Refrigerator compressor start-stop control method based on data processing
By collecting refrigerator operation data to calculate the rewarming thermal inertia index and attention weight, and combining it with a long short-term memory network for weighted fusion, adaptive start-stop control of the refrigerator compressor was achieved, solving the problem of temperature instability and improving control accuracy and response speed.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DA PAN ELECTRIC APPLIANCE IND CO LTD
- Filing Date
- 2026-04-10
- Publication Date
- 2026-05-12
AI Technical Summary
Existing data processing-based refrigerator compressor control methods cannot effectively cope with changes in the internal load state of the refrigerator and external interference, resulting in unstable temperature.
By collecting multi-dimensional time-series data during refrigerator operation, calculating the thermal inertia index, and determining the attention weight by combining the door opening and closing frequency, the historical information and short-term state in the long short-term memory network are weighted and fused to achieve adaptive start-stop control of the compressor.
It improves the adaptability of refrigerator compressor start/stop control and the stability of internal temperature, while taking into account both temperature prediction accuracy and response speed.
Smart Images
Figure CN122015413A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of refrigerator control technology, and in particular to a refrigerator compressor start-stop control method based on data processing. Background Technology
[0002] As a refrigeration device that maintains a constant low temperature environment, a refrigerator primarily relies on the intermittent operation of its compressor to regulate the internal temperature. Its control strategy directly impacts the preservation of food and the energy consumption of the equipment. With the development of intelligent technology, data-processing-based intelligent control is gradually replacing traditional mechanical temperature control.
[0003] Currently, patent application CN119665526A discloses a refrigerator and a refrigerator compressor operation control method. The refrigerator includes: a cabinet with a door at its opening, and at least one storage compartment for storing items; a compressor for compressing and circulating refrigerant; a compartment temperature sensor for collecting the temperature of the storage compartment; an ambient temperature sensor for collecting the ambient temperature; and a controller. The method includes: acquiring the compartment temperature, ambient temperature, and a current set temperature; constructing a feature vector based on the compartment temperature, ambient temperature, and current set temperature; inputting the feature vector into a pre-trained neural network model for calculation; obtaining an operating frequency output; and controlling the compressor to operate at the operating frequency.
[0004] The above method directly inputs the real-time collected compartment temperature, ambient temperature, and current set temperature into a pre-trained neural network model, and controls the compressor based on the output operating frequency. However, during the operation of the refrigerator, changes in the internal load state and external interference from frequent door opening and closing will affect the refrigerator's cooling effect, making it impossible for the compressor to achieve adaptive control, which in turn leads to unstable internal temperature. Summary of the Invention
[0005] To address the technical issues of low accuracy and low product yield in data-processing-based refrigerator compressor start-stop control, this application provides a data-processing-based refrigerator compressor start-stop control method that can improve the adaptability of refrigerator compressor start-stop control and the stability of internal temperature.
[0006] In a first aspect, this application provides a refrigerator compressor start-stop control method based on data processing. The control method includes: collecting multi-dimensional time-series data during refrigerator operation, including internal temperature, ambient temperature, and door open / close status; calculating the rewarming thermal inertia index based on the time required for a unit temperature rise inside the refrigerator in response to the compressor switching to a stop state; statistically analyzing the refrigerator door opening / closing frequency within a preset time window to determine the attention weight of historical information, wherein the attention weight is negatively correlated with the opening / closing frequency and positively correlated with the rewarming thermal inertia index, and the preset time window begins when the compressor switches to a stop state and ends at the current control moment; weightedly fusing the historical information and short-term states at the current control moment in a long short-term memory network based on the attention weight to obtain a fused feature vector at the current control moment, wherein the historical information is the average of multiple short-term states prior to the current control moment; inputting the fused feature vector into a predictive regression layer to obtain a predicted temperature for future moments; and implementing compressor start-stop control based on the predicted temperature and the real-time internal temperature.
[0007] By calculating the thermal inertia index in response to the compressor switching to the stop state and determining the attention weight based on the door opening and closing frequency, the historical information and short-term state of the current control moment in the long short-term memory network are weighted and fused. When the thermal inertia is large and there is little interference, the historical trend is emphasized to filter noise, and when there is frequent interference, the short-term state is emphasized to capture sudden changes, thereby realizing the adaptive start-stop control of the refrigerator compressor.
