Refrigerator and humidity prediction method thereof

Abstract

This application relates to a refrigerator and a method for predicting its humidity. The refrigerator includes a cabinet; a refrigeration system; an internal sensing sensor for collecting environmental state data inside the refrigerator; an external sensing sensor for collecting external environmental data of the environment in which the refrigerator is located; the environmental state data and / or the external environmental data include branch selection reference data; at least one controller connected to the sensing sensors and configured to: acquire the environmental state data and external environmental data collected by the refrigerator during the current time period; select at least one target prediction branch from at least one humidity prediction branch in a humidity prediction network according to the branch selection reference data; input the environmental state data into each target prediction branch for humidity prediction to obtain the humidity prediction value under the corresponding target prediction branch; and determine the target humidity of the refrigerator at the current time based on each humidity prediction value. This method can improve the accuracy of humidity prediction.

Description

Technical Field

[0001] This application relates to the field of refrigerators, and in particular to a refrigerator and a method for predicting humidity therefrom. Background Technology

[0002] With the rapid development of smart home technology, refrigerators, as core devices for household food management, are constantly improving their level of intelligence. Among these improvements, accurate humidity sensing inside the refrigerator is crucial for extending the shelf life of food and maintaining a suitable storage environment.

[0003] Traditional techniques can directly use ambient temperature and humidity as model input parameters and retrain a single model using more comprehensive mixed-condition data. However, different condition data can interfere with each other during training, making it difficult for the model to converge. Ultimately, the prediction accuracy decreases instead of increasing, failing to solve the accuracy problem in high-temperature and high-humidity scenarios. Summary of the Invention

[0004] This application provides a refrigerator and a humidity prediction method thereof to improve the accuracy of humidity prediction in refrigerators.

[0005] In a first aspect, some embodiments provide a refrigerator, including:

[0006] The enclosure has at least one compartment.

[0007] The refrigeration system, located inside the refrigerator, is configured to transfer heat from the inside of the refrigerator to the outside through the circulation of refrigerant.

[0008] Internal sensing sensors are used to collect environmental status data inside the refrigerator;

[0009] External sensing sensors are used to collect external environmental data of the environment in which the refrigerator is located; the environmental status data and / or the external environmental data include branch selection reference data;

[0010] At least one controller, connected to the sensing sensor, is configured to:

[0011] Acquire environmental status data and external environment data collected by the refrigerator within the current time period;

[0012] Based on the reference data for branch selection, at least one target prediction branch is selected from at least one humidity prediction branch in the humidity prediction network;

[0013] Environmental state data are input into each target prediction branch to predict humidity, and the predicted humidity value under the corresponding target prediction branch is obtained.

[0014] Based on the predicted humidity values, determine the target humidity for the refrigerator at the current moment.

[0015] The refrigerator provided in the above embodiments can comprehensively capture multi-source information affecting the internal humidity of the refrigerator by simultaneously collecting environmental state data and external environmental data, laying a data foundation for accurate decision-making. By dynamically selecting appropriate prediction branches using branch selection reference data, the most suitable computing resources can be adaptively called according to different working conditions or scenarios, avoiding redundant overhead caused by full branch operation, thereby reducing the computing power and power consumption of real-time inference. By inputting environmental state data into multiple target prediction branches in parallel for humidity prediction, complementary prediction results from different modeling perspectives can be obtained at the same time, effectively mitigating the prediction bias of a single model under boundary conditions. By fusing the prediction values ​​of each branch to determine the target humidity, the advantages of each branch in feature representation can be combined to improve the stability and accuracy of the final output, thereby achieving higher accuracy of refrigerator internal humidity sensing capability at a lower cost without the need for physical humidity sensors.

[0016] Secondly, some embodiments also provide a refrigerator humidity prediction method, including:

[0017] Acquire environmental status data and external environment data collected by the refrigerator within the current time period; the environmental status data and / or external environment data include branch selection reference data;

[0018] Based on the reference data for branch selection, at least one target prediction branch is selected from at least one humidity prediction branch in the humidity prediction network;

[0019] Environmental state data are input into each target prediction branch to predict humidity, and the predicted humidity value under the corresponding target prediction branch is obtained.

[0020] Based on the predicted humidity values, determine the target humidity for the refrigerator at the current moment.

[0021] The refrigerator humidity prediction method provided in the above embodiments can comprehensively capture multi-source information affecting the internal humidity of the refrigerator by simultaneously collecting environmental state data and external environmental data, laying a data foundation for accurate decision-making. By dynamically selecting appropriate prediction branches using branch selection reference data, the most suitable computing resources can be adaptively called according to different working conditions or scenarios, avoiding redundant overhead caused by full branch operation, thereby reducing the computing power and power consumption of real-time inference. By inputting environmental state data into multiple target prediction branches in parallel for humidity prediction, complementary prediction results from different modeling perspectives can be obtained at the same time, effectively mitigating the prediction bias of a single model under boundary conditions. By fusing the prediction values ​​of each branch to determine the target humidity, the advantages of each branch in feature representation can be combined to improve the stability and accuracy of the final output, thereby achieving higher accuracy of refrigerator internal humidity sensing capability at a lower cost without the need for physical humidity sensors.

[0022] Thirdly, some embodiments also provide a refrigerator humidity prediction device, including:

[0023] The acquisition module is used to acquire environmental status data and external environment data collected by the refrigerator during the current time period.

[0024] The selection module is used to select at least one target prediction branch from at least one humidity prediction branch in the humidity prediction network based on reference data selected from the branch selection reference data.

[0025] The prediction module is used to input environmental state data into each target prediction branch to predict humidity and obtain the humidity prediction value under the corresponding target prediction branch.

[0026] The determination module is used to determine the target humidity of the refrigerator at the current moment based on the various humidity prediction values.

[0027] The refrigerator humidity prediction device provided in the above embodiments can comprehensively capture multi-source information affecting the internal humidity of the refrigerator by simultaneously collecting environmental state data and external environmental data, laying a data foundation for accurate decision-making. By dynamically selecting appropriate prediction branches using branch selection reference data, the device can adaptively call the most suitable computing resources according to different working conditions or scenarios, avoiding redundant overhead caused by full-branch operation, thereby reducing the computing power and power consumption of real-time inference. By inputting environmental state data into multiple target prediction branches in parallel for humidity prediction, complementary prediction results from different modeling perspectives can be obtained at the same time, effectively mitigating the prediction bias of a single model under boundary conditions. By fusing the prediction values ​​of each branch to determine the target humidity, the advantages of each branch in feature representation can be combined to improve the stability and accuracy of the final output, thereby achieving higher accuracy of refrigerator internal humidity sensing capability at a lower cost without the need for a physical humidity sensor.

[0028] Fourthly, some embodiments also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0029] Acquire environmental status data and external environment data collected by the refrigerator within the current time period;

[0030] Based on the reference data for branch selection, at least one target prediction branch is selected from at least one humidity prediction branch in the humidity prediction network;

[0031] Environmental state data are input into each target prediction branch to predict humidity, and the predicted humidity value under the corresponding target prediction branch is obtained.

