Refrigerator intelligent temperature control method, system and server

Abstract

The application provides a refrigerator intelligent temperature control method, system and server, the method comprises the following steps: obtaining historical running state data of a refrigerator to train a time sequence prediction model to obtain a refrigerator simulation model, and constructing a training environment by using the refrigerator simulation model; inputting current state data of the training environment into an initial decision model to obtain running control parameters, inputting the running control parameters and environment working condition data into the refrigerator simulation model to obtain running time sequence data tracks of the refrigerator; updating the initial decision model by using a reinforcement learning algorithm in the training environment according to predicted temperature time sequence data and set temperature time sequence data to obtain a trained decision model; and loading the trained decision model on a refrigerator control system to output a control instruction according to real-time collected state data to drive functional components to perform actions. The application solves the problems of temperature control lag and high energy consumption caused by traditional temperature control relying on online trial and error of the refrigerator, and realizes precise temperature control and energy efficiency optimization.

Description

Technical Field

[0001] This application relates to the field of smart home appliance control technology, and in particular to a method, system and server for smart temperature control of a refrigerator. Background Technology

[0002] As an essential household appliance for food preservation, the refrigerator's temperature control performance directly affects the food's freshness and energy consumption. With increasing user demands for smart and energy-efficient features, traditional refrigerator temperature control technology is no longer sufficient to meet the complex and ever-changing usage scenarios, necessitating more precise and efficient control solutions to enhance product competitiveness.

[0003] There is an existing intelligent temperature control solution for refrigerators. This solution collects measured temperature data and set temperature data from each compartment of the refrigerator, and directly uses reinforcement learning algorithms to update the decision model online on a real refrigerator. This optimizes operating control parameters such as compressor speed, solenoid valve direction, and heater on / off status, attempting to adapt the refrigerator to different operating conditions through continuous trial and error.

[0004] However, the decision-making model updates in existing technical solutions rely on online trial-and-error interaction on real refrigerators, which makes it impossible to pre-verify the control strategy in an offline environment. The trial-and-error process is prone to causing drastic temperature fluctuations in each compartment, posing a safety risk of frequent compressor start-stop, and the training efficiency is low. At the same time, this method is difficult to coordinate the optimization of compressor speed and temperature control effect while meeting the temperature control accuracy, resulting in low overall energy efficiency and increased unnecessary energy consumption costs. Summary of the Invention

[0005] This invention provides a method, system, and server for intelligent temperature control of refrigerators, to solve the technical problems in the prior art, such as temperature control lag, high energy consumption, safety risks, and low training efficiency caused by decision model updates relying on online trial and error with real refrigerators.

[0006] To achieve the above objectives, in a first aspect, the present invention provides a method for intelligent temperature control of a refrigerator, comprising: Acquire historical operating status data of the refrigerator, including measured temperature data and set temperature data of each compartment, environmental condition data, and operating status data of each functional component. The time series prediction model is trained using historical operational status data to obtain a refrigerator simulation model, and a training environment is built using the refrigerator simulation model. The current state data of the training environment is input into the initial decision model to obtain the running control parameters; The operating control parameters and the current environmental conditions of the refrigerator are input into the refrigerator simulation model to obtain the refrigerator's runtime sequence data trajectory. The runtime sequence data trajectory includes at least the predicted temperature time series data of each refrigerator compartment. Based on the predicted temperature time series data and the set temperature time series data, the initial decision model is updated in the training environment using a reinforcement learning algorithm to obtain the trained decision model. The trained decision-making model is integrated into the refrigerator control system. Based on the real-time collected status data, the trained decision-making model outputs control commands to drive each functional component to perform corresponding operating actions, so that the temperature inside the refrigerator reaches the set temperature.

[0007] Preferably, the measured temperature data of each compartment of the refrigerator includes the temperature of the refrigerator compartment, the temperature of the freezer compartment, and the temperature of the variable temperature compartment; Environmental operating data includes: ambient temperature, ambient humidity, and refrigerator cooling load status; The operating status data of functional components include: compressor operating frequency, compressor start / stop status, compressor cumulative running time, duct fan speed, and the opening / closing status of the refrigeration solenoid valve; Obtain historical operating status data of the refrigerator, including: According to the preset sampling period, the refrigerator compartment temperature, freezer compartment temperature, variable compartment temperature, set temperature data, ambient temperature, ambient humidity, refrigerator cooling load status, compressor operating frequency, compressor start / stop status, compressor cumulative running time, air duct fan speed, and the opening and closing status of the refrigeration solenoid valve are collected, and time sequence alignment is performed based on the collection timestamp to form status data with time sequence marks.

[0008] Preferably, it further includes: The first sample is obtained by acquiring historical running status data within the first time window according to the preset interception period. The first sample is sorted chronologically and duplicate data is removed to obtain the second sample. The temperature outliers in the second sample are removed to obtain the third sample, where the temperature outliers are data that exceed the preset temperature threshold. The actual operating status data of each functional component and the corresponding operating parameter values ​​are used to reconstruct features to obtain the fourth sample; The fourth sample is time-aligned to obtain the fifth sample; The fifth sample was resampled to generate a refrigerator time-series data sequence.

