processing algorithms.

[0057] Visual information can include color and morphological information. Color information can be obtained by converting the acquired image data from the RGB color space to the HSV color space for analysis. Compared to the RGB model, the HSV model better reflects the intuitive attributes of object color and effectively filters out interference caused by changes in light intensity. This model can accurately quantify color shifts on the surface of food, such as the spoilage characteristics of meat changing from bright red to brown or mold spots. Morphological information includes, but is not limited to, the rate of change of the food's geometric contour (i.e., shrinkage rate) and the degree of juice seepage. By comparing the current edge contour of the food in the image data with its initial template, the volume shrinkage ratio is calculated; simultaneously, color segmentation algorithms are used to identify whether there are abnormally high humidity reflective areas or liquid residues on or around the food surface, thereby determining whether there is juice loss or spoilage.

[0058] Continue to refer to Figure 1The gas sensing module 32 is positioned along the airflow path within the refrigeration chamber 11, such as near the fan outlet or return air inlet, to ensure rapid detection of changes in the gas composition within the chamber. This gas sensing module 32 is configured to monitor and quantitatively detect the concentration values ​​of various characteristic putrefactive gases, i.e., target gases, in real time.

[0059] Specifically, the target gases may include indicative gases highly correlated with food spoilage, such as ethylene, ammonia, hydrogen sulfide, and volatile basic nitrogen. The gas sensing module 32 integrates multiple sensor elements with specific responses to different gases, such as metal oxide semiconductor sensors and electrochemical sensors, to simultaneously collect instantaneous concentration data of each of the aforementioned target gases. These concentration data reflect the overall degree of spoilage within the storage room.

[0060] Understandably, the freshness detection device 30, which includes a vision module 31 and a gas sensing module 32, is only one embodiment of this application. In other embodiments, the freshness detection device 30 may include any one or any combination of a vision module, a gas sensing module, and a spectral detection module.

[0061] In some embodiments, such as Figure 1 As shown, the refrigerator also includes a power supply unit 40, such as a rechargeable battery or a supercapacitor module. The power supply unit 40 is used to power relevant devices in the refrigerator during power outages, enabling the detection of compartment temperature, freshness characteristics, and recording of relevant information during power outages.

[0062] Specifically, the power supply unit 40 can be coupled to the refrigerator's main power supply circuit and configured to automatically and seamlessly switch power during unexpected external power grid outages to supply power to the temperature detection device 20, freshness detection device 30, and controller 50. This ensures that compartment temperature and freshness information can be continuously collected and key timestamps and physical parameters can be recorded during power outages, preventing data loss or monitoring blind spots due to power interruptions. When the refrigerator is normally connected to the power grid, the power supply unit 40 can automatically switch to charging mode, charging itself by the main power supply to maintain a fully charged or high-charge state for use in the event of the next sudden power outage.

[0063] In some embodiments, such as Figure 2 As shown, the refrigerator also includes a display 60. The display 60 may be embedded in a prominent position on the outside or inside of the refrigerator door, and is communicatively connected to the controller 50, configured to visualize the generated decision information.

[0064] Specifically, the display 60 employs a hierarchical visual coding mechanism combined with interactive detail display in its interface presentation. When displaying a thumbnail of the refrigerator compartment layout, the display 60 uses a red-yellow-green three-color warning light system to mark the food stored in each location. Red represents high risk, yellow represents medium risk, and green represents low risk or no risk. This design allows users to intuitively and quickly identify potential food safety hazards through color mapping without opening the refrigerator door. The display 60 also responds to user touch or click operations on specific food icons, further displaying a decision analysis report for that food in a pop-up window. This decision analysis report not only includes final handling suggestions, such as "discard immediately" or "consume within 24 hours," but can also selectively display auxiliary decision-making evidence, such as temperature rise curves during power outages, morphological changes detected by the visual module 31, and the concentration of putrefactive gases detected by the gas sensing module 32.

[0065] like Figure 2 As shown, the refrigerator also includes a controller 50. The controller 50 can be the refrigerator's original main control board, or it can be an independent control module added to implement this application. The controller 50, in its hardware architecture, includes at least a processor, non-volatile memory such as Flash or EEPROM, and an integrated AI inference engine, such as dedicated circuitry based on NPU or GPU acceleration.