[0008] Preferably, after collecting multidimensional time-series data during refrigerator operation, the control method further includes: performing a moving average filtering process on the collected internal temperature to remove random noise; recording the change in refrigerator door status as a door opening / closing state when the duration exceeds a preset threshold; and normalizing the processed internal temperature, ambient temperature, and door opening / closing state to obtain input data for inputting into the long short-term memory network.
[0009] Preferably, the calculation of the rewarming thermal inertia index based on the time required for a unit temperature rise inside the refrigerator includes: in response to the compressor switching to a stop state, recording the moment when the compressor is in the stop state as the start time, and recording the start time and the start temperature inside the refrigerator; continuously monitoring until the current control time is reached, recording the current control time as the end time, and recording the end time and the end temperature inside the refrigerator; calculating the time difference between the end time and the start time, and using the absolute value of the difference between the end temperature inside the refrigerator and the start temperature inside the refrigerator as the effective temperature rise; and using the ratio of the time difference to the effective temperature rise as the rewarming thermal inertia index at the current control time.
[0010] Quantifying the heat load of items stored in the refrigerator into specific numerical indicators reflects the time required for a unit temperature rise. This allows the system to identify whether the refrigerator is in a light-load state with rapid temperature rise and low inertia, or in a heavy-load state with slow temperature rise and high inertia, providing a data foundation for subsequent control processes.
[0011] Preferably, the effective temperature rise further includes: calculating the difference between the temperature inside the cut-off chamber and the temperature inside the starting chamber as a first temperature difference; calculating the average ambient temperature and the average temperature inside the chamber between the cut-off time and the starting time; using the difference between the average ambient temperature and the average temperature inside the chamber as a second temperature difference; subtracting the heat transfer coefficient ratio and the product of the second temperature difference from the first temperature difference, and using the absolute value of the difference as the effective temperature rise.
[0012] The heat exchange component caused by the temperature difference between the inside and outside of the refrigerator is removed from the total temperature rise, so that the corrected effective temperature rise only reflects the heat absorption capacity of the items stored inside the refrigerator.
[0013] Preferably, the method for determining the heat transfer coefficient ratio includes: controlling the refrigerator in an unloaded state to operate sequentially at at least two different preset ambient temperatures; for each preset ambient temperature, obtaining the natural temperature rise value within a preset calibration time window during compressor shutdown, and calculating the average temperature difference between the average ambient temperature and the average internal temperature within the calibration time window; calculating the ratio of the natural temperature rise value to the average temperature difference at each preset ambient temperature, and determining the arithmetic mean of each ratio as the heat transfer coefficient ratio.
[0014] The heat transfer coefficient ratio of the refrigerator compartment was accurately obtained using experimental data, providing an accurate reference parameter for effective temperature rise correction and improving the accuracy of the reheat thermal inertia index.
[0015] Preferably, the step of determining the attention weight of historical information by statistically analyzing the opening and closing frequency of the refrigerator door within a preset time window includes: calculating a confidence coefficient, wherein the confidence coefficient is negatively correlated with the opening and closing frequency; calculating the difference between the rewarming thermal inertia index and the reference thermal inertia index, and dividing the difference by the reference thermal inertia index to obtain a normalized difference; and obtaining the attention weight based on the confidence coefficient and the normalized difference.
[0016] Preferably, attention weights Satisfying the relation: ;in, Current control time The thermal inertia index during warming, The baseline thermal inertia index; This is the confidence coefficient.
[0017] Preferably, the current control time fused feature vector Satisfying the relation: ; in, Current control time Attention weights Current control time Historical information Current control time A short-term state.
[0018] This allows the feature vector input to the prediction regression layer to contain both the stable patterns of long-term evolution and the current real-time perturbation information, thereby improving the accuracy of temperature prediction.
[0019] Preferably, the compressor start-stop control based on the predicted temperature and the real-time chamber temperature includes: starting the compressor in response to the difference between the predicted temperature and the set temperature being greater than the maximum allowable temperature.
[0020] Preferably, the training process of the Long Short-Term Memory (LSTM) network includes: collecting historical operating data of the refrigerator as a training set, using multi-dimensional time-series data of historical moments as input to the LSM network, and using the actual internal temperature of the refrigerator at the next future moment as a label; calculating the mean squared error loss between the predicted temperature output by the LSM network and the label; and iteratively updating the LSM network using the backpropagation algorithm until the mean squared error loss is less than a preset convergence threshold, or the number of iterations is greater than a preset number.