[0009] Preferably, the time-series prediction model is trained using historical operational state data to obtain a refrigerator simulation model, including: Obtain a historical sequence consisting of refrigerator time-series data sequences within a historical time window; Input historical data into the time series prediction model and output a predicted series of temperature data for the next time window. Calculate the error between the predicted sequence and the measured temperature data within the corresponding time window in the refrigerator time-series data sequence; The time series prediction model is updated using the error to obtain the refrigerator simulation model.

[0010] Preferably, the current state data of the training environment is input into the initial decision model to obtain the operating control parameters, including: The refrigerator compartment temperature, freezer compartment temperature, variable temperature compartment temperature, compressor operating frequency, compressor start / stop status, compressor cumulative running time, air duct fan speed, refrigeration solenoid valve opening / closing status, ambient temperature, ambient humidity, and refrigerator refrigeration load status are input into the initial decision model to obtain the operating control parameters. The operating control parameters include compressor speed, solenoid valve direction, and heater on / off status.

[0011] Preferably, the operating control parameters and the current environmental conditions of the refrigerator are input into the refrigerator simulation model to obtain the refrigerator's operating sequence data trajectory, including: A feature vector is constructed using compressor speed, solenoid valve direction, heater on / off status, and the refrigerator's current ambient temperature, ambient humidity, and refrigerator cooling load status. The feature vector is input into the refrigerator simulation model to obtain the runtime sequence data trajectory of the refrigerator.

[0012] Preferably, based on the predicted temperature time series data and the set temperature time series data, a reinforcement learning algorithm is used to update the initial decision model in the training environment to obtain a trained decision model, including: Calculate the first temperature difference between the predicted temperature time series data of the refrigerator compartment and the set temperature time series data of the refrigerator compartment, and calculate the second temperature difference between the predicted temperature time series data of the freezer compartment and the set temperature time series data of the freezer compartment; The first temperature difference, the second temperature difference, and the ratio of the compressor speed to the compressor's preset maximum speed are input into the reward function to obtain the reward value; Based on the reward value, a reinforcement learning algorithm is used to update the initial decision model in the training environment to obtain a fully trained decision model.

[0013] Preferably, calculating the temperature deviation between the predicted temperature time series data and the set temperature time series data includes: the reward function expression is:

[0014] Wherein, ΔT1 is the first temperature difference between the predicted temperature time series data of the refrigerator compartment and the set temperature time series data of the refrigerator compartment, ΔT2 is the second temperature difference between the predicted temperature time series data of the freezer compartment and the set temperature time series data of the freezer compartment, Si is the current speed of the compressor, Smax is the preset maximum speed of the compressor, and w1, w2, and w3 are preset weighting coefficients.

[0015] In a second aspect, the present invention provides a refrigerator intelligent temperature control system, applied to the refrigerator intelligent temperature control method of the first aspect, comprising: a data acquisition terminal deployed on the refrigerator body, an edge computing node communicatively connected to the data acquisition terminal, and a training server communicatively connected to the edge computing node; The data acquisition terminal is configured to collect historical operating status data and current operating status data of the refrigerator. The status data includes measured temperature data and set temperature data of each compartment of the refrigerator, environmental condition data, and operating status data of each functional component. The edge computing node is configured to: acquire status data collected by the data acquisition terminal and send the status data to the training server; The training server is configured to: train the time series prediction model using historical running state data to obtain a refrigerator simulation model, and build a training environment using the refrigerator simulation model; The current state data of the training environment is input into the initial decision model to obtain the running control parameters; The operating control parameters and the current environmental conditions of the refrigerator are input into the refrigerator simulation model to obtain the refrigerator's runtime sequence data trajectory. The runtime sequence data trajectory includes at least the predicted temperature time series data of each refrigerator compartment. Based on the predicted temperature time series data and the set temperature time series data, the initial decision model is updated in the training environment using a reinforcement learning algorithm to obtain the trained decision model. Deploy the trained decision model to edge computing nodes; The edge computing node is also configured to: invoke the trained decision model, and output control commands based on the real-time collected status data to drive each functional component to perform corresponding operating actions, so that the temperature inside the refrigerator reaches the set temperature.

[0016] Thirdly, the present invention provides a server including a processor and a memory, the memory storing computer-executable instructions that can be executed by the processor, the processor executing the computer-executable instructions to implement the refrigerator intelligent temperature control method of any one of the first aspects.

[0017] As can be seen from the above technical solutions, this application provides a method, system, and server for intelligent temperature control of a refrigerator. This method obtains a refrigerator simulation model by training a time-series prediction model using historical operating state data of the refrigerator, and then constructs a training environment using the refrigerator simulation model. In the training environment, a reinforcement learning algorithm is used to update the initial decision model, resulting in a trained decision model. The decision model is then integrated into the refrigerator control system, which outputs control commands based on real-time collected state data to drive functional components to perform operational actions. This application solves the problems of temperature control lag, high energy consumption, safety risks, and low training efficiency caused by traditional reinforcement learning temperature control relying on online trial and error with real refrigerators by constructing a training environment using a refrigerator simulation model, thereby achieving precise temperature control and energy efficiency optimization. Attached Figure Description

[0018] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a refrigerator intelligent temperature control method provided in an embodiment of this application; Figure 2 A breakdown flowchart of step S100 in the method provided in the embodiments of this application; Figure 3 A breakdown flowchart of step S200 in the method provided in the embodiments of this application; Figure 4 A breakdown flowchart of step S300 in the method provided in the embodiments of this application; Figure 5 A breakdown flowchart of step S400 in the method provided in the embodiments of this application; Figure 6 A breakdown flowchart of step S500 in the method provided in the embodiments of this application; Figure 7 This is a schematic diagram of the structure of the intelligent temperature control system for a refrigerator provided in an embodiment of this application. Detailed Implementation

[0020] 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.