[0066] The controller 50 achieves closed-loop control of the refrigerator's entire operating condition through the following electrical connections and signal interaction mechanisms: The controller 50 is electrically connected to the power supply unit 40. In the event of an unexpected power grid interruption, the power supply unit 40 provides uninterrupted backup power to the controller 50, ensuring that the controller 50 can maintain minimum operational and recording functions during power outages, preventing data loss or program crashes. The controller 50 is electrically connected to the temperature detection device 20 to obtain accurate temperature distribution information within the refrigeration compartment 11 in real time. The controller 50 is electrically connected to the vision module 31 and the gas sensing module 32, receiving food shape / color information extracted by the vision module 31 and target gas concentration data detected by the gas sensing module 32. The controller 50 is electrically connected to the display 60. Decision information generated by the AI ​​inference engine, such as handling suggestions and risk levels, is processed by the controller 50 and transmitted to the display 60 for visualization rendering. The controller 50 is also connected to core electronic control components of the refrigeration system, such as the compressor. Based on the deviation between the current compartment temperature and the target set value, the controller 50 generates a corresponding PWM (Pulse Width Modulation) control signal and sends it to the compressor drive circuit to control the compressor's start-up, shutdown, or frequency conversion adjustment, thereby dynamically maintaining the temperature in the refrigeration compartment 11 within the preset range. Furthermore, the controller 50 can be extended to connect to other electronically controlled loads within the refrigerator, such as the fan, damper, and ice maker, to coordinate the complex operating logic of the entire refrigerator.

[0067] Controller 50 is configured to execute stored food management methods, such as Figure 3 As shown, the food storage management method includes at least steps S310 to S340, which are described in detail below.

[0068] In step S310, the power outage impact vector is constructed.

[0069] When the power grid supply is normal, the controller 50 controls the temperature detection device 20 to collect the temperature of each room at a preset sampling frequency, such as once per minute, and stores the collected data in a non-volatile memory.

[0070] When a power outage signal is detected, the power supply unit 40 automatically switches to power the temperature detection device 20, the freshness detection device 30, and the controller 50. The controller 50 records the power outage time T_off and starts a power outage timer. During the power outage, the temperature detection device 20 continuously collects the room temperature, and the controller 50 writes the temperature data to a non-volatile memory in real time.

[0071] When a power restoration signal is detected, the controller 50 records the power restoration time T_on, stops the power outage timer, reads the temperature data during the power outage from the non-volatile memory, and further constructs the power outage impact vector V_outage.

[0072] In some embodiments, the power outage impact vector is constructed based on at least one of the following parameters: power outage duration, temperature rise of the cooling compartment during the power outage, integral of compartment temperature over time during the power outage, maximum value of compartment temperature during the power outage, and rate of change of compartment temperature after power is restored.

[0073] The power outage duration is the time difference from the moment the power grid was de-energized (T_off) to the moment the power was restored (T_on), i.e., power outage duration D = T_on - T_off. The power outage duration reflects the cumulative time that food has experienced an uncontrolled temperature environment.

[0074] The temperature rise of the refrigeration compartment during a power outage is the difference between the compartment temperature Temp_on,i at the moment power is restored and the compartment temperature Temp_off,i at the moment of power failure. That is, the temperature rise ΔT_i during a power outage is equal to Temp_on,i - Temp_off,i, where i represents the i-th refrigeration compartment, ΔT_i is the temperature rise of the i-th refrigeration compartment, Temp_off,i is the compartment temperature of the i-th compartment at the moment of power failure, and Temp_on,i is the compartment temperature of the i-th compartment at the moment power is restored. The temperature rise of the refrigeration compartment during a power outage directly characterizes the intensity of the thermal shock and reflects the degree to which the ambient temperature affects the shelf life of food after the refrigeration system fails.

[0075] The integral of the room temperature over time during the power outage is the integral of the deviation of the real-time room temperature T(t) from the target set temperature T_target,i over time during the power outage period, which is also the integral of the room temperature over time during the power outage period. Where T_i(t) is the actual temperature of the i-th compartment at time t, and T_target,i is the target temperature of the i-th compartment. This integral characterizes the cumulative degree of deviation of the compartment temperature from the target temperature during the power outage, i.e., the comprehensive heat load. Compared to a single duration or temperature difference, this integral parameter can more comprehensively quantify the comprehensive thermal stress borne by the food throughout the entire power outage.

[0076] The highest temperature recorded in the compartment during the power outage is defined as T_max,i = max_{t∈[T_off,T_on]} T_i(t), where T_i(t) is the actual temperature of the i-th compartment at time t. This parameter characterizes the highest temperature reached in the i-th compartment during the power outage and is used to determine whether the safe storage threshold for a specific food has been exceeded.

[0077] The rate of change of the compartment temperature after power restoration is the slope at which the compartment temperature falls back to the target set range, i.e., R_i = (T_max,i - T_stable,i) / (t_stable - T_on), where T_stable,i is the actual temperature when the temperature stabilizes after power restoration, and t_stable is the moment when the temperature reaches a stable state. This parameter characterizes the recovery capability of the refrigeration system.

[0078] In some embodiments, constructing a power outage impact vector includes: normalizing the duration of the power outage, the temperature rise of the cooling room during the power outage, the integral of the room temperature over time during the power outage, the highest value of the room temperature during the power outage, and the rate of change of the room temperature after power is restored, and then constructing the power outage impact vector based on the normalized parameters.