[0032] Based on the predicted humidity values, determine the target humidity for the refrigerator at the current moment.

[0033] The readable storage medium provided in the above embodiments stores a computer program that, when executed by a processor, simultaneously collects environmental state data and external environmental data. This allows for the comprehensive capture of multi-source information affecting the humidity inside the refrigerator, laying a data foundation for accurate decision-making. By dynamically selecting appropriate prediction branches using branch selection reference data, the most suitable computing resources can be adaptively called according to different operating conditions or scenarios, avoiding redundant overhead caused by full-branch computation, thereby reducing the computing power and power consumption of real-time inference. Parallel input of environmental state data into multiple target prediction branches for humidity prediction enables the acquisition of complementary prediction results from different modeling perspectives at the same time, effectively mitigating the prediction bias of a single model under boundary conditions. By fusing the prediction values ​​of each branch to determine the target humidity, the advantages of each branch in feature representation can be combined to improve the stability and accuracy of the final output, thereby achieving higher precision humidity sensing capability inside the refrigerator at a lower cost without the need for a physical humidity sensor.

[0034] Fifthly, some embodiments also provide a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0035] Acquire environmental status data and external environment data collected by the refrigerator within the current time period;

[0036] Based on the reference data for branch selection, at least one target prediction branch is selected from at least one humidity prediction branch in the humidity prediction network;

[0037] Environmental state data are input into each target prediction branch to predict humidity, and the predicted humidity value under the corresponding target prediction branch is obtained.

[0038] Based on the predicted humidity values, determine the target humidity for the refrigerator at the current moment.

[0039] The computer program provided in the above embodiments, when executed by the processor, can comprehensively capture multi-source information affecting the humidity inside the refrigerator by simultaneously collecting environmental state data and external environmental data, laying a data foundation for accurate decision-making. By dynamically selecting appropriate prediction branches using branch selection reference data, the most suitable computing resources can be adaptively called according to different working conditions or scenarios, avoiding redundant overhead caused by full-branch operation, thereby reducing the computing power and power consumption of real-time inference. By inputting environmental state data into multiple target prediction branches in parallel for humidity prediction, complementary prediction results from different modeling perspectives can be obtained at the same time, effectively mitigating the prediction bias of a single model under boundary conditions. By fusing the prediction values ​​of each branch to determine the target humidity, the advantages of each branch in feature representation can be combined to improve the stability and accuracy of the final output, thereby achieving higher accuracy of refrigerator internal humidity sensing capability at a lower cost without the need for physical humidity sensors. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 A schematic block diagram of a first refrigerator structure provided for some embodiments of this application;

[0042] Figure 2 A schematic block diagram of a second refrigerator structure provided for some embodiments of this application;

[0043] Figure 3 Schematic block diagram of the structure of the processing device in the refrigerator provided in some embodiments of this application;

[0044] Figure 4 A schematic block diagram of a third refrigerator structure provided for some embodiments of this application;

[0045] Figure 5 A schematic block diagram of a fourth refrigerator structure provided in some embodiments of this application;

[0046] Figure 6A A flowchart illustrating a first method for predicting refrigerator humidity provided in some embodiments of this application;

[0047] Figure 6B A schematic diagram of a humidity prediction network structure provided in some embodiments of this application;

[0048] Figure 7A flowchart illustrating a target prediction branch selection step provided for some embodiments of this application;

[0049] Figure 8A A flowchart illustrating a weight selection and determination step provided for some embodiments of this application;

[0050] Figure 8B A schematic diagram of a weight determination function provided for some embodiments of this application;

[0051] Figure 9 A flowchart illustrating the training steps of a humidity prediction network provided in some embodiments of this application;

[0052] Figure 10A A flowchart illustrating a second refrigerator humidity prediction method provided in some embodiments of this application;

[0053] Figure 10B A target prediction branch structure diagram is provided for some embodiments of this application;

[0054] Figure 11 A schematic flowchart of a refrigerator humidity prediction device provided in some embodiments of this application;

[0055] Figure 12 This is an internal structural diagram of a computer device provided for some embodiments of this application. Detailed Implementation

[0056] The embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described below do not represent all embodiments consistent with this application. They are merely examples of systems and methods consistent with some aspects of this application as detailed in the claims.

[0057] 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.

[0058] The terms "first," "second," "third," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar or related objects or entities, and do not necessarily imply a specific order or sequence, unless otherwise specified. It should be understood that such terms are interchangeable where appropriate.

[0059] The terms “comprising” and “having”, and any variations thereof, are intended to cover but not exclude inclusion, for example, a product or device that includes a range of components is not necessarily limited to all of the components that are clearly listed, but may include other components that are not clearly listed or that are inherent to such product or device.

[0060] The term "module" refers to any known or subsequently developed hardware, software, firmware, artificial intelligence, fuzzy logic, or combination of hardware and / or software code that is capable of performing the functions associated with that element.

[0061] The refrigerator 1 provided in this application will now be described with reference to the accompanying drawings. The overall structure of the refrigerator 1 is as follows: Figure 1 As shown. Refrigerator 1 includes a cabinet 10 and a processing device 20.

[0062] like Figure 2 As shown, the housing 10 has at least one storage compartment.

[0063] Storage rooms are typically divided into freezer rooms and refrigerator rooms (referred to as refrigerator rooms). They can also be further divided into chambers with special functions, such as chambers for storing fruits and vegetables. Refrigerator rooms can maintain a temperature range of approximately 4°C to store food, medicine, or biological agents in a refrigerated state. Freezer rooms can maintain a temperature range of approximately -18°C to store food, medicine, or biological agents in a frozen state.

[0064] The storage compartment has an opening that can be opened and closed via a door 11 hinged to the outer casing, or via a drawer 12. When a refrigerator compartment and a freezer compartment are provided, one opening can be opened and closed via a door (e.g., the refrigerator compartment), and the other opening can be opened and closed via a drawer 12 (e.g., the freezer compartment).

[0065] The housing 10 employs a vapor compression refrigeration cycle to generate energy for maintaining the target temperature. The refrigeration cycle consists of a compressor 161, a condenser, a throttling device, and an evaporator. The refrigeration cycle involves a series of processes, including compression, condensation, expansion, and evaporation, to cool the storage compartment and maintain an ideal low-temperature storage environment inside.

[0066] In a vapor compression refrigeration cycle, a low-temperature, low-pressure refrigerant enters the compressor 161, which compresses it into a high-temperature, high-pressure refrigerant gas and discharges the compressed refrigerant gas. The discharged refrigerant gas flows into the condenser, where the condenser condenses the compressed refrigerant into a liquid phase, and the heat is released to the surrounding environment through the condensation process.

[0067] The throttling device causes the high-temperature, high-pressure liquid refrigerant formed in the condenser to expand into a low-pressure liquid refrigerant. The evaporator evaporates the refrigerant that has expanded in the throttling device and returns the low-temperature, low-pressure refrigerant gas to the compressor 161. The evaporator can achieve a cooling effect by exchanging heat with the material to be cooled through the latent heat of refrigerant evaporation. In this application, the evaporator exchanges heat with air to form air for cooling the storage compartment, thereby cooling the storage compartment. The throttling device can be a capillary tube.