[0021] 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.

[0022] In the following description, the terms "first," "second," etc., 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 with "first," "second," etc., 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.

[0023] 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.

[0024] In the field of intelligent temperature control technology for refrigerators, existing methods for optimizing refrigerator temperature control performance are typically based on fixed rules or PID linear adjustment logic. These methods trigger the start and stop of various functional components such as the compressor, fan, heater, and solenoid valve based on the deviation between the measured temperature data and the set temperature data of a single compartment. However, during refrigerator operation, it is affected by multiple factors, including environmental conditions (such as ambient temperature, humidity, and refrigerator cooling load), door opening and closing disturbances, and fluctuations in food load. Strong coupling exists between compartments and functional components, making it difficult for traditional control methods to accurately match dynamic load changes. This results in large temperature fluctuations, high energy consumption, and compromised food preservation quality.

[0025] To improve the intelligence of refrigerator temperature control, existing technologies have developed control schemes based on reinforcement learning algorithms. These schemes collect measured and set temperature data from each compartment of the refrigerator and update the decision-making model online directly on a real refrigerator to optimize operating control parameters such as compressor speed, solenoid valve direction, and heater on / off status. However, this approach relies on online trial-and-error interaction on a real refrigerator, making it impossible to pre-validate the control strategy in an offline environment. The trial-and-error process can easily lead to drastic temperature fluctuations in each compartment, posing a safety risk of frequent compressor start-stop cycles, and also results in low training efficiency. Furthermore, this method struggles to coordinate compressor speed and temperature control performance while maintaining temperature control accuracy, leading to lower overall energy efficiency and increased unnecessary energy costs. In addition, due to the lack of predictive models that reflect the complex thermodynamic response of refrigerators, existing technologies cannot predict temperature evolution trends under different control actions during the design phase, resulting in delayed temperature control response and difficulty adapting to complex and changing user scenarios.

[0026] To address the aforementioned technical problems, this application provides a method for intelligent temperature control in a refrigerator, see below. Figure 1 ,include: S100: Obtain historical operating status data of the refrigerator.

[0027] The status data includes measured temperature data and set temperature data for each compartment of the refrigerator, environmental operating condition data, and operating status data of each functional component.

[0028] Specifically, acquiring historical operating status data of the refrigerator refers to acquiring measured temperature data, set temperature data, environmental condition data, and operating status data of various functional components during the refrigerator's historical operation. Among them, the measured temperature data reflects the actual temperature of each compartment of the refrigerator, the set temperature data reflects the target temperature set by the user, the environmental condition data reflects the external environment and the refrigerator's own load status, and the operating status data of various functional components reflects the working status of components such as the compressor, fan, heater, and solenoid valve.

[0029] In some embodiments, see Figure 2 Measured temperature data for each compartment of the refrigerator: including the temperature of the refrigerator compartment, the temperature of the freezer compartment, and the temperature of the variable temperature compartment; Environmental operating data includes: ambient temperature, ambient humidity, and refrigerator cooling load status; The operating status data of functional components include: compressor operating frequency, compressor start / stop status, compressor cumulative running time, duct fan speed, and the opening / closing status of the refrigeration solenoid valve; Obtain historical operating status data of the refrigerator, including: S101. Collect data on the refrigerator compartment temperature, freezer compartment temperature, variable compartment temperature, set temperature data, ambient temperature, ambient humidity, refrigerator cooling load status, compressor operating frequency, compressor start / stop status, compressor cumulative running time, air duct fan speed, and the opening / closing status of the refrigeration solenoid valve according to the preset sampling period, and perform time sequence alignment based on the collection timestamp to form status data with time sequence markers.

[0030] Specifically, at the moment of each preset sampling period, the following data are simultaneously collected: refrigerator compartment temperature, freezer compartment temperature, variable temperature compartment temperature, set temperature data, ambient temperature, ambient humidity, refrigerator cooling load status, compressor operating frequency, compressor start / stop status, compressor cumulative runtime, duct fan speed, and the opening / closing status of the refrigeration solenoid valve. The preset sampling period can be set to 1 second, or it can be set to other values ​​within the range of 0.5 to 2 seconds depending on the response characteristics of the temperature sensor and the processing capability of the main control board. Each piece of data is appended with a timestamp of the current moment, forming a raw state record. After multiple consecutive sampling periods, multiple raw state records with timestamps are obtained. These raw state records are sorted in ascending order according to the order of their timestamps. Multiple raw state records with the same timestamp are merged into one state record. After sorting and merging, the state records are obtained in ascending order of their timestamps, forming state data with time stamps.

[0031] In some embodiments, after obtaining historical running status data, the method further includes: The first sample is obtained by acquiring historical running status data within the first time window according to the preset interception period. The first sample is sorted chronologically and duplicate data is removed to obtain the second sample. The temperature outliers in the second sample are removed to obtain the third sample, where the temperature outliers are data that exceed the preset temperature threshold. The actual operating status data of each functional component and the corresponding operating parameter values ​​are used to reconstruct features to obtain the fourth sample; The fourth sample is time-aligned to obtain the fifth sample; The fifth sample was resampled to generate a refrigerator time-series data sequence.