[0079] The constructed power outage impact vector is: V_outage = [D, ΔT_1, A_1, T_max,1, R_1, ΔT_2, A_2, ..., R_n], where D is the duration of the power outage, ΔT_1 is the temperature rise of cooling room 1 during the power outage, A_1 is the integral of room temperature 1 with respect to time during the power outage, T_max,1 is the highest value of room temperature 1 during the power outage, R_1 is the rate of change of room temperature 1 after power is restored, ΔT_2 is the temperature rise of cooling room 2 during the power outage, A_2 is the integral of room temperature 2 with respect to time during the power outage, and R_n is the room temperature after power is restored.

[0080] In step S320, a food attribute vector is constructed, which is used to characterize the individual attribute features of the food.

[0081] In some embodiments, the food attribute vector is constructed based on a first attribute and a second attribute.

[0082] The first attribute is a basic, general attribute that describes the inherent information and spatial location of the food. It is typically entered manually by the user when the food is placed in the refrigerator or initially determined by a visual module. The first attribute includes at least one of the following: type, storage volume, and spatial coordinates. Type refers to the biological species it belongs to, such as seafood, leafy vegetables, pasta, or fruit. Storage volume is a volume parameter, such as 500g. Spatial coordinates are the food's placement within the refrigeration compartment, such as the upper shelf of the freezer or a drawer in the refrigerator.

[0083] The second attribute is related to freshness, which is directly related to the food's spoilage process and can be dynamically adjusted during power outages. This second attribute includes at least one of the following: freshness score at the time of power outage, storage date, expiration date, packaging type, freeze resistance, and thaw sensitivity. The freshness score at the time of power outage can be automatically generated by the visual module based on image analysis or manually entered by the user, and can be divided into 0-100 points. Packaging type can be vacuum-packed, sealed container, ordinary cling film, or unpackaged. Freeze resistance is the duration or temperature threshold that food can withstand at a set temperature, and can be divided into high, medium, and low. Thawing sensitivity characterizes the food's fragility to repeated freeze-thaw cycles, such as leafy vegetables > root vegetables > frozen meats, and can be divided into high, medium, and low.

[0084] The aforementioned first and second attribute information is encoded into a food attribute vector L_j, stored in non-volatile memory, and a mapping relationship between food and its placement location is established.

[0085] In some embodiments, constructing a food attribute vector includes: normalizing the first attribute and the second attribute, and then constructing the food attribute vector based on the normalized attribute parameters.

[0086] In some embodiments, during periods of normal power supply, the dynamic attributes of the food are continuously updated via the freshness detection device 30. Specifically, the vision module 31 periodically acquires images of the food and extracts color and morphological features, such as shrinkage rate and juice seepage degree, using a pre-trained convolutional neural network to update the freshness score. The gas sensing module 32 periodically detects the concentration of target gases around the food, establishes and updates the odor feature baseline of the food under normal conditions, and the second attribute may include this odor feature baseline. The aforementioned dynamically updated data serves as an extended dimension of the food attribute vector and participates in subsequent network model inference.

[0087] In step S330, a freshness feature vector is constructed based on the freshness information detected by the freshness detection device. The freshness feature vector is used to characterize the freshness of the food.

[0088] In some embodiments, the freshness information detected by the freshness detection device includes visual information and odor information. The visual information may include color information and morphological information, and the odor information includes the concentration of the target gas. Accordingly, the freshness feature vector includes a visual feature vector and a gas feature vector. Constructing the freshness feature vector based on the freshness information detected by the freshness detection device includes: constructing a visual feature vector based on the visual information and constructing a gas feature vector based on the odor information.

[0089] In step S340, the power outage impact vector, food attribute vector, and freshness feature vector are input into the network model, and decision information is generated through the network model. The decision information includes at least food disposal recommendations.

[0090] exist Figure 3 In the illustrated embodiment, a power module supplies power to relevant devices in the refrigerator during a power outage, enabling the detection of compartment temperature, freshness characteristics, and recording of relevant information during the outage. Based on this, a power outage impact vector is constructed to measure the impact of the power outage on the food. A food attribute vector is constructed to represent the individual attribute characteristics of the food, and a freshness feature vector is constructed to represent the freshness characteristics of the food. The power outage impact vector, food attribute vector, and freshness feature vector are input into a network model for deep fusion. The network model directly generates decision information containing disposal suggestions, enabling differentiated and accurate assessments of foods with different characteristics. Users can perform targeted disposal without professional knowledge, effectively improving food storage safety and intelligent management.