[0068] A filter is also installed downstream of the condenser. The filter is used to filter impurities in the refrigerant, improve the heat exchange efficiency of the refrigeration unit, and reduce the risk of pipe blockage.

[0069] A liquid receiver can also be installed on the suction side of the compressor 161. The liquid receiver is used to separate the refrigerant into gas and liquid phases. The liquid receiver is a shell-shaped component. The gas-liquid mixed refrigerant fluid enters the liquid receiver for basic phase separation. The gas enters the gas passage and undergoes gravity settling to separate droplets, while the liquid enters the liquid space and separates into bubbles. The gas flows out from the gas outlet and is then drawn into the compressor 161, preventing the compressor 161 from carrying liquid in the suction and reducing the service life of the compressor 161.

[0070] The compressor 161 and condenser can be located at the lower rear of the housing, while the evaporator can be located at the rear of the housing corresponding to the storage compartment. The evaporator and condenser can also be arranged in other locations according to the industrial design of the housing 10, which will not be listed here. The location where the evaporator is located has sufficient space to allow air to flow. The air is driven by the fan 162 to deliver the air generated by the evaporator for cooling the storage compartment to the target location and to draw in air from the storage compartment, forming an air circulation. In one or more embodiments of this application, the fan 162 includes a refrigeration fan and a freezing fan. In one or more embodiments of this application, the fan 162 can also be configured in conjunction with the condenser.

[0071] In one or more embodiments of this application, the evaporator may also be divided into two parts for the refrigerator compartment and the freezer compartment, referred to as the refrigerator compartment cooler and the freezer compartment cooler.

[0072] A defrosting element is provided in the housing 10. The defrosting element is configured to generate heat for defrosting the evaporator, thereby putting the evaporator in a defrosting state. In one or more embodiments of this application, the defrosting element includes a defrosting heater 163, which may be an electric heating tape or an electric heater. In one or more embodiments of this application, the defrosting element may also be a combination of an electric heating tape or an electric heater, and a heat exchanger or heat exchange piping. When defrosting conditions are met, the heat exchange piping is opened, and the high-temperature, high-pressure refrigerant discharged from the compressor 161 enters the heat exchange piping, exchanges heat with the surrounding air, raises the air temperature, and further provides heat to melt the frost layer on the evaporator surface, thereby putting the evaporator in a defrosting state. The heat exchange piping may be located below the evaporator, utilizing the principle that hot air has a lower density and rises to guide the air to remove the ice or frost layer on the evaporator. The defrosting element composed of an electric heating tape or an electric heater may also be located around the evaporator in other positions, such as above or to one side of the evaporator.

[0073] A display 164 is installed on the cabinet 10.

[0074] The cabinet 10 is equipped with a refrigeration system, which is configured to transfer heat from the inside of the refrigerator to the outside through the circulation of refrigerant.

[0075] like Figure 3 As shown in the figure, the hardware configuration of the processing device 20 is as follows. The processing device 20 includes components such as a processor 201, volatile memory 203, non-volatile memory 202, display device 204, operation device 205, communication interface 206, and drive device 207, which are interconnected via a bus 208. The processor 201 can be a dedicated processor 201, a central processing unit, etc. The processor 201 can access the storage unit to execute instructions or application programs stored in the storage unit to achieve related functions. The display device 204 is a display device 204 for displaying various information, the operation device 205 is an operation device for receiving various operations, and the drive device 207 is a hardware terminal that interacts with the storage medium. In one or more embodiments of this application, the storage medium includes media such as CD-ROM, floppy disk, and optical-magnetic-optical disk that record information in an optical, electrical, or magnetic manner. The storage medium can also be a semiconductor memory such as ROM or flash memory that records information in an electrical manner.

[0076] In one or more embodiments of this application, the processing device 20 may be a controller 13. The controller 13 is disposed in the housing 10.

[0077] In one or more embodiments of this application, the processing device 20 may be communicatively connected to the controller 13, for example, by a terminal device 15 and / or a cloud server 14.

[0078] In one or more embodiments of this application, some functions of the processing device 20 may be implemented by the controller 13, and some functions may be implemented by the terminal device 15 and / or the cloud server 14.

[0079] Controller 13 can communicate with terminal device 15 and / or server 14. The network between controller 13 and terminal device 15, or between controller 13 and server 14, can be the Internet, cellular network, Wi-Fi network, low power wide area network (LPWAN), WAN, LAN, etc., based on standards and protocols such as LoRa, Sigfox, and NB-IoT.

[0080] The cabinet 10 can be used in home environments to store daily necessities such as food and cold drinks; it can also be used in commercial places such as restaurants, hotels, supermarkets, and convenience stores to store ingredients, food, and drinks to meet customer needs; and it can also be used in places such as hospitals and laboratories to store medicines and biological samples to meet medical and scientific research needs.

[0081] Server 14 can provide various network services, such as resource and data access for refrigerator 1 controller 13 and terminal device 15. Server 14 has higher performance and reliability. Server 14 can connect to multiple refrigerator 1 controllers 13, multiple terminal devices 15, and other smart home appliance terminals.

[0082] Terminal device 15 is an electronic device with intelligent functions. It can connect to the aforementioned networks to achieve functions such as remote control, data exchange, and human-computer interaction. Terminal device 15 includes smartphones, tablets, smart speakers, wearable devices, smart home appliances (such as smart TVs), and smart in-vehicle devices, etc. The interaction methods between terminal device 15 and users include, but are not limited to: operating on the screen with a finger or stylus, performing various operations through buttons, voice control, gesture control, iris recognition, and facial recognition, etc.

[0083] In one or more embodiments of this application, the housing 10 is communicatively connected to the sensor assembly 30. At least a portion of the sensors in the sensor assembly 30 are disposed within the housing 10.

[0084] like Figure 4 and Figure 5 As shown, in one or more embodiments of this application, the sensor assembly 30 includes at least one temperature sensor; the temperature sensor may include at least one of a compartment temperature sensor 31, an evaporator temperature sensor 32, and an ambient temperature sensor 33.

[0085] For example, the compartment temperature sensor 31 includes a refrigerator compartment temperature sensor 311 and a freezer compartment temperature sensor 312. The refrigerator compartment temperature sensor 311 is installed in the refrigerator compartment of the cabinet 10 to detect the temperature of the refrigerator compartment; the freezer compartment temperature sensor 312 is installed in the freezer compartment of the cabinet 10 to detect the temperature of the freezer compartment.

[0086] In one or more embodiments of this application, the compartment temperature sensor 31 further includes a fruit and vegetable compartment temperature sensor 313.

[0087] In one or more embodiments of this application, the compartment temperature sensor 31 further includes a variable temperature compartment temperature sensor 314.

[0088] For example, an evaporator temperature sensor 32 is disposed on the evaporator for detecting the temperature of the evaporator.