[0032] Specifically, data within a first time window is obtained from the time-stamped state data according to a preset extraction period. All state records within this time window are treated as a whole, referred to as the first sample. The state records in the first sample are sorted in ascending order of their acquisition timestamps. The sorted state records are then traversed to check for duplicate records with identical acquisition timestamps. If duplicate records exist, one record is retained and the rest are deleted, resulting in the second sample after removing duplicate data. The preset temperature thresholds include a minimum temperature threshold and a maximum temperature threshold. The minimum temperature threshold is lower than the lowest temperature that each compartment can reach during normal refrigerator operation, and the maximum temperature threshold is higher than the highest temperature that each compartment can reach during normal refrigerator operation. Each state record in the second sample is traversed. If the refrigerator compartment temperature, freezer compartment temperature, or variable temperature compartment temperature in the state record is lower than the minimum temperature threshold or higher than the maximum temperature threshold, the state record is identified as an abnormal record and deleted from the second sample. After outlier removal, the third sample is obtained. The actual operating status data from the third sample—compressor operating frequency, compressor start / stop status, compressor cumulative runtime, duct fan speed, and the opening / closing status of the refrigerant solenoid valve—are reconstructed with their corresponding operating parameter values. This is achieved by multiplying the compressor start / stop status by the compressor operating frequency and combining the opening / closing status of the refrigerant solenoid valve with direction encoding, thus fusing the status data and operating parameter values ​​into a single feature value, resulting in the fourth sample. Since the timestamps of the status records in the fourth sample may not be entirely consistent, the status records in the fourth sample are sorted in ascending order according to their timestamps. For records with missing or inaccurate timestamps, missing feature values ​​are filled in using linear interpolation or nearest neighbor matching, ensuring that each timestamp has a complete feature record, resulting in the time-aligned fifth sample. The fifth sample is then resampled at fixed resampling intervals, converting the original timestamp sequence into a sequence with equal time intervals. At each resampling moment, the feature data for that moment is obtained by averaging or taking the nearest neighbor value, generating a refrigerator time-series data sequence.

[0033] S200. Use historical operational status data to train the time series prediction model to obtain a refrigerator simulation model, and use the refrigerator simulation model to build a training environment.

[0034] Specifically, historical sequences are obtained from the refrigerator's time-series data. Each historical sequence contains measured temperature data, set temperature data, environmental condition data, and operational status data of various functional components at multiple consecutive time points. These historical sequences are input into the time-series prediction model, which outputs a predicted sequence of temperature data for a future period. Measured temperature data within the same time window as the predicted sequence is obtained from the refrigerator's time-series data. The error between the predicted sequence and the measured temperature data is calculated, and this error is used to update the parameters of the time-series prediction model. This process is repeated until the error meets a preset condition. The resulting time-series prediction model is then the refrigerator simulation model. The refrigerator simulation model can output corresponding predicted temperature time-series data based on the input control parameters and environmental condition data. After training, this refrigerator simulation model is used as the training environment for reinforcement learning. The decision model interacts within this simulation environment, and the actions output by the decision model are returned to the next state by the refrigerator simulation model, without needing to execute actions on a real refrigerator.

[0035] In some embodiments, see Figure 3 The time-series prediction model is trained using historical operational state data to obtain a refrigerator simulation model, including: S201. Obtain a historical sequence consisting of refrigerator time-series data sequences within a historical time window; S202. Input the historical sequence into the time series prediction model and output the predicted sequence of temperature data in the next time window; S203. Calculate the error between the predicted sequence and the measured temperature data within the corresponding time window in the refrigerator time series data sequence; S204. Update the time series prediction model using the error to obtain the refrigerator simulation model.

[0036] Specifically, a continuous data sequence is extracted from the refrigerator's time-series data according to the length of a historical time window; this data sequence is called the historical sequence. The length of the historical time window can be set to 60 minutes, or it can be set to other values ​​within the range of 30 to 120 minutes, depending on the refrigerator's thermal response characteristics. The historical sequence contains measured temperature data, set temperature data, environmental condition data, and operating status data of various functional components at multiple time points within this time window. The historical sequence is organized into a tensor format required for model input and input into the time-series prediction model. The time-series prediction model performs forward calculations on the input historical sequence and outputs a predicted sequence of temperature data for a future period; the length of this period is called the next time window. The length of the next time window can be set to 5 minutes, or it can be set to other values ​​within the range of 1 to 10 minutes, depending on the prediction requirements. Each predicted value in the predicted sequence corresponds to the temperature at a time point within the next time window. Measured temperature data within the same time window as the predicted sequence is obtained from the refrigerator's time-series data and used as the true value. The difference between the predicted temperature value and the measured temperature value at each time point in the predicted sequence is calculated. The absolute values ​​of all differences are summed or squared and then averaged to obtain the error between the predicted sequence and the measured temperature data. This error is used to calculate the gradient of each network parameter in the time-series prediction model using the backpropagation algorithm. The values ​​of the network parameters are updated according to the gradient direction, so that the predicted sequence is closer to the measured temperature data in the next forward calculation. This process of obtaining historical sequences, inputting the model, outputting the predicted sequence, calculating the error, and updating the parameters is repeated, traversing all possible historical sequences in the refrigerator time-series data. Each complete traversal is called a training epoch. After multiple training epochs, for example, 8 to 20 epochs, training stops when the error no longer decreases or the decrease is less than a preset threshold. The resulting time-series prediction model is the refrigerator simulation model.