[0021] The technical solution of this application has the following beneficial technical effects: By collecting multi-dimensional time-series data during refrigerator operation and quantifying the current heat load characteristics of the refrigerator using the thermal inertia index during compressor shutdown, and calculating attention weights by combining the door opening and closing frequency within a preset time window, the system dynamically weights and fuses historical information and short-term states in the long short-term memory network. This achieves adaptive adjustment of attention to long-term historical trends and current instantaneous states. When the heat load is large or the environment is stable, it can maintain stable control based on historical patterns, while responding quickly to short-term state changes when there are frequent door openings or sudden load changes. It also balances temperature prediction accuracy and response speed, improving the adaptability of refrigerator compressor start-stop control and the stability of internal temperature. Attached Figure Description
[0022] Figure 1 This is a flowchart of a refrigerator compressor start-stop control method based on data processing according to an embodiment of this application.
[0023] Figure 2 This is a comparison chart of prediction errors according to embodiments of this application. Detailed Implementation
[0024] According to the first aspect of this application, this application provides a refrigerator compressor start-stop control method based on data processing. The refrigerator can be a vehicle refrigerator or a household refrigerator, and this application does not impose any restrictions. Figure 1 This is a flowchart of a refrigerator compressor start / stop control method based on data processing, according to an embodiment of this application. Figure 1 As shown, the refrigerator compressor start-stop control method based on data processing includes steps S101 to S105, which are described in detail below.
[0025] S101, Collect multi-dimensional time-series data during the operation of the refrigerator, including the internal temperature, ambient temperature and door open / close status.
[0026] In one embodiment, a built-in sensor array in the refrigerator is first used to collect multi-dimensional time-series data in real time during the refrigerator's operation at a preset sampling frequency. The multi-dimensional time-series data specifically includes: internal temperature, ambient temperature, and door open / closed status signal. The door open / closed status signal is a binary signal, with 0 indicating closed and 1 indicating open. The sampling frequency can be set to collect multi-dimensional data once every 10 seconds, and the refrigerator compressor start / stop control frequency can be set to a preset multiple of the sampling frequency. For example, the refrigerator compressor start / stop control frequency is 30 times the sampling frequency, meaning the refrigerator compressor start / stop control is executed once every 5 minutes.
[0027] After collecting multidimensional time-series data during refrigerator operation, preprocessing is required to ensure data quality. Specifically, the collected internal temperature is subjected to a moving average filter to remove random noise and retain the true temperature change trend. If the duration of a refrigerator door state change exceeds a preset threshold, the change is recorded as a valid door opening / closing state, thus avoiding misjudgments caused by poor sensor contact or mechanical vibration. For example, the preset threshold is 500ms. The processed internal temperature, ambient temperature, and door opening / closing state are then normalized to obtain input data for the Long Short-Term Memory (LSTM) network.
[0028] The Min-Max algorithm can be used for normalization.
[0029] Thus, by preprocessing the multidimensional time-series data, the impact of hardware noise and dimensional differences on control accuracy is eliminated, providing a data foundation for the subsequent start-stop control of the compressor.
[0030] S102, in response to the compressor switching to the stop state, calculates the rewarming thermal inertia index based on the time required for a unit temperature rise inside the refrigerator.
[0031] In one embodiment, the quantity and type of food stored in the refrigerator vary at different times, resulting in differences in the time required for a unit temperature rise within the refrigerator. For example, storing room-temperature beverages and heated food will have different effects on the time required for a unit temperature rise within the refrigerator. The rewarming thermal inertia index is used to quantify the heat load of the items stored in the refrigerator at any given time, i.e., the time required for a unit temperature rise; at any given control time, the rewarming thermal inertia index corresponding to that control time must be calculated.
[0032] Specifically, the calculation of the rewarming thermal inertia index based on the time required for a unit temperature rise inside the refrigerator includes: in response to the compressor switching to the stop state, recording the moment when the compressor is in the stop state as the start time, and recording the start time and the start temperature inside the refrigerator; continuously monitoring until the current control time is reached, recording the current control time as the end time, and recording the end time and the end temperature inside the refrigerator; calculating the time difference between the end time and the start time, and using the absolute value of the difference between the end temperature inside the refrigerator and the start temperature inside the refrigerator as the effective temperature rise; and using the ratio of the time difference to the effective temperature rise as the rewarming thermal inertia index at the current control time.