[0089] In one or more embodiments of this application, the evaporator temperature sensor includes a refrigerator compartment cooler temperature sensor 321 and a freezer compartment cooler temperature sensor 322.

[0090] In one or more embodiments of this application, the ambient temperature sensor 33 includes an indoor temperature sensor 331.

[0091] In one or more embodiments of this application, the ambient temperature sensor 33 includes an indoor temperature sensor 331 and an outdoor temperature sensor (not shown). The outdoor temperature can also be obtained by querying a server.

[0092] In one or more embodiments of this application, the sensor assembly 30 further includes a humidity sensor.

[0093] In one or more embodiments of this application, the sensor assembly 30 further includes a door switch sensor to detect the opening and closing of the door 11.

[0094] In one or more embodiments of this application, the sensor assembly 30 may also include other sensors, such as vibration sensors, weight sensors, etc.

[0095] In one or more embodiments of this application, the sensor assembly 30 further includes an electrical parameter sensor 34. The number of electrical parameter sensors 34 is not limited, and the electrical parameter sensors 34 can be used to detect one or more of the following: electrical charge, peak electrical charge, valley electrical charge, current, voltage, and energy efficiency.

[0096] In one or more embodiments of this application, the electrically driven actuator 16 in the housing 10 includes a compressor 161, a fan 162, and a defrost heater 163.

[0097] In one or more embodiments of this application, the electrically driven actuators 16 in the housing 10 include a compressor 161, a fan 162, a defrost heater 163, a display 164, and may also include, for example, a water pump in an ice-making module and a motor in an ice-crushing module.

[0098] Currently, virtual humidity sensing within a refrigerator can be achieved by constructing a model using existing operating parameters (such as evaporator temperature, compressor start / stop, fan speed, defrosting status, and ambient temperature). For example, by collecting time-series data such as compressor start / stop cycles, fan runtime, and evaporator temperature changes, a nonlinear mapping relationship with the internal humidity can be established, thus enabling humidity prediction under sensorless conditions. This method achieves relatively ideal prediction accuracy under normal operating conditions. However, in practical applications, refrigerators face complex and variable external environments. Studies show that when a refrigerator is in a high ambient temperature (e.g., above 32°C) and high ambient humidity (e.g., above 80%RH), the prediction accuracy of the virtual humidity sensor decreases significantly. Taking a certain existing virtual humidity sensor algorithm as an example, under ambient temperature of 25°C and ambient humidity of 40%RH, its 24-hour average prediction deviation is approximately 4.34%; while when the ambient temperature rises to 32°C and the ambient humidity rises to 80%RH, as shown in Table 1, the prediction deviation can climb to over 10%, and even exceed 11% at some settings. This is mainly because under high temperature and humidity conditions, the refrigerator's cooling load increases significantly, and the dynamic characteristics such as compressor operating time and evaporator frosting and defrosting cycles differ fundamentally from those under normal operating conditions. Existing single models struggle to simultaneously cover the complex nonlinear mapping relationships under multiple operating conditions. To address this issue, an intuitive improvement approach is to add ambient temperature and humidity as new input parameters while maintaining the model structure, and to expand the training data collection range to cover high temperature and humidity scenarios, thereby retraining the single model. However, experimental results show that training with mixed data under different ambient temperature and humidity conditions leads to difficulty in model convergence, mutual interference between data from different operating conditions, and the prediction bias under high ambient temperature and humidity conditions not only fails to decrease but further worsens. For example, under 32°C and 80%RH conditions, the bias for some settings rises to 18.93%, as shown in Table 2 below. This indicates that a single model architecture is difficult to accommodate the differentiated dynamic characteristics under multiple operating conditions, and a technical solution capable of differentiated modeling for different operating conditions is urgently needed.

[0099] Table 1. Detection Accuracy of Virtual Humidity Sensor Algorithm

[0100]

[0101] Table 2. Data training results under different ambient humidity conditions.

[0102]

[0103] To overcome the above problems, in some alternative embodiments, see Figure 6A A method for predicting humidity in a refrigerator is provided, which can be applied to the controller in the refrigerator and may include the following steps:

[0104] S601 acquires environmental status data and external environment data collected by the refrigerator during the current time period.

[0105] Among them, environmental state data refers to the set of raw operating parameters continuously collected by various sensing sensors inside the refrigerator (such as temperature, fan speed, door open / closed, etc.) within the current time period. It is the most basic input of the model, containing multiple dimensions of physical measurements and their raw sequences of changes over time.

[0106] External environmental data refers to data collected by external sensors installed on the refrigerator, used to characterize the external environment in which the refrigerator operates. As a reference for comparing the internal environment, this data specifically includes, but is not limited to, ambient humidity and ambient temperature.

[0107] For example, the environmental status data in this embodiment may include temperature environmental status data and refrigerator environmental status data. The temperature environmental status data may further include refrigerator temperature data and refrigerator temperature difference data. The refrigerator temperature data may include the set refrigerator temperature, the set freezer temperature, the actual refrigerator temperature, the actual freezer temperature, the refrigerator evaporator temperature, and the freezer evaporator temperature. The refrigerator temperature difference data may include the refrigerator temperature difference and the freezer temperature difference. The refrigerator environmental status data may include the refrigerator status status, compressor power, refrigerator cooling on / off status, refrigerator door on / off status, refrigerator fan speed, and humidifier water tank setting.

[0108] For example, the controller is communicatively connected to an internal sensing sensor; the sensing sensor collects environmental state data within the current time period and sends this environmental state data to the controller; the controller receives this environmental state data. Additionally, the controller is communicatively connected to an external sensing sensor; the sensing sensor collects external environmental data within the current time period and sends this external environmental data to the controller; the controller receives this external environmental data.

[0109] It should be noted that in this embodiment, the environmental state data can be preprocessed based on the input layer of the humidity prediction network. For example, the input size of the environmental state data is [batchsize, D], where batchsize represents the number of input batches, typically 1, and D represents the data dimension of the environmental state data. After accumulating N environmental state data points within the current time period, the environmental state data is integrated and normalized to make the input size [batchsize, D, N], where N represents the time dimension, indicating that the humidity value at this moment is inferred from the N consecutive environmental state data points (collected at 30-second intervals, N / 2 minutes of data from the past).

[0110] S602 selects at least one target prediction branch from at least one humidity prediction branch in the humidity prediction network based on reference data selected from the branch selection reference data.

[0111] The reference data for branch selection refers to specific data parameters used in the "environmental status data and / or external environment data" collected by the refrigerator to select at least one target prediction branch from multiple humidity prediction branches in the humidity prediction network. This data serves as the basis or triggering condition for branch selection, reflecting the current operating conditions or environmental characteristics of the refrigerator, thereby dynamically determining which one or more prediction branches will participate in humidity prediction.