[0037] S300. Input the current state data of the training environment into the initial decision model to obtain the running control parameters.

[0038] Specifically, the current state data of the training environment includes all state information output by the training environment at the current moment. This current state data is organized into a one-dimensional feature vector and input into the initial decision model. The initial decision model is a policy network whose input layer node number equals the dimension of the state data. The initial decision model performs forward computation on the input state data, and after nonlinear transformation through multiple hidden layers, the output layer outputs operating control parameters. These operating control parameters are instruction data used to control the actions of various functional components of the refrigerator, including compressor-related control commands, solenoid valve-related control commands, and heater-related control commands. In the early stages of training, the network parameters of the initial decision model are randomly initialized, therefore the output operating control parameters are also random. As training progresses, the parameters of the initial decision model are gradually optimized, and the output operating control parameters will enable the refrigerator simulation model to return increasingly higher reward values.

[0039] In some embodiments, see Figure 4 The current state data of the training environment is input into the initial decision model to obtain the running control parameters, including: S301. Input the refrigerator compartment temperature, freezer compartment temperature, variable compartment temperature, compressor operating frequency, compressor start / stop status, compressor cumulative running time, duct fan speed, refrigeration solenoid valve opening / closing status, ambient temperature, ambient humidity, and refrigerator refrigeration load status into the initial decision model to obtain the operating control parameters.

[0040] The operating control parameters include compressor speed, solenoid valve direction, and heater on / off status.

[0041] Specifically, the following data are read from the current state data output by the training environment: refrigerator temperature, freezer temperature, variable temperature compartment temperature, compressor operating frequency, compressor start / stop status, compressor cumulative runtime, fan speed, solenoid valve opening / closing status, ambient temperature, ambient humidity, and refrigerator cooling load status. These data are then concatenated into a one-dimensional feature vector in a preset order. The dimension of this feature vector is equal to the number of state data points. This feature vector is then input into the initial decision model. The initial decision model performs forward computation on the input feature vector and outputs operating control parameters. Among these parameters, the compressor speed is a continuous value representing the target speed of the compressor drive motor; the solenoid valve direction is a discrete value indicating whether the refrigerant flows to the refrigerator or freezer compartment; and the heater on / off status is a Boolean value indicating whether the heater is powered on. In the initial training phase, the network parameters of the initial decision model are randomly initialized, therefore the output operating control parameters are also random. As training progresses, the parameters of the initial decision model are gradually optimized, and the output operating control parameters will enable the refrigerator simulation model to return increasingly higher reward values.

[0042] S400: Input the operating control parameters and the current environmental conditions of the refrigerator into the refrigerator simulation model to obtain the refrigerator's operating sequence data trajectory.

[0043] The runtime sequence data trajectory includes at least the predicted temperature time series data for each refrigerator compartment.

[0044] Specifically, the operating control parameters output from the initial decision model are combined with the refrigerator's current environmental operating condition data and then input into the refrigerator simulation model. The refrigerator simulation model performs forward calculations based on the input data and outputs the refrigerator's operating sequence data trajectory. This operating sequence data trajectory includes at least the predicted temperature time-series data for each refrigerator compartment, recording the predicted temperature values ​​for the refrigerator compartment, freezer compartment, and variable temperature compartment at various future time points. The predicted temperature time-series data in the operating sequence data trajectory is used to calculate the reward value in subsequent steps. In addition, the operating sequence data trajectory may also include the predicted temperature time-series data for the evaporator, the predicted operating status time-series data for the compressor, and the predicted speed time-series data for the fan; time-series data from other dimensions are used to monitor the refrigerator's operating status.

[0045] In some embodiments, see Figure 5 The operating control parameters and the refrigerator's current environmental operating conditions are input into the refrigerator simulation model to obtain the refrigerator's runtime sequence data trajectory, including: S401. Construct a feature vector using compressor speed, solenoid valve direction, heater on / off status, and the current ambient temperature, ambient humidity, and refrigerator cooling load status of the refrigerator.

[0046] Specifically, the compressor speed, solenoid valve direction, and heater on / off status are read from the operating control parameters output by the initial decision model. Ambient temperature, ambient humidity, and refrigerator cooling load are read from the refrigerator's current environmental condition data. These values ​​are normalized, transforming each value to a preset range so that parameters with different dimensions can participate in the calculation on the same scale. The normalized values ​​are then concatenated according to a preset order to form a one-dimensional feature vector. Each dimension of this feature vector corresponds to a specific operating control parameter or environmental condition data item.

[0047] S402. Input the feature vector into the refrigerator simulation model to obtain the refrigerator's runtime sequence data trajectory.