[0033] For the current control moment thermal inertia index upon rewarming Satisfying the relation: ; in, This is the current control time, which is also the cutoff time. At the starting time, For effective temperature rise, The minimum non-zero temperature rise value, which is 0.1 degrees Celsius, is used to prevent the denominator from being zero. The effective temperature rise... ; and These are the temperatures inside the stop chamber and the start chamber, respectively. The thermal inertia index during reheating. This represents the time required for the internal temperature to rise by 1 degree Celsius. A higher value indicates a greater heat load and thermal inertia inside the refrigerator.
[0034] It should be noted that if the time interval between the start time and the end time is less than the minimum statistical duration, the reference thermal inertia index will be used as the reheat thermal inertia index at the current control time, and the minimum statistical duration is 5 minutes.
[0035] In another embodiment, since ambient temperature affects the temperature rise inside the refrigerator, in order to accurately eliminate the influence of ambient temperature on the temperature rise inside the refrigerator and improve the accuracy of the rewarming thermal inertia index, the effective temperature rise further includes: calculating the difference between the cut-off temperature inside the refrigerator and the starting temperature inside the refrigerator as a first temperature difference; calculating the average ambient temperature and the average temperature inside the refrigerator between the cut-off time and the starting time; using the difference between the average ambient temperature and the average temperature inside the refrigerator as a second temperature difference; subtracting the heat transfer coefficient ratio and the product of the second temperature difference from the first temperature difference, and using the absolute value of the difference as the effective temperature rise.
[0036] Specifically, effective temperature rise Satisfying the relation: ; in, and These are the temperatures inside the cutoff chamber and the temperatures inside the starting chamber, respectively. and These are the average ambient temperature and the average temperature inside the chamber, respectively. This represents the proportion of the heat transfer coefficient, where... The heat exchange caused by the temperature difference between the average ambient temperature and the average internal temperature was quantified, and the influence of ambient temperature on the internal temperature change was corrected, thus improving the effective temperature rise. The thermal inertia index, which accurately reflects the refrigerator's heat absorption capacity, can accurately quantify the refrigerator's heat load.
[0037] The heat transfer coefficient ratio can be determined experimentally. Specifically, the method for determining the heat transfer coefficient ratio includes: controlling the refrigerator in an unloaded state to operate sequentially at at least two different preset ambient temperatures; for each preset ambient temperature, obtaining the natural temperature rise value within a preset calibration time window during compressor shutdown, and calculating the average temperature difference between the average ambient temperature and the average internal temperature within the calibration time window; calculating the ratio of the natural temperature rise value to the average temperature difference at each preset ambient temperature, and determining the arithmetic mean of each ratio as the heat transfer coefficient ratio.
[0038] Thus, the heat transfer coefficient ratio determined by experiments on refrigerators under no-load conditions ensures that the product of the heat transfer coefficient ratio and the average temperature difference accurately reflects the heat exchange caused by the temperature difference between the average ambient temperature and the average internal temperature, thus ensuring the accuracy of the reheat thermal inertia index.
[0039] S103, Statistically determine the opening and closing frequency of the refrigerator door within a preset time window to determine the attention weight of historical information. The attention weight is negatively correlated with the opening and closing frequency and positively correlated with the thermal inertia index of the temperature return. The preset time window starts when the compressor switches to the stop state and ends at the current control time.
[0040] In one embodiment, since the compressor start-stop control is performed every 5 minutes, if the current control time is 10 minutes after the compressor switched to the stop state, the length of the preset time window is 10 minutes; if the current control time is 5 minutes after the compressor switched to the stop state, the length of the preset time window is 5 minutes.
[0041] Specifically, the method of determining the attention weight of historical information by statistically analyzing the opening and closing frequency of the refrigerator door within a preset time window includes: calculating a confidence coefficient, which is negatively correlated with the opening and closing frequency; calculating the difference between the rewarming thermal inertia index and the reference thermal inertia index, and dividing the difference by the reference thermal inertia index to obtain a normalized difference; and obtaining the attention weight based on the confidence coefficient and the normalized difference.