[0112] In this context, the humidity prediction branch refers to multiple parallel sub-models or sub-network modules that constitute the humidity prediction network. Each branch is designed to predict humidity for specific operating conditions or data characteristics. In the specific implementation, different humidity prediction branches can adopt the same or similar network structures (such as including time feature extractors, state feature extractors, pooling layers, and output heads), but they are trained independently using different training samples to learn the humidity mapping rules under different collection conditions (such as high ambient temperature and high humidity scenarios versus non-high ambient temperature and high humidity scenarios).

[0113] For example, such as Figure 6B The schematic diagram of the humidity prediction network structure shown illustrates that, when the reference data for the branch is the ambient humidity, the humidity prediction network in this embodiment can include two humidity prediction branches: a normal humidity branch (branch A) and a high humidity branch (branch B). The humidity ranges corresponding to different humidity prediction branches are different. In this embodiment, based on the smoothing fusion layer in the humidity prediction network, a humidity prediction branch that matches the humidity range to which the ambient humidity belongs can be determined as the target prediction branch.

[0114] It should be noted that the model prediction results are reasonably selected and fused based on the refrigerator's operating status data. Considering the differences between the prediction results of the two paths, frequent switching of the model path may cause jumps in the model prediction values ​​when the ambient humidity is near the threshold, affecting the prediction accuracy. Therefore, a smoothing fusion layer is designed for smoothing.

[0115] S603 inputs environmental state data into each target prediction branch to perform humidity prediction, and obtains the humidity prediction value under the corresponding target prediction branch.

[0116] The humidity prediction value refers to the single prediction result of the humidity inside the refrigerator at the current moment, which is output by a selected target prediction branch after the environmental state data collected by the refrigerator is input into that branch. Each selected target prediction branch will independently output a corresponding humidity prediction value. These prediction values ​​come from different modeling perspectives or different operating condition adaptation branches and may differ.

[0117] In some embodiments, environmental state data is input into each target prediction branch for humidity prediction; the target prediction branch performs humidity prediction based on the environmental state data to obtain the humidity prediction value under the corresponding target prediction branch.

[0118] S604 determines the target humidity of the refrigerator at the current moment based on the predicted humidity values.

[0119] In one alternative embodiment, when the target prediction branch is a single branch, the humidity prediction value is used as the target humidity of the refrigerator at the current moment.

[0120] In one alternative embodiment, when there are at least two target prediction branches, the target humidity of the refrigerator at the current moment is determined by the weighted sum of the humidity prediction values ​​and the selection weights corresponding to the respective target prediction branches.

[0121] For example, for each humidity prediction value, a selection weight that matches the target prediction branch corresponding to the humidity prediction value is determined; the product between the humidity prediction value and the selection weight is determined; and the image quality between all the products corresponding to the humidity prediction values ​​is taken as the target humidity of the refrigerator at the current moment.

[0122] In the above embodiments, the target humidity of the refrigerator at the current moment is determined by weighted summation of the humidity prediction values ​​output by each target prediction branch and their corresponding selection weights. This fully integrates the modeling advantages of different prediction branches under their respective favorable operating conditions, and uses weights to prioritize the prediction results of high-confidence branches while retaining the auxiliary correction role of low-confidence branches. Compared with a single model or hard switching, this scheme effectively avoids the jump in prediction values ​​at the model switching boundary, significantly improving the smoothness, robustness, and overall accuracy of humidity prediction under all operating conditions. Especially under complex boundary conditions such as high ambient temperature and high humidity, it can achieve more stable and accurate humidity perception with lower computational overhead.

[0123] In the above embodiments, by simultaneously collecting environmental state data and external environmental data, multi-source information affecting the humidity inside the refrigerator can be comprehensively captured, laying a data foundation for accurate decision-making. By dynamically selecting appropriate prediction branches using branch selection reference data, the most suitable computing resources can be adaptively called according to different working conditions or scenarios, avoiding redundant overhead caused by full-branch operation, thereby reducing the computing power and power consumption of real-time inference. By inputting environmental state data into multiple target prediction branches in parallel for humidity prediction, complementary prediction results from different modeling perspectives can be obtained at the same time, effectively mitigating the prediction bias of a single model under boundary conditions. By fusing the prediction values ​​of each branch to determine the target humidity, the advantages of each branch in feature representation can be combined to improve the stability and accuracy of the final output, thereby achieving higher precision humidity sensing capability inside the refrigerator at a lower cost without the need for a physical humidity sensor.

[0124] Based on the technical solutions of the above embodiments, some optional embodiments are also provided. In these optional embodiments, the step of selecting at least one target prediction branch from at least one humidity prediction branch according to the branch selection reference data in the above embodiments is refined.

[0125] See Figure 7 The target prediction branch selection steps shown include:

[0126] S701 determines the selection weight of each humidity prediction branch under the branch selection reference data.

[0127] The selection weight refers to the numerical value used to quantify the priority and contribution of each humidity prediction branch in the current operating conditions (characterized by the branch selection reference data) to be adopted or involved in the fusion. Each humidity prediction branch can correspond to a selection weight, which can be obtained by looking up a table based on a preset first mapping relationship, or by inputting the branch selection reference data into the weight determination function corresponding to each branch in real time.

[0128] In some embodiments, the selection weight of each humidity prediction branch is determined based on the smoothing fusion layer in the humidity prediction network and the reference data selected by the branch.

[0129] S702 selects at least one target prediction branch from at least one humidity prediction branch according to each selection weight.

[0130] In one alternative embodiment, the humidity prediction branch with the highest selection weight can be selected as the target prediction branch.

[0131] In one optional embodiment, the humidity prediction branches can also be sorted in descending order of selection weight; a preset number of humidity prediction branches with the highest selection weights are selected as target prediction branches.

[0132] In an optional embodiment, humidity prediction models whose selection weights exceed a preset weight lower limit can be selected as target prediction branches. In the above embodiment, by setting a preset weight lower limit, only humidity prediction branches with weights exceeding this lower limit are selected as target prediction branches. This effectively filters out redundant branches with low confidence and small contribution under the current operating conditions, avoiding interference from low-quality prediction results on the final humidity fusion value. Compared to using all branches without selection or relying solely on a single branch, this scheme significantly reduces computational consumption and latency during model inference while ensuring prediction reliability, thus improving the system's real-time response capability. Furthermore, by adjusting the threshold of the preset weight lower limit, the stringency of branch selection can be flexibly controlled to adapt to the differentiated accuracy and efficiency requirements of different application scenarios, thereby achieving high accuracy, low overhead, and high robustness in refrigerator humidity prediction under complex operating conditions.

[0133] In one optional embodiment, a flag bit corresponding to each selection weight can also be determined; based on the flag bits of each humidity prediction branch, at least one target prediction branch is selected from at least one humidity prediction branch. For example, in this embodiment, the flag bit can be set to "on" when the selection weight is 1, and set to "off" when the selection weight is 0. Alternatively, in this embodiment, the flag bit can be set to "off" when the selection weight is 0, and set to "on" when the selection weight is not 0; this embodiment does not limit the scope of the specification.

[0134] It should be noted that the branch weights, model parameters and model structures of different humidity prediction branches in this embodiment can be the same or different, and this embodiment does not limit them.