[0048] Specifically, the feature vector constructed by S401 is passed as input data to the input interface of the refrigerator simulation model. After receiving the feature vector, the refrigerator simulation model executes its internal forward computation process. Based on the learned thermodynamic mapping relationship, the refrigerator simulation model performs layer-by-layer calculations on the input feature vector, ultimately outputting the refrigerator's runtime sequence data trajectory. The runtime sequence data trajectory includes the predicted temperature time-series data of each refrigerator compartment at multiple future time points. The predicted temperature time-series data records the predicted temperature values ​​of the refrigerator compartment, freezer compartment, and variable temperature compartment at various future time points. The obtained runtime sequence data trajectory is used to calculate the reward value in subsequent steps.

[0049] S500: Based on the predicted temperature time series data and the set temperature time series data, the initial decision model is updated in the training environment using a reinforcement learning algorithm to obtain the trained decision model.

[0050] Specifically, predicted temperature time-series data is extracted from the refrigerator's runtime time-series data, and set temperature time-series data is extracted from the refrigerator's historical operating status data. The predicted and set temperature time-series data are then input into a reinforcement learning algorithm, and the initial decision model is iteratively updated multiple times in the training environment until the initial decision model converges, resulting in a trained decision model.

[0051] In some embodiments, see Figure 6 Based on the predicted temperature time series data and the set temperature time series data, a reinforcement learning algorithm is used to update the initial decision model in the training environment, resulting in a trained decision model, including: S501. Calculate the first temperature difference between the predicted temperature time series data of the refrigerator compartment and the set temperature time series data of the refrigerator compartment, and calculate the second temperature difference between the predicted temperature time series data of the freezer compartment and the set temperature time series data of the freezer compartment. S502. Input the first temperature difference, the second temperature difference, and the ratio of the compressor speed to the preset maximum compressor speed into the reward function to obtain the reward value; S503. Based on the reward value, a reinforcement learning algorithm is used to update the initial decision model in the training environment to obtain a trained decision model.

[0052] Specifically, predicted temperature time-series data for the refrigerator compartment and freezer compartment are extracted from the refrigerator's operational time-series data. Set temperature time-series data for the refrigerator compartment and freezer compartment are extracted from the refrigerator's historical operational status data. The temperature difference at each time point is obtained by subtracting the temperature values ​​from the predicted and set temperature time-series data for the refrigerator compartment. The average or sum of all time-point temperature differences is then calculated to obtain the first temperature difference. Similarly, a second temperature difference is calculated between the predicted and set temperature time-series data for the freezer compartment. The compressor speed is read from the operational control parameters to obtain the preset maximum compressor speed. The ratio of the compressor speed to the preset maximum compressor speed is calculated. The first temperature difference, the second temperature difference, and the ratio of the compressor speed to the preset maximum compressor speed are input into a reward function, which outputs a scalar value as the reward value. The reward value increases as the temperature deviation decreases and the compressor speed decreases. The reward value is then passed to a reinforcement learning algorithm, which updates the network parameters of the initial decision model based on the reward value. Repeat the above process until the initial decision model converges, resulting in a fully trained decision model.

[0053] In some embodiments, calculating the temperature deviation between the predicted temperature time series data and the set temperature time series data includes: the reward function expression is:

[0054] Wherein, ΔT1 is the first temperature difference between the predicted temperature time series data of the refrigerator compartment and the set temperature time series data of the refrigerator compartment, ΔT2 is the second temperature difference between the predicted temperature time series data of the freezer compartment and the set temperature time series data of the freezer compartment, Si is the current speed of the compressor, Smax is the preset maximum speed of the compressor, and w1, w2, and w3 are preset weighting coefficients.

[0055] S600: The trained decision-making model is integrated into the refrigerator control system. Based on the real-time collected status data, the trained decision-making model outputs control commands to drive each functional component to perform corresponding operating actions, so that the temperature inside the refrigerator reaches the set temperature.

[0056] Specifically, the trained decision-making model is deployed to the edge computing nodes or the main control board of the refrigerator control system. During actual operation, the trained decision-making model acquires real-time status data from the refrigerator according to a preset control cycle. This data includes measured temperature data of each compartment, set temperature data, environmental condition data, and operational status data of each functional component. The preset control cycle can be set to 1 minute or other values ​​within the range of 30 seconds to 5 minutes, depending on the actual temperature control requirements. The trained decision-making model performs forward inference calculations on the real-time acquired status data and outputs the corresponding operating control parameters for the current moment. These operating control parameters include compressor speed, solenoid valve direction, and heater on / off status. The operating control parameters are converted into specific control commands: compressor speed commands are sent to the compressor inverter driver, solenoid valve direction commands are sent to the solenoid valve drive circuit, and heater on / off commands are sent to the heater relay. The compressor adjusts its motor speed according to the received speed commands, the solenoid valve switches the refrigerant flow direction according to the direction commands, and the heater decides whether to energize and heat according to the on / off commands. After each functional component executes the control commands, the temperature of each compartment of the refrigerator changes. Temperature sensors collect new temperature data in real time and feed it back to the trained decision-making model. The trained decision-making model outputs operating control parameters again based on the new temperature data, forming a closed-loop control. Since the trained model has been thoroughly trained in a simulation environment, its output operating control parameters can adapt to dynamic conditions such as changes in ambient temperature, door opening and closing disturbances, and fluctuations in food load. This ensures that the measured temperature data of each compartment of the refrigerator remains stable near the set temperature, thereby achieving the control target of reaching the set internal temperature. Throughout the entire operation, the trained decision-making model only performs forward inference calculations and does not perform online updates, thus avoiding the safety risks associated with online trial and error.