[0091] In some embodiments, the freshness feature vector includes a visual feature vector and a gas feature vector. That is, the power outage impact vector, food attribute vector, visual feature vector, and gas feature vector are used as inputs to the network model. The power outage impact vector is denoted as V_outage, which quantifies the degree of disturbance to the physical environment of the refrigerator caused by the power outage event. As mentioned above, the power outage impact vector is constructed from multi-dimensional parameters such as the duration of the power outage, the temperature rise of the compartment, and the temperature-time integral, characterizing the intensity of thermal shock suffered by each refrigeration compartment during the power outage. The food attribute vector is denoted as L_j, which characterizes the individual attributes of the j-th food item, integrating its basic general attributes and freshness-related attributes. The visual feature vector is denoted as I_j, which is extracted from the image data collected by the vision module 31 using a convolutional neural network (CNN). Preferably, a pre-trained deep network such as ResNet-50 or VGG-16 can be used to extract deep semantic features of the image, including but not limited to color histograms, texture features, and morphological changes, thereby quantifying the apparent freshness of the food. The odor feature vector, denoted as G_j, is derived from the time-series data of the target gas concentration detected by the gas sensing module. This time-series data is then processed by a one-dimensional convolutional neural network (1D-CNN) to extract features from the odor sequence, capturing the evolution trend and peak characteristics of the gas concentration over time. The odor feature vector G_j and the visual feature vector I_j together constitute the freshness feature vector F_j.

[0092] The network model performs deep coupling analysis on the input power outage impact vector V_outage, food attribute vector L_j, odor feature vector G_j, and visual feature vector I_j to generate decision information.

[0093] In some embodiments, such as Figure 4 As shown, the steps for generating decision information through a network model include at least the following steps S410 to S430, which are described in detail below.

[0094] In step S410, the impact weights of the power outage event on different foods are determined based on the food attribute vector and the power outage impact vector.

[0095] In step S410, the model introduces a cross-modal attention mechanism to quantify the differentiated impact intensity of the power outage event on different foods in the refrigerator. By performing a nonlinear mapping between the food attribute vector L_j (including freeze resistance, thawing sensitivity, etc.) and the power outage impact vector V_outage (including temperature rise amplitude, duration, etc.), the personalized impact weight aj for the j-th food is calculated.

[0096] Specifically, for foods with poor freeze resistance, such as leafy vegetables and seafood, even a slight temperature rise will be assigned a higher impact weight value; while for foods with strong freeze resistance or those nearing their shelf life, their weight coefficients will be adjusted accordingly.

[0097] In step S420, the power outage impact vector, food attribute vector, and freshness feature vector are weighted and fused based on the impact weights to obtain a comprehensive feature vector.

[0098] After obtaining the differential impact weights aj, the input data is no longer simply concatenated; instead, a weighted fusion operation is performed. Specifically, the impact weights aj are applied to the power outage impact vector V_outage, the food attribute vector L_j, the odor feature vector G_j, and the visual feature vector I_j, respectively, so that the features with higher weights dominate in subsequent calculations.

[0099] Subsequently, the weighted data is integrated into a high-dimensional comprehensive feature vector Zj. This vector not only includes the physical environment history and current biochemical state of the food, but also embeds the model's subjective assessment of the risk of specific foods, providing a data foundation for the final accurate decision-making.

[0100] In some embodiments, the impact weights of a power outage event on different foods are determined based on food attribute vectors and power outage impact vectors, including: based on the function αj=Softmax(Wa [V_outage;L_j]) determines the impact weights of the power outage event on different foods; where αj is the impact weight of the power outage event on the j-th food, Wa is the learnable attention weight matrix, V_outage is the power outage impact vector, L_j is the food attribute vector of the j-th food, and [;] indicates vector concatenation operation.

[0101] The Softmax attention mechanism is used to calculate the weights of the differential impact, enabling the network model to adaptively learn the different responses of different foods to the same power outage impact. This eliminates the need for manually preset fixed thresholds, automates and automates the weight allocation, and enhances the model's generalization ability.

[0102] In step S430, decision information is generated based on the comprehensive feature vector.

[0103] In step S430, decision information is generated based on the comprehensive feature vector. This can be achieved by using a preset classification or regression algorithm to map the feature vector into human-understandable decision information.

[0104] exist Figure 4 In the embodiment shown, by determining the differentiated impact weights based on the food attribute vector and the power outage impact vector, and by weighting and fusing the features of each modality, a deep coupling analysis of the historical impact of power outages and individual differences in food is achieved. This enables the network model to automatically focus on foods that are significantly affected by power outages, and to obtain higher risk assessment weights even with small temperature rises, thus significantly improving the accuracy of decision-making information.

[0105] In some embodiments, the decision information further includes a remaining shelf life confidence level and / or a risk level, wherein the remaining shelf life confidence level represents the probability that the food is safe to eat, and the risk level represents the risk of food spoilage; the decision information is generated through a network model, including: generating a remaining shelf life confidence level through a network model; and generating a risk level and disposal recommendations based on the remaining shelf life confidence level.

[0106] In some embodiments, such as Figure 5 As shown, the steps for generating decision information through a network model include at least the following steps S510 to S520, which are described in detail below.

[0107] In step S510, the remaining shelf life confidence score is generated using a network model.

[0108] The remaining shelf life confidence score can be generated by using a network model. This can be achieved by using a network model based on a comprehensive feature vector Zj and outputting a continuous quantitative value, namely the remaining shelf life confidence score P, through a regression algorithm or a fully connected layer.