[0135] In the above embodiments, by first determining the selection weight of each humidity prediction branch under the branch selection reference data (such as ambient humidity, ambient temperature, etc.), and then selecting at least one target prediction branch from multiple branches based on the weight, intelligent branch selection and resource adaptation based on the current operating conditions can be achieved. Compared with the fixed use of all branches or random selection, this scheme can dynamically call the prediction branch with the best performance or the highest confidence according to different environmental conditions, effectively avoiding redundant calculations and potential noise interference caused by low-relevance branches, thereby significantly reducing the computing power and power consumption of real-time inference while ensuring prediction accuracy. At the same time, by excluding branches with too low weights through the weight selection mechanism, the signal-to-noise ratio and stability of the fusion result can be further improved, enabling the refrigerator to obtain highly reliable humidity prediction results in an efficient and lightweight manner under various complex operating conditions.

[0136] Based on the technical solutions of the above embodiments, some optional embodiments are also provided. In these optional embodiments, the step of determining the selection weight of each humidity prediction branch under the branch selection reference data in S701 of the above embodiments is refined.

[0137] See Figure 8A The steps for determining the selection weights shown include:

[0138] Based on the first mapping relationship, S801 finds the preset weights corresponding to each humidity prediction branch under the branch selection reference data, and uses them as the selection weights for the corresponding humidity prediction branches.

[0139] The first mapping relationship includes the preset weights of each humidity prediction branch under different preset branch selection reference data.

[0140] For example, the reference data for branch selection is determined; based on the first mapping relationship, the preset weight corresponding to each humidity prediction branch under the reference data for branch selection is found; and the preset weight is used as the selection weight of the corresponding humidity prediction branch.

[0141] For example, such as Figure 8BThe diagram illustrates the weight determination function. In this embodiment, based on the first mapping relationship, the preset weights corresponding to each humidity prediction branch under the branch selection reference data are found and used as the selection weights for the corresponding humidity prediction branches. That is, as shown in the diagram, taking ambient humidity as the branch selection reference data as an example, when the ambient humidity is less than 70% RH, the selection weight of humidity prediction branch A can be determined to be 1, while the selection weight of humidity prediction branch B is 0. In this case, humidity prediction branch A can be selected as the target humidity branch. When the ambient humidity is greater than 80% RH, the selection weight of humidity prediction branch A can be determined to be 0, while the selection weight of humidity prediction branch B is 1. In this case, humidity prediction branch B can be selected as the target humidity branch. When the ambient humidity is greater than 70% RH but less than 80% RH, the selection weights of humidity prediction branch A and humidity prediction branch B can be determined based on the corresponding first mapping relationship in the diagram. In this case, both humidity prediction branch A and humidity prediction branch B can be selected as target humidity branches.

[0142] It should be noted that in this embodiment, when the selection weight of any humidity prediction branch is 1 and the selection weight of other humidity prediction branches is 0, humidity prediction can be performed only based on the humidity prediction branch with a selection weight of 1 to reduce the computational resources occupied by humidity prediction; alternatively, humidity prediction can be performed on all humidity prediction branches. In this case, since the selection weight of other humidity prediction branches is 0, the humidity prediction results of other humidity prediction branches will not affect the current humidity value.

[0143] In the above embodiments, by pre-constructing a first mapping relationship, reference data for different branch selections (such as ambient humidity and ambient temperature) are associated with preset weights for each humidity prediction branch. During actual inference, the selection weights of each branch can be quickly obtained by simply looking up this mapping relationship based on the currently collected branch selection reference data. Compared to methods that require online weight calculation (such as real-time solution using a weight determination function), this scheme significantly reduces the complexity and time consumption of on-site calculations, improving the model's inference efficiency and real-time response capability. Simultaneously, the preset weights can be obtained based on numerous prior experiments or offline optimization, ensuring the rationality and stability of weight assignment and avoiding numerical fluctuations or abnormal jumps that may occur with online calculations. Therefore, this scheme achieves lower computational overhead and higher operating efficiency while ensuring branch selection accuracy and prediction precision, making it particularly suitable for resource-constrained embedded refrigerator controller environments.

[0144] In some embodiments, the branch selection reference data can be input into the weight determination function corresponding to each humidity prediction branch to obtain the selection weight of the corresponding humidity prediction branch under the branch selection reference data.

[0145] In this design, the reference data for branch selection is ambient humidity data, and the weight determination function is a monotonic function of the ambient humidity data. By using ambient humidity data as the reference data for branch selection and setting the weight determination function to its monotonicity, the design leverages ambient humidity—a key factor directly affecting humidity changes within the refrigerator—to achieve dynamic control over the physical meaning of the predicted branch selection weights. When ambient humidity increases, the monotonic function increases the weights of predicted branches for high-humidity scenarios and decreases the weights of branches for low-humidity scenarios, and vice versa. This ensures that the weight allocation follows a regular and interpretable unidirectional trend with changes in ambient humidity. This design avoids the prediction instability caused by irregular weight jumps with ambient humidity and eliminates the need for complex multivariate fusion calculations. While maintaining model simplicity, it significantly improves the refrigerator's adaptive humidity prediction capability and generalization performance across a wide humidity range.

[0146] For example, the branch selection reference data is input into the weight determination function corresponding to each humidity prediction branch; the weight determination function determines the selection weight of the corresponding humidity prediction branch under the branch selection reference data based on the branch selection reference data.

[0147] In the above embodiments, by configuring a corresponding weight determination function for each humidity prediction branch, and inputting the branch selection reference data into the function in real time to dynamically calculate the selection weight, the weight values ​​of each branch can be continuously and adaptively generated according to the current operating conditions, rather than relying on a fixed preset weight table. Compared with the static mapping method, this scheme can more flexibly capture the nonlinear or continuous change relationship between the branch selection reference data and the branch confidence level, avoiding the accuracy loss caused by the discretization of preset weights. It is especially suitable for scenarios where the branch selection reference data fluctuates frequently near the threshold boundary, thereby achieving smoother and more accurate branch fusion and prediction result output.

[0148] Based on the technical solutions of the above embodiments, some optional embodiments are also provided, in which the steps of the above embodiments are refined.

[0149] See Figure 9 The training steps for the humidity prediction network shown include:

[0150] The S901 acquires multiple training sample data and the humidity labels corresponding to each training sample data.

[0151] The training sample data includes branch training samples used to independently train each humidity prediction branch, as well as joint training samples used to jointly train the humidity prediction network; the collection conditions for the training sample data corresponding to different branch training samples are different.

[0152] In some embodiments, sample environmental state data collected by the sample refrigerator during the time period in which the sample time is located are obtained. The sample environmental state data is then classified to obtain multiple training sample data.

[0153] For example, when the sample environmental state data includes sample data collected under environmental humidity of 0-95% RH, the sample environmental state data can be used as training samples. Furthermore, this embodiment can also classify the sample environmental state data based on environmental humidity. For example, sample environmental state data of 0-75% RH can be used as training sample data specifically for training the humidity prediction branch for normal humidity; sample environmental state data of 75-95% RH can be used as training sample data specifically for training the humidity prediction branch for high humidity.