[0057] In some embodiments, see Figure 7 This application also provides a refrigerator intelligent temperature control system, applied to the refrigerator intelligent temperature control method provided in the above embodiments, including: The data acquisition terminal is deployed on the refrigerator body, the edge computing node is connected to the data acquisition terminal, and the training server is connected to the edge computing node. The data acquisition terminal is configured to collect historical operating status data and current operating status data of the refrigerator. The status data includes measured temperature data and set temperature data of each compartment of the refrigerator, environmental condition data, and operating status data of each functional component. The edge computing node is configured to: acquire status data collected by the data acquisition terminal and send the status data to the training server; The training server is configured to: train the time series prediction model using historical running state data to obtain a refrigerator simulation model, and build a training environment using the refrigerator simulation model; The current state data of the training environment is input into the initial decision model to obtain the running control parameters; The operating control parameters and the current environmental conditions of the refrigerator are input into the refrigerator simulation model to obtain the refrigerator's runtime sequence data trajectory. The runtime sequence data trajectory includes at least the predicted temperature time series data of each refrigerator compartment. Based on the predicted temperature time series data and the set temperature time series data, the initial decision model is updated in the training environment using a reinforcement learning algorithm to obtain the trained decision model. Deploy the trained decision model to edge computing nodes; The edge computing node is also configured to: invoke the trained decision model, and output control commands based on the real-time collected status data to drive each functional component to perform corresponding operating actions, so that the temperature inside the refrigerator reaches the set temperature.

[0058] For details, please refer to the description in the above-mentioned embodiments of the intelligent temperature control method for refrigerators, which will not be repeated here.

[0059] Based on the above-described intelligent temperature control method for refrigerators, some embodiments of this application also provide a server, including a processor and a memory. The memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the intelligent temperature control method for refrigerators described above.

[0060] Specifically, the server also includes a bus and a communication interface. The processor, communication interface, and memory are connected via the bus. The memory may include high-speed random access memory or non-volatile memory, such as disk storage. Communication with other network elements is achieved through at least one communication interface, which can use the Internet, wide area network, local area network, or metropolitan area network. The bus may be an industry-standard architecture bus, a peripheral component interconnection standard bus, or an extended industry-standard architecture bus, etc. The processor may be a general-purpose processor, including a central processing unit, a network processor, etc.; or it may be a digital signal processor, an application-specific integrated circuit, a field-programmable gate array or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of this application can be directly executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. The storage medium is located in the memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the refrigerator intelligent temperature control method of the aforementioned embodiments. The server executes instructions from memory to perform functions such as acquiring historical operating status data of the refrigerator, training a time-series prediction model to obtain a refrigerator simulation model, constructing a training environment using the refrigerator simulation model, updating the initial decision model in the training environment to obtain a trained decision model, and deploying the decision model to edge computing nodes. The server can be deployed in a cloud data center or in an enterprise's internal server cluster, interacting with the local edge computing nodes of the refrigerator via a communication network.

[0061] The above specific embodiments further illustrate the purpose, technical solution and beneficial effects of this application. It should be understood that the above are only specific embodiments of this application and are not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of this application should be included within the scope of protection of this application.

Claims

1. A method for intelligent temperature control in a refrigerator, characterized in that, include: Acquire historical operating status data of the refrigerator, including measured temperature data and set temperature data of each compartment of the refrigerator, environmental condition data, and operating status data of each functional component; The time-series prediction model is trained using the historical operational state data to obtain a refrigerator simulation model, and a training environment is constructed using the refrigerator simulation model. The current state data of the training environment is input into the initial decision model to obtain the running control parameters; The operating control parameters and the current environmental operating conditions data of the refrigerator are input into the refrigerator simulation model to obtain the operating sequence data trajectory of the refrigerator. The operating sequence data trajectory includes at least the predicted temperature time series data of each of the refrigerator compartments. Based on the predicted temperature time series data and the set temperature time series data, a reinforcement learning algorithm is used to update the initial decision model in the training environment to obtain a trained decision model. The trained decision model is integrated into the refrigerator control system. Based on the real-time collected status data, the trained decision model outputs control commands to drive each functional component to perform corresponding operating actions, so that the temperature inside the refrigerator reaches the set temperature.

2. The intelligent temperature control method for a refrigerator according to claim 1, characterized in that, The measured temperature data for each compartment of the refrigerator include the temperature of the refrigerator compartment, the temperature of the freezer compartment, and the temperature of the variable temperature compartment. The environmental operating data includes: ambient temperature, ambient humidity, and refrigerator cooling load status; The operating status data of the functional components include: compressor operating frequency, compressor start / stop status, compressor cumulative running time, duct fan speed, and the opening / closing status of the refrigeration solenoid valve. The acquisition of historical operating status data of the refrigerator includes: According to the preset sampling period, the refrigerator compartment temperature, freezer compartment temperature, variable compartment temperature, set temperature data, ambient temperature, ambient humidity, refrigerator cooling load status, compressor operating frequency, compressor start / stop status, compressor cumulative running time, air duct fan speed, and the opening / closing status of the refrigeration solenoid valve are collected, and time sequence alignment is performed based on the collection timestamp to form status data with time sequence markers.