[0109] The confidence score P for the remaining shelf life can range from 0 to 100 or be a probability value of 0 to 1. It represents the network model's comprehensive safety assessment of food after a power outage. It integrates not only the physical heat load during the power outage, such as temperature-time integrals, but also visually recognized changes in appearance, such as juice seepage, and the concentration of spoilage gases detected by gas sensors. This multi-dimensional comprehensive evaluation result characterizes the likelihood that the food is still safe to eat.

[0110] In step S520, a risk level and disposal recommendations are generated based on the confidence level of the remaining shelf life.

[0111] After obtaining the quantified confidence level P of the remaining shelf life, it is further transformed into two types of decision labels that users can intuitively understand through preset mapping rules or classifiers: risk level and disposal recommendations.

[0112] In some embodiments, the risk levels include high risk, medium risk, low risk, and no risk. In step S520, based on the different threshold ranges in which the remaining shelf life confidence level P falls, it is qualitatively classified into high risk, medium risk, low risk, or no risk. For example, when the remaining shelf life confidence level P is lower than a certain safety threshold, such as 30%, it is classified as high risk.

[0113] In some embodiments, disposal recommendations include immediate disposal, consumption within a preset time, continued storage, and consumption after cooking. In step S520, a specific action instruction can be generated by calling a pre-set decision knowledge base based on the risk level and the specific remaining shelf life confidence level P. For example, for food classified as high-risk and with obvious spoilage characteristics detected by the visual module, an immediate disposal recommendation is generated; for food classified as medium-risk and with acceptable freeze resistance, a disposal recommendation is output that it should be cooked and consumed today and should not be refrozen.

[0114] By generating confidence levels and risk levels for remaining shelf life, the abstract model output is transformed into probabilistic indicators and hierarchical classifications that users can intuitively understand. This enables users to quickly grasp the food safety status and provides a quantitative basis for multi-level decision-making, such as prioritizing the handling of high-risk foods, thereby enhancing the user interaction experience.

[0115] To address the limitation of existing technologies in assessing the cumulative effects of multiple short-term power outages, the network model incorporates a time-series analysis mechanism with memory capabilities. In some embodiments, for food products that have experienced multiple power outages, the historical remaining shelf life confidence score generated from the previous power outage is obtained. This score, along with the current power outage impact vector, food attribute vector, food attribute vector, freshness feature vector, and freshness feature vector, is input into the network model to generate the remaining shelf life confidence score corresponding to the current power outage, thus achieving a correlation analysis of the cumulative effect.

[0116] In some embodiments, such as Figure 6 As shown, the steps for generating the confidence level of the remaining shelf life through the network model include at least the following steps S610 to S620, which are described in detail below.

[0117] In step S610, the confidence level of the historical remaining shelf life generated by the network model for the previous power outage event is obtained.

[0118] In step S610, in response to the occurrence of this power outage event, the historical state data stored in the non-volatile memory is first accessed. Specifically, the historical remaining shelf life confidence score for the j-th food item, output by the network model at the end of the previous power outage event, is read. This historical remaining shelf life confidence score represents the residual freshness level of the food item before the current power outage event.

[0119] In step S620, the remaining shelf life confidence level corresponding to this power outage event is generated based on the historical remaining shelf life confidence level, the power outage impact vector, the food attribute vector, and the freshness feature vector.

[0120] That is, the historical remaining shelf life confidence score is used as a time-series input feature and jointly processed with the power outage impact vector, food attribute vector and freshness feature vector of the current power outage event to generate the remaining shelf life confidence score corresponding to the current power outage event.

[0121] The structure and training process of the network model will be explained below.

[0122] The network model consists of an input layer, a parallel feature extraction layer, a cross-modal attention fusion layer, and an output layer.

[0123] The input layer receives four types of data as input: the power outage impact vector V_outage, the food attribute vector L_j, the odor feature vector G_j, and the visual feature vector I_j.

[0124] Parallel feature extraction layers process each input vector separately. The power outage impact vector V_outage and the food attribute vector L_j are processed by fully connected layers; the visual feature vector I_j is processed by a pre-trained ResNet-50 network to extract deep features; and the odor feature vector G_j is processed by a 1D-CNN to extract temporal variation features.

[0125] The cross-modal attention fusion layer determines the differentiated impact weights of a power outage event on different foods based on the food attribute vector L_j and the power outage impact vector V_outage. Specifically, for the j-th food: αj = Softmax(Wa [V_outage;L_j]); where αj is the weight of the power outage event on the j-th food, Wa is the learnable attention weight matrix, V_outage is the power outage impact vector, L_j is the food attribute vector of the j-th food, and [;] indicates vector concatenation operation.

[0126] This mechanism enables the network model to learn automatically that foods with low freeze resistance and high thawing sensitivity (such as salmon and broccoli) receive higher influence weights even with small temperature rises, while foods with high freeze resistance (such as apples and root vegetables) receive lower influence weights even with large temperature rises.