[0154] It should be noted that in this embodiment, the actual humidity value of the sample refrigerator during the time period at which the sample is located can be measured by an external high-precision humidity sensor; the external high-precision humidity sensor is fixed on the inner wall of the refrigerator and collects humidity data at equal time intervals (30s).

[0155] It should be noted that the sample environmental status data in this embodiment can take into account at least one of the following factors: ambient temperature (data was collected at ambient temperatures of 16, 25, and 32°C respectively), ambient humidity (data was collected at ambient humidity of 40%, 60%, and 80%RH respectively), temperature setting (data was collected at typical temperature settings, such as default [4, -18], high [2, -25], medium [5, -20]), and door open / close status (considering that opening the refrigerator door will significantly affect the humidity inside the refrigerator, data collection should include two stages: stable operation and door opening / closing).

[0156] S902 For each current humidity prediction branch, fix the model parameters of the current humidity prediction branch, and train the model of the current humidity prediction branch based on the branch training samples that match the humidity prediction branch until the model accuracy of the current humidity prediction branch meets the branch training completion condition, and determine that the current humidity prediction branch training is complete.

[0157] In some embodiments, for each current humidity prediction branch, the model parameters of the current humidity prediction branch are frozen; the current humidity prediction branch is trained only based on the branch training samples that match the humidity prediction branch; and the current humidity prediction branch is determined to be trained when the model accuracy of the current humidity prediction branch meets the condition that the branch training is completed.

[0158] With all humidity prediction branches having completed model training, S903 trains the humidity prediction network to be adjusted based on joint training samples until the model accuracy of the humidity prediction network meets the network training completion conditions, thus determining that the humidity prediction network training is complete.

[0159] In some embodiments, when all humidity prediction branches have completed model training, the humidity prediction network to be adjusted is trained based on joint training samples; when the model accuracy of the humidity prediction network meets the network training completion condition, the humidity prediction network training is determined to be complete.

[0160] In the above embodiments, by adopting a phased training strategy, each humidity prediction branch is first trained independently using training samples from branches with different collection conditions. This allows each branch to focus on learning the humidity mapping patterns under specific operating conditions (such as high ambient temperature and humidity, or normal operating conditions), avoiding the model's difficulty in convergence caused by interference between data from different operating conditions. After the branch training is completed, the entire humidity prediction network is fine-tuned using joint training samples, enabling each branch to achieve synergistic integration while maintaining its own specialization. This improves the overall prediction consistency and generalization ability of the network under all operating conditions. Compared to a single model mixed data training method, this scheme effectively solves the problem of accuracy degradation caused by interference between data from multiple operating conditions, significantly enhances the robustness and prediction accuracy of the model under boundary conditions such as high ambient temperature and humidity, and reduces optimization difficulty and accelerates convergence speed through phased fixed parameter training. This lays a solid model foundation for refrigerators to achieve high-precision and high-reliability virtual humidity perception in all scenarios.

[0161] Based on the technical solutions of the above embodiments, some optional embodiments are also provided, in which the humidity prediction process of the refrigerator is described in detail.

[0162] See Figure 10A The refrigerator humidity prediction method shown includes:

[0163] S1001 acquires environmental status data and external environment data collected by the refrigerator during the current time period;

[0164] S1002 determines the selection weight of each humidity prediction branch under the branch selection reference data;

[0165] Optionally, the method for determining the selection weight of each humidity prediction branch under the branch selection reference data can be as follows: based on the first mapping relationship, find the preset weight corresponding to each humidity prediction branch under the branch selection reference data, and use it as the selection weight of the corresponding humidity prediction branch; wherein, the first mapping relationship includes the preset weight corresponding to each humidity prediction branch under different preset branch selection reference data.

[0166] Optionally, the selection weights of each humidity prediction branch under the branch selection reference data can also be determined by inputting the branch selection reference data into the weight determination function corresponding to each humidity prediction branch, thereby obtaining the selection weights of the corresponding humidity prediction branch under the branch selection reference data. The branch selection reference data is the ambient humidity data; the weight determination function is a monotonic function of the ambient humidity data.

[0167] S1003 selects at least one target prediction branch from at least one humidity prediction branch according to each selection weight.

[0168] Optionally, selecting at least one target prediction branch from at least one humidity prediction branch based on each selection weight can be achieved by selecting humidity prediction models whose corresponding selection weights exceed a preset lower limit as target prediction branches. Alternatively, at least one target prediction branch can be selected from at least one humidity prediction branch based on the magnitude relationship between the selection weights.

[0169] S1004 inputs environmental status data into each target prediction branch to perform humidity prediction, and obtains the humidity prediction value under the corresponding target prediction branch.

[0170] S1005 determines the target humidity of the refrigerator at the current moment by weighting the selected weights of each humidity prediction value and the corresponding target prediction branch.

[0171] For example, such as Figure 10B The target prediction branch structure diagram shown represents the environmental state data [batchsize, D, N] input to the humidity prediction network, where D represents the data dimension of a single environmental state data point, and N represents the time window size. The input data first enters the temporal feature extractor (i.e., the temporal feature extraction layer in the humidity prediction network). This module employs a multi-scale structure design, consisting of three parallel 1D temporal convolutional layers (Conv1d) with kernel sizes of 3, 5, and 7, respectively, to capture temporal dependencies across different time spans. After merging the outputs of each convolution, global average pooling is used to compress the time dimension N to 1, achieving dimensionality reduction and aggregation of temporal features. The pooled features are flattened and transformed into intermediate features of size [batchsize, D], which are then input into the state feature extractor, composed of cascaded fully connected layers, to mine the intrinsic state correlations between multi-dimensional sensor parameters. Finally, the output head integrates the state features and outputs the predicted humidity value inside the refrigerator at the current moment.

[0172] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0173] Based on the same inventive concept, this application also provides a refrigerator humidity prediction device for implementing the refrigerator humidity prediction method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more refrigerator humidity prediction device embodiments provided below can be found in the limitations of the refrigerator humidity prediction method described above, and will not be repeated here.

[0174] In one exemplary embodiment, such as Figure 11 As shown, a refrigerator humidity prediction device is provided, including: an acquisition module 1101, a selection module 1102, a prediction module 1103, and a determination module 1104. Wherein,

[0175] The acquisition module 1101 is used to acquire environmental status data and external environment data collected by the refrigerator during the current time period.

[0176] The selection module 1102 is used to select at least one target prediction branch from at least one humidity prediction branch in the humidity prediction network based on the branch selection reference data.

[0177] The prediction module 1103 is used to input environmental state data into each target prediction branch to predict humidity, and obtain the humidity prediction value under the corresponding target prediction branch.

[0178] The determination module 1104 is used to determine the target humidity of the refrigerator at the current moment based on the predicted humidity values.

[0179] In some embodiments, the selection module 1102 is further configured to determine the selection weight of each humidity prediction branch under the branch selection reference data; and select at least one target prediction branch from at least one humidity prediction branch according to each selection weight.