3. The intelligent temperature control method for a refrigerator according to claim 1, characterized in that, Also includes: The historical running status data within the first time window is obtained according to a preset interception period to obtain the first sample; The first sample is sorted chronologically and duplicate data is removed to obtain the second sample. The temperature outliers in the second sample are removed to obtain the third sample, wherein the temperature outliers are data that exceed a preset temperature threshold. The actual operating status data of each functional component and the corresponding operating parameter values ​​are used to reconstruct features to obtain the fourth sample; The fourth sample is time-aligned to obtain the fifth sample; The fifth sample is resampled to generate a refrigerator time-series data sequence.

4. The intelligent temperature control method for a refrigerator according to claim 3, characterized in that, The step of training the time-series prediction model using the historical operational state data to obtain a refrigerator simulation model includes: Obtain a historical sequence consisting of the refrigerator time-series data sequence within the historical time window; The historical sequence is input into the time series prediction model, and the predicted sequence of temperature data in the next time window is output. Calculate the error between the predicted sequence and the measured temperature data within the corresponding time window in the refrigerator time-series data sequence; The time-series prediction model is updated using the error to obtain the refrigerator simulation model.

5. The intelligent temperature control method for a refrigerator according to claim 2, characterized in that, The step of inputting the current state data of the training environment into the initial decision model to obtain the running control parameters includes: The refrigerator compartment temperature, freezer compartment temperature, variable temperature compartment temperature, compressor operating frequency, compressor start / stop status, compressor cumulative running time, air duct fan speed, refrigeration solenoid valve opening / closing status, ambient temperature, ambient humidity, and refrigerator refrigeration load status are input into the initial decision model to obtain the operating control parameters. The operating control parameters include compressor speed, solenoid valve direction, and heater on / off status.

6. The intelligent temperature control method for a refrigerator according to claim 5, characterized in that, The step of inputting the operating control parameters and the current environmental operating condition data of the refrigerator into the refrigerator simulation model to obtain the refrigerator's operating sequence data trajectory includes: A feature vector is constructed using the compressor speed, the solenoid valve direction, and the heater switch status, along with the current ambient temperature, ambient humidity, and refrigerator cooling load status of the refrigerator. The feature vector is input into the refrigerator simulation model to obtain the runtime sequence data trajectory of the refrigerator.

7. The intelligent temperature control method for a refrigerator according to claim 5, characterized in that, Based on the predicted temperature time-series data and the set temperature time-series data, a reinforcement learning algorithm is used to update the initial decision model in the training environment to obtain a trained decision model, including: Calculate the first temperature difference between the predicted temperature time series data of the refrigerator compartment and the set temperature time series data of the refrigerator compartment, and calculate the second temperature difference between the predicted temperature time series data of the freezer compartment and the set temperature time series data of the freezer compartment; The first temperature difference, the second temperature difference, and the ratio of the compressor speed to the preset maximum compressor speed are input into the reward function to obtain the reward value; Based on the reward value, a reinforcement learning algorithm is used to update the initial decision model in the training environment to obtain a fully trained decision model.

8. The intelligent temperature control method for a refrigerator according to claim 7, characterized in that, The calculation of the temperature deviation between the predicted temperature time series data and the set temperature time series data includes: the reward function expression is: ； Wherein, ΔT1 is the first temperature difference between the predicted temperature time series data of the refrigerator compartment and the set temperature time series data of the refrigerator compartment, ΔT2 is the second temperature difference between the predicted temperature time series data of the freezer compartment and the set temperature time series data of the freezer compartment, Si is the current speed of the compressor, Smax is the preset maximum speed of the compressor, and w1, w2, and w3 are preset weighting coefficients.

9. A refrigerator intelligent temperature control system, applied to the refrigerator intelligent temperature control method according to any one of claims 1-8, characterized in that, include: A data acquisition terminal deployed on the refrigerator body, an edge computing node communicatively connected to the data acquisition terminal, and a training server communicatively connected to the edge computing node; The data acquisition terminal is configured to: collect historical operating status data and current operating status data of the refrigerator. The status data includes measured temperature data and set temperature data of each compartment of the refrigerator, environmental condition data, and operating status data of each functional component. The edge computing node is configured to: acquire the status data collected by the data acquisition terminal and send the status data to the training server; The training server is configured to: train the time-series prediction model using the historical running state data to obtain a refrigerator simulation model, and construct a training environment using the refrigerator simulation model; The current state data of the training environment is input into the initial decision model to obtain the running control parameters; The operating control parameters and the current environmental operating conditions data of the refrigerator are input into the refrigerator simulation model to obtain the operating sequence data trajectory of the refrigerator. The operating sequence data trajectory includes at least the predicted temperature time series data of each of the refrigerator compartments. Based on the predicted temperature time series data and the set temperature time series data, a reinforcement learning algorithm is used to update the initial decision model in the training environment to obtain a trained decision model. The trained decision model is deployed to the edge computing node; The edge computing node is also configured to: invoke the trained decision model, and output control commands based on the real-time collected status data to drive each functional component to perform corresponding operating actions, so that the temperature inside the refrigerator reaches the set temperature.

10. A server, characterized in that, The device includes a processor and a memory, the memory storing computer-executable instructions that can be executed by the processor, the processor executing the computer-executable instructions to implement the refrigerator intelligent temperature control method according to any one of claims 1 to 8.

Images