[0042] Wherein, the confidence coefficient Satisfying the relation: ; This is the attenuation factor, with a value of 0.5; The opening and closing frequency of the refrigerator door within a preset time window; when the opening and closing frequency... An increase indicates that the reheat thermal inertia index is significantly affected by opening and closing the refrigerator door, and the confidence coefficient of the reheat thermal inertia index decreases exponentially. When the refrigerator door opening and closing frequency is 0 within the preset time window, the reheat thermal inertia index can accurately reflect the heat load inside the refrigerator, and the confidence coefficient at this time is [value missing]. The value is 1.
[0043] In other embodiments, a lower frequency limit can be set. When the switching frequency is not greater than the lower frequency limit, it is assumed that the thermal inertia index of the recirculation is not affected by the opening and closing of the refrigerator door, and the confidence coefficient is set to 1. When the switching frequency is greater than the lower frequency limit, it is assumed that the thermal inertia index of the recirculation is affected by the opening and closing of the refrigerator door, and the confidence coefficient is set to 1. ; This is the lower limit of frequency and can be set to 5.
[0044] The attenuation factor The value of the attenuation factor is determined through an engineering calibration method based on boundary condition constraints. The calibration process for the attenuation factor is described below: Under a set constant ambient temperature and fixed load or no-load condition, the refrigerator is controlled to operate without door opening interference within a preset time window. Refrigerator operating data is collected, and the thermal inertia index of the stable state is calculated, which is used as a reference value. Under the same constant ambient temperature and load condition, a specific number of door opening interferences are applied within the preset time window, and the disturbed thermal inertia index corresponding to the current number of door openings is calculated. Based on the absolute value of the difference between the disturbed thermal inertia index and the reference value, the relative deviation rate is calculated. A distortion threshold is preset, and the relative deviation rate is compared with the distortion threshold. In response to the relative deviation rate being greater than the distortion threshold, it is determined that the current thermal inertia index can no longer reflect the true thermal characterization of the load and is in a distorted state. At this time, the number of door openings corresponding to the triggering of the distortion state is recorded, and this number of door openings is calibrated as the limiting door opening frequency. The limiting door opening frequency is... With distortion threshold Substituting into the exponential decay formula of the confidence coefficient, we obtain Taking the natural logarithm of both sides of the equation, we obtain the attenuation factor. The calculation formula is as follows: For example, laboratory tests have shown that if the refrigerator door is opened and closed 6 times within a preset time window of 5 minutes, which is considered the maximum opening frequency, the refrigerator door will not open more than 6 times. At this point, the thermal inertia index is already distorted; a distortion threshold is set for this point. Substituting it into the calculation formula, we get... ; and then the attenuation factor The value is fixed at 0.5.
[0045] Then, attention weights are calculated using the confidence coefficient and the normalized thermal inertia difference. These attention weights are used to control the degree of trust the Long Short-Term Memory (LSTM) network has in historical information, which corresponds to the current control moment in the LSM network. The average of the previous 5 short-term states, at the current control time. Attention weights Satisfying the relation: ; in, The rewarming thermal inertia index at the current control moment. The reference thermal inertia index is the thermal inertia index of the refrigerator when it is unloaded; the reference thermal inertia index is the factory calibration value and can be pre-calibrated.
[0046] Understandably, when Greater than At this time, the time required for a unit temperature rise is relatively long, and the temperature change inside the refrigerator is relatively gradual. This forces the Long Short-Term Memory (LSTM) network to pay more attention to historical information, i.e., the current control moment. The average of several previous short-term states is mainly based on historical temperature change trends to predict future temperatures, ignoring short-term temperature disturbances caused by occasional door openings or defrosting; conversely, when Not greater than At this time, the time required for a unit temperature rise is relatively short, and the temperature inside the refrigerator changes rapidly. In order for the Long Short-Term Memory Network to output accurate temperature predictions, it is forced to pay more attention to the short-term state at the current moment in order to capture the rapid changes in the temperature inside the refrigerator and ensure the timeliness of the cooling response.
[0047] This can be considered as the warming thermal inertia exponent corrected for the confidence coefficient. When the confidence coefficient approaches 0, the warming thermal inertia exponent is in an uncertain state. The value of approaches 0.5, meaning that it combines historical information and the short-term state at the current moment to achieve temperature prediction for the future moment.