[0180] In some embodiments, the selection module 1102 is further configured to select each humidity prediction model whose corresponding selection weight exceeds a preset weight lower limit as the target prediction branch.

[0181] In some embodiments, the selection module 1102 is further configured to select at least one target prediction branch from at least one humidity prediction branch according to the size relationship between the selection weights.

[0182] In some embodiments, the determining module 1104 is further configured to determine the target humidity of the refrigerator at the current moment based on the weighted sum of the selected weights corresponding to each humidity prediction value and the corresponding target prediction branch.

[0183] In some embodiments, the selection module 1102 is further configured to find the preset weights corresponding to each humidity prediction branch under the branch selection reference data based on the first mapping relationship, and use them as the selection weights of the corresponding humidity prediction branches; wherein, the first mapping relationship includes the preset weights corresponding to each humidity prediction branch under different preset branch selection reference data.

[0184] In some embodiments, the selection module 1102 is further configured to input the branch selection reference data into the weight determination function corresponding to each humidity prediction branch to obtain the selection weight of the corresponding humidity prediction branch under the branch selection reference data; the branch selection reference data is the ambient humidity data; the weight determination function is a monotonic function of the ambient humidity data.

[0185] In some embodiments, the refrigerator humidity prediction device further includes: a training module, configured to acquire multiple training sample data and humidity labels corresponding to each training sample data; wherein the multiple training sample data includes branch training samples for independently training each humidity prediction branch, and joint training samples for jointly training the humidity prediction network; the acquisition conditions of the training sample data corresponding to different branch training samples are different; for each current humidity prediction branch, the model parameters of the current humidity prediction branch are fixed, and the current humidity prediction branch is trained based on the branch training samples that match the humidity prediction branch, until the model accuracy of the current humidity prediction branch meets the branch training completion condition, and the current humidity prediction branch training is determined to be complete; when all humidity prediction branches have completed model training, the humidity prediction network to be adjusted is trained based on the joint training samples, until the model accuracy of the humidity prediction network meets the network training completion condition, and the humidity prediction network training is determined to be complete.

[0186] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 12As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When executed by the processor, the computer program implements a method for predicting humidity in a vacuum drawer. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0187] Those skilled in the art will understand that Figure 12 The structures shown are merely block diagrams of some structures related to the embodiments of this application and do not constitute a limitation on the computer devices on which the embodiments of this application are applied. Specific computer devices may include more or fewer components than those shown in the figures, or combine certain components, or have different component arrangements.

[0188] In one alternative embodiment, Figure 12 The computer device shown may be the aforementioned refrigerator. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0189] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0190] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0191] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with relevant regulations.

[0192] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0193] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0194] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A refrigerator, characterized in that, include: The enclosure has at least one compartment. A refrigeration system, located inside the cabinet, is configured to transfer heat from inside the refrigerator to the outside through the circulation of refrigerant. An internal sensing sensor is used to collect environmental status data inside the refrigerator; An external sensing sensor is used to collect external environmental data of the environment in which the refrigerator is located; the environmental status data and / or the external environmental data include branch selection reference data; At least one controller, connected to the sensing sensor, is configured to: Acquire environmental status data and external environment data collected by the refrigerator during the current time period; Based on the reference data selected from the branch, at least one target prediction branch is selected from at least one humidity prediction branch in the humidity prediction network; The environmental state data is input into each of the target prediction branches to predict humidity, and the humidity prediction value under the corresponding target prediction branch is obtained. Based on the predicted humidity values, the target humidity of the refrigerator at the current moment is determined.

2. The refrigerator according to claim 1, characterized in that, When the controller performs the operation of selecting at least one target prediction branch from at least one humidity prediction branch based on the reference data selected from the branch, it is configured to: Determine the selection weight of each humidity prediction branch under the reference data selected by the branch; Based on the selection weights, at least one target prediction branch is selected from at least one humidity prediction branch.

3. The refrigerator according to claim 2, characterized in that, The step of selecting at least one target prediction branch from at least one humidity prediction branch according to each of the selection weights includes: Each humidity prediction model whose selection weight exceeds the preset weight lower limit is selected as the target prediction branch.

4. The refrigerator according to claim 2, characterized in that, The step of selecting at least one target prediction branch from at least one humidity prediction branch according to each of the selection weights includes: Based on the relationship between the selected weights, at least one target prediction branch is selected from at least one humidity prediction branch.

5. The refrigerator according to claim 2, characterized in that, Based on the predicted humidity values, the target humidity of the refrigerator at the current moment is determined, including: The target humidity of the refrigerator at the current moment is determined by the weighted sum of the selected weights of each humidity prediction value and the corresponding target prediction branch.

6. The refrigerator according to any one of claims 2-5, characterized in that, Determine the selection weight of each humidity prediction branch under the reference data selected by the branch, including: Based on the first mapping relationship, find the preset weight corresponding to each humidity prediction branch under the branch selection reference data, and use it as the selection weight of the corresponding humidity prediction branch. The first mapping relationship includes preset weights corresponding to each humidity prediction branch under different preset branch selection reference data.

7. The refrigerator according to claim 6, characterized in that, The determination of the selection weight of each humidity prediction branch under the reference data of the branch selection includes: The reference data for branch selection is input into the weight determination function corresponding to each humidity prediction branch to obtain the selection weight of the corresponding humidity prediction branch under the reference data for branch selection.

8. The refrigerator according to claim 7, characterized in that, The reference data for selecting the branch is the ambient humidity data; the weight determination function is a monotonic function of the ambient humidity data.

9. The refrigerator according to any one of claims 1-5, characterized in that, The controller is configured to: when executing the training method for the humidity prediction network. Acquire multiple training sample data and humidity labels corresponding to each training sample data; wherein, the multiple training sample data includes branch training samples used to independently train each humidity prediction branch, and joint training samples used to jointly train the humidity prediction network; the collection conditions of the training sample data corresponding to different branch training samples are different. For each current humidity prediction branch, the model parameters of the current humidity prediction branch are fixed, and the model of the current humidity prediction branch is trained based on the branch training samples that match the current humidity prediction branch until the model accuracy of the current humidity prediction branch meets the branch training completion condition, and the training of the current humidity prediction branch is determined to be completed. With all humidity prediction branches having completed model training, the humidity prediction network to be adjusted is trained based on the joint training samples until the model accuracy of the humidity prediction network meets the network training completion condition, at which point the humidity prediction network training is considered complete.

10. A method for predicting refrigerator humidity, characterized in that, The method includes: Acquire environmental status data and external environment data collected by the refrigerator during the current time period; the environmental status data and / or the external environment data include branch selection reference data. Based on the reference data selected from the branch, at least one target prediction branch is selected from at least one humidity prediction branch in the humidity prediction network; The environmental state data is input into each of the target prediction branches to predict humidity, and the humidity prediction value under the corresponding target prediction branch is obtained. Based on the predicted humidity values, the target humidity of the refrigerator at the current moment is determined.

Images