[0127] Based on the differential influence weight α_j, the feature vectors of each modality are weighted and fused to obtain the comprehensive feature vector H_j: H_j = α_j · (FC(V_outage) + FC(L_j) + CNN_visual(I_j) + CNN_gas(G_j)); where FC represents a fully connected layer, and CNN_visual and CNN_gas represent the visual and odor feature extraction networks, respectively.

[0128] The output layer generates the following decision information based on the comprehensive feature vector H_j: remaining shelf life confidence score P_j (0-100 points), representing the probability that the food is still safe to eat in its current state; risk level, based on P_j: P_j<30 is high risk, 30≤P_j<60 is medium risk, 60≤P_j<85 is low risk, and P_j≥85 is no risk; disposal suggestion, directly mapped from the risk level: if the risk level is high, the disposal suggestion is to discard immediately; if the risk level is medium risk, the disposal suggestion is to consume within 24 hours; if the risk level is low risk, the disposal suggestion is to continue freezing; and if the risk level is no risk, the disposal suggestion is to store normally.

[0129] In some embodiments, the output layer generates a list of treatment recommendations sorted in descending order of risk level, the list containing decision information for each food item. For example, the list of treatment recommendations is as follows: High risk (act immediately) Frozen salmon: Temperature rose to -2°C for 4 hours during the power outage. A1 assessment showed: surface juice seepage and abnormal TVB-N concentration. Recommendation: Discard immediately; do not consume.

[0130] Medium risk (consume as soon as possible) Frozen Dumplings (Freezer Compartment): During the power outage, the temperature rose to -5℃. AI assessment showed that the stability of the dough and filling structure was acceptable. Recommendation: Cook and consume within one day; do not refrozen.

[0131] Low risk (can continue to be stored) Apples in the refrigerator compartment: During the power outage, the temperature rose to 10°C. AI assessment shows that the temperature rise has minimal impact on the apples. Recommendation: Normal storage is possible.

[0132] Regarding network model training, a three-level progressive strategy of pre-training, personalized fine-tuning, and federated aggregation is adopted to adapt to the usage habits of different users and the performance differences of different refrigerators, while taking into account both data privacy and the improvement of model performance.

[0133] Specifically, in the initial deployment of the network model, the basic model is first pre-trained using massive public datasets. These datasets contain spoilage time-series data of various foods under different temperature conditions. By learning from rich and diverse general data, the model can establish a basic understanding of food preservation patterns, temperature sensitivity, and common spoilage characteristics, and possess preliminary decision-making capabilities.

[0134] As users begin actual use of the refrigerator, the process enters a personalized fine-tuning phase to adapt to their unique habits. User feedback on model decisions is recorded, such as confirmation that food has been discarded or that it was consumed normally, and this feedback is used as weakly labeled data. Subsequently, this data is continuously used for local learning, fine-tuning the model parameters to adapt to the user's storage habits, door opening and closing frequency, and the refrigerator's actual operating performance.

[0135] A federated learning mechanism is introduced, under strict user authorization, to encrypt and upload locally trained model updates—specifically, parameter increments or changes—to the cloud. The cloud server securely aggregates these model updates from different users, continuously optimizing the model while protecting privacy. Finally, the optimized network model is distributed to each refrigerator terminal.

[0136] In some embodiments, after generating decision information through a network model, the placement locations of foods with different risk levels are also indicated by different colors on a display, and a decision analysis report corresponding to the food is displayed in response to the user's operation on the food icon. The decision analysis report records the basis for generating the decision information.

[0137] The display 60 uses red, yellow, and green to indicate the placement of food items at different risk levels, with red representing high risk, yellow representing medium risk, and green representing low risk or no risk. Users can click on a food icon to view a decision analysis report, which records the basis for generating the decision information, including the parameter values ​​of the power outage impact vector, freshness test data, and the model reasoning process.

[0138] In some embodiments, decision information is also pushed to the user's mobile terminal APP via a communication module (not shown in the figure) to generate a "Power Outage Event Impact Report", so that the user can remotely monitor the refrigerator status and make arrangements in advance even when they are away from home.

[0139] Figure 7 A flowchart of a method for managing stored food in a refrigerator, according to another embodiment of this application, is shown.

[0140] like Figure 7 As shown, under normal power supply conditions, the refrigerator system uses a vision module, a gas sensing module, and a temperature detection device to perform all-weather multimodal monitoring of the refrigeration compartments, continuously updating the digital files of the food inside the refrigerator, including food attributes and freshness. Once an external power grid interruption occurs, the monitoring system powered by the power supply unit immediately activates, accurately recording the moment of power failure (T_off) and the temperature of each compartment. During the power outage, non-volatile memory is used to continuously record key temperature change trajectories, ensuring data integrity.

[0141] Once power is restored (T_on), the data analysis process begins immediately. First, the controller calculates the physical impact parameters during the power outage, including the duration of the outage, the temperature rise in each compartment, the temperature-time integral, and the highest temperature value. These parameters are then normalized to construct the outage impact vector V_outage, characterizing the environmental disturbance. Simultaneously, the food attribute vector L_j is retrieved or updated, and the current freshness feature vector F_j is extracted using the visual module and the gas sensing module.

[0142] Next, the three vectors constructed above are input into a pre-trained deep neural network model. Internally, the model first uses a cross-modal attention mechanism to calculate the differentiated impact weights of the power outage event on different foods; for example, seafood is given a higher weight. Then, the model performs weighted fusion of the feature vectors from each modality to generate a comprehensive feature vector, and finally outputs a quantified confidence level of the remaining shelf life. Based on this confidence level, a risk level that users can directly understand is further mapped and generated, such as high / medium / low, along with specific disposal recommendations, such as immediate disposal or consumption within 24 hours.

[0143] Finally, the generated decision information is pushed to the user through a multi-dimensional interactive interface. The refrigerator door display uses red, yellow, and green to intuitively identify the locations of food items with different risk levels. Users can click on specific icons to view detailed decision analysis reports, including power outage temperature rise curves and identification criteria. In addition, a "Power Outage Event Impact Report" can be pushed to users via an app for remote management. After users handle the food according to the suggestions, their feedback (such as confirming disposal) will be fed back to the system as weakly labeled data for continuous model learning and optimization, thus forming a complete closed loop from perception, analysis, decision-making to feedback.

[0144] In some embodiments, the refrigerator may not be equipped with a freshness detection device 30. By inputting the power outage impact vector and food attribute vector into the network model, the network model generates decision information based on the power outage impact vector and food attribute vector.

[0145] In an embodiment where the refrigerator is not equipped with a freshness detection device 30, the step of generating decision information through a network model may include: determining the impact weight of a power outage event on different foods based on food attribute vectors and power outage impact vectors; weighting and fusing the power outage impact vector and food attribute vector based on the impact weights to obtain a comprehensive feature vector; and generating decision information based on the comprehensive feature vector.

[0146] In embodiments where the refrigerator is not equipped with a freshness detection device 30, the step of generating decision information through a network model may include: obtaining the historical remaining shelf life confidence score generated by the network model for the previous power outage event; and generating the remaining shelf life confidence score corresponding to the current power outage event based on the historical remaining shelf life confidence score, the power outage impact vector, and the food attribute vector.

[0147] In some embodiments, the refrigerator may not have a power supply unit 40, but it may have network communication capabilities. The controller 50 uploads the time of power outage, the time of power restoration, and the corresponding temperature data to a cloud server. The cloud server simulates the temperature change curve, constructs a power outage impact vector, performs inference through a network model deployed in the cloud, and sends the decision information back to the refrigerator or user terminal.

[0148] The following example illustrates the workflow of this application, using the case of a user's refrigerator experiencing a power outage and then having its power restored after 6 hours.

[0149] Refrigerator configuration: Freezer target temperature -18℃, refrigerator target temperature 4℃. Food: Salmon (seafood, low freezing tolerance) on the upper shelf of the freezer, frozen dumplings (noodles, medium freezing tolerance) on the middle shelf of the freezer; broccoli (leafy greens, extremely low freezing tolerance) in the refrigerator drawer, and apples (fruit, high freezing tolerance) on the middle shelf of the refrigerator.

[0150] The data was collected as follows:

[0151] Food attributes (partial) are as follows:

[0152] Network model inference: Construct the power outage impact vector, obtain the attribute vectors and freshness feature vectors of each food item, and input them into the network model. The cross-modal attention fusion layer calculates the differential impact weights. Salmon: α=0.32 (highest, seafood has extremely low freeze tolerance) Broccoli: α=0.28 (second highest, leafy vegetables are sensitive to temperature) Frozen dumplings: α=0.22 (moderate, average tolerance to wheat-based foods) Apple: α=0.18 (lowest, fruit has strong tolerance to short-term temperature rise) The output decision information is as follows:

[0153] User handling: The refrigerator display showed a warning, and the app pushed a decision analysis report. Following the advice, the user discarded the salmon, chose broccoli for dinner, and ate the dumplings and apple as usual. The user reported that everything was normal after eating the food via the app; the controller recorded this feedback and fine-tuned the network model.

[0154] In summary, this application overcomes the limitations of focusing on only a single dimension, such as temperature or current state, by establishing a coupled analysis mechanism that combines the historical impact of power outages with individual differences in food products. Utilizing an attention mechanism, the model can automatically learn the differentiated impact weights of different foods on the same power outage event, thereby achieving a scientific and precise assessment of the freshness and safety of each food product, significantly improving the accuracy of risk identification.

[0155] At the decision-making level, this application directly generates actionable disposal recommendations, such as immediate disposal or consumption within the same day, along with the risk level. This significantly lowers the barrier to understanding for ordinary users, enabling even those lacking professional food safety knowledge to make correct decisions based on the guidance, effectively reducing the health risks caused by accidentally consuming spoiled food.

[0156] Furthermore, this application constructs a dual guarantee system of self-evolution and resource conservation. On the one hand, by introducing continuous learning and federated learning mechanisms, the model can be continuously optimized based on actual user feedback and usage habits, becoming more accurate with use and adapting to the performance differences of different refrigerators; on the other hand, accurate risk assessment effectively avoids two extreme situations caused by uncertainty: over-disposal and underestimation of risk.

[0157] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of this application is limited only by the appended claims.

Claims

1. A refrigerator, characterized in that, include: The enclosure defines a refrigeration chamber. A temperature detection device is installed in the refrigeration room to detect the room temperature of the refrigeration room; The controller, electrically connected to the temperature detection device, is configured to perform: Construct a power outage impact vector, which is based on at least one of the following parameters: power outage duration, temperature rise of the cooling room during the power outage, integral of the room temperature with respect to time during the power outage, maximum value of the room temperature during the power outage, and rate of change of the room temperature after power is restored. Construct a food attribute vector, which is used to characterize the individual attribute features of the food. The power outage impact vector and the food attribute vector are input into the network model, and decision information is generated through the network model. The decision information includes at least the disposal suggestions for the food.

2. The refrigerator according to claim 1, characterized in that, The food attribute vector is constructed based on a first attribute and a second attribute. The first attribute includes at least one of type, storage quantity, and spatial coordinates. The second attribute includes at least one of freshness score at the time of power failure, storage date, shelf life expiration date, packaging type, freeze resistance, and thawing sensitivity.

3. The refrigerator according to claim 1, characterized in that, Also includes: A freshness detection device is installed in the refrigeration room to detect the freshness information of food in the refrigeration room; The controller is electrically connected to the freshness detection device and is configured to perform the following: construct a freshness feature vector based on the freshness information detected by the freshness detection device, the freshness feature vector being used to characterize the freshness of the food; and input the power outage impact vector, the food attribute vector, and the freshness feature vector into a network model to generate decision information through the network model.

4. The refrigerator according to claim 3, characterized in that, The freshness information includes at least one of visual information, odor information, and spectral information; the freshness detection device includes at least one of a visual module, a gas sensing module, and a spectral detection module. The visual module is used to acquire image data of the food to extract the visual information, the gas sensing module is used to detect the concentration of a target gas to obtain the odor information, the target gas includes at least one of ethylene, ammonia, hydrogen sulfide, and volatile basic nitrogen, and the spectral detection module is used to acquire the spectral information of the food.

5. The refrigerator according to claim 3, characterized in that, Decision information is generated through the network model, including: Based on the food attribute vector and the power outage impact vector, the impact weight of the power outage event on different foods is determined. Based on the aforementioned influence weights, the power outage impact vector, the food attribute vector, and the freshness feature vector are weighted and fused to obtain a comprehensive feature vector; Decision information is generated based on the comprehensive feature vector.

6. The refrigerator according to any one of claims 1 to 5, characterized in that, The decision information also includes a confidence level of the remaining shelf life and / or a risk level, whereby the confidence level of the remaining shelf life represents the probability that the food is safe to eat, and the risk level represents the risk of food spoilage; the decision information generated through the network model includes: The remaining shelf life confidence score is generated using the network model. Based on the confidence level of the remaining shelf life, the risk level and the disposal recommendations are generated.

7. The refrigerator according to claim 6, characterized in that, The remaining shelf life confidence score is generated through the network model, including: Obtain the historical remaining shelf life confidence score generated by the network model for the last power outage event; Based on the historical remaining shelf life confidence score and the received target vector, the remaining shelf life confidence score corresponding to this power outage event is generated. The target vector includes at least the power outage impact vector and the food attribute vector.

8. The refrigerator according to claim 6, characterized in that, The refrigerator includes a display, and the controller is electrically connected to the display and is configured to: after generating decision information through the network model, use different colors to identify the placement locations of food items with different risk levels on the display, and in response to the user's operation on the food icons, display a decision analysis report corresponding to the food item, wherein the decision analysis report records the basis for generating the decision information.

9. The refrigerator according to any one of claims 1 to 5, characterized in that, The controller is configured to: record user feedback on the decision information, and adjust the network model based on the feedback information; or The controller is configured to upload the network model to the cloud and adjust the network model through a federated learning mechanism.

10. A method for managing stored food in a refrigerator, characterized in that, The refrigerator's cooling compartment is equipped with a temperature detection device, which is used to detect the compartment temperature of the cooling compartment; the food storage management method includes: Construct a power outage impact vector, which is based on at least one of the following parameters: power outage duration, temperature rise of the cooling room during the power outage, integral of the room temperature with respect to time during the power outage, maximum value of the room temperature during the power outage, and rate of change of the room temperature after power is restored. Construct a food attribute vector, which is used to characterize the individual attribute features of the food. The power outage impact vector and the food attribute vector are input into the network model, and decision information is generated through the network model. The decision information includes at least the disposal suggestions for the food.

Images