[0048] In this way, by using the thermal inertia index and confidence coefficient to determine the time required for a unit temperature rise, attention can be allocated to the short-term state and historical information at the current moment, thereby improving the accuracy of temperature prediction for future moments while ensuring the timeliness of the cooling response.
[0049] S104, based on the attention weights, the historical information and short-term state of the current control moment in the long short-term memory network are weighted and fused to obtain the fused feature vector of the current control moment.
[0050] In one embodiment, the input of the Long Short-Term Memory (LSTM) network is multidimensional time-series data, and the output is the predicted temperature of the multidimensional time-series data at the next future time. The LSTM network includes an input layer, a feature extraction layer, and a prediction and regression layer. The input layer is used to receive multidimensional data at each time step in the multidimensional time-series data. The feature extraction layer includes multiple LSTM units, and each LSTM unit includes a forget gate, an input gate, and an output gate. The output gate is used to output the short-term state at each time step. The prediction and regression layer is a fully connected layer used to regress the average state of the historical information and the short-term states at the current time step to the predicted temperature at the next future time step. The historical information is the average of multiple short-term states before the current time step, for example, the average of five short-term states before the current time step.
[0051] It should be noted that, since the Long Short-Term Memory (LSTM) network needs to be deployed on the refrigerator, fewer LSTM units can be set in the feature extraction layer to reduce the number of parameters in the LSM network; for example, the number of LSTM units can be set to 8. To enable the LSM network to learn the mapping relationship between multi-dimensional time-series data and the predicted temperature at the next future time, the LSM network needs to be trained. The training process of the LSM network includes: collecting historical operating data of the refrigerator as a training set; using the multi-dimensional time-series data of historical moments as input to the LSM network; using the actual internal temperature of the refrigerator at the next future time as a label; calculating the mean squared error loss between the predicted temperature output by the LSM network and the label; iteratively updating the LSM network using the backpropagation algorithm until the mean squared error loss is less than a preset convergence threshold, or the number of iterations is greater than a preset number. The preset convergence threshold is 0.01, and the preset number of iterations is 100. It should also be noted that during the training phase, the attention weights of each historical information in the multi-dimensional time-series data of historical moments are also acquired.
[0052] Current control time fused feature vector Satisfying the relation: ; in, Current control time Attention weights Current control time Historical information Current control time The short-term state; the historical information The short-time state is used to characterize the patterns of slow change learned from long-term history. Used to characterize the current control moment The input information carries rapidly changing instantaneous information.
[0053] S105, the fused feature vector is input into the prediction regression layer to obtain the predicted temperature at future times; the compressor start-stop control is realized based on the predicted temperature and the real-time internal temperature.
[0054] In one embodiment, the fused feature vector integrates long-term patterns and short-term disturbances in temperature changes. Inputting this fused feature vector into the prediction regression layer accurately outputs the predicted temperature for future times. See also... Figure 2 The figure shows a comparison between the prediction error in the Long Short-Term Memory Network (LSTM) of this application and the prediction error in the standard Long Short-Term Memory Network (LSTM). As can be seen from the figure, by weighting historical information and short-term states, the prediction error of future temperature can be reduced.
[0055] The compressor start-stop control based on the predicted temperature and the real-time chamber temperature includes: starting the compressor in response to the difference between the predicted temperature and the set temperature being greater than the maximum allowable temperature.
[0056] Understandably, the set temperature is the temperature suitable for food preservation. After the compressor stops running, the temperature inside the refrigerator will gradually rise. When the difference between the predicted temperature and the set temperature is greater than the maximum allowable temperature, it means that the temperature inside the refrigerator at a future time will not meet the minimum requirements for food preservation. At this time, the compressor is started to lower the temperature inside the refrigerator. The maximum allowable temperature is usually 1.2 times the set temperature.
[0057] After the compressor is started, the control method further includes: switching the compressor to a stop state in response to the internal temperature remaining below the set temperature for a preset duration. For example, the preset duration is the refrigerator compressor start-stop control frequency, i.e., the compressor is switched to a stop state when the internal temperature remains below the set temperature for 5 minutes.
[0058] In this way, the refrigerator compressor can be adaptively started and stopped.
[0059] It should be noted that the scope of protection of this patent application shall be determined by the appended claims.
Claims
1. A refrigerator compressor start / stop control method based on data processing, characterized in that, The control method includes: collecting multi-dimensional time-series data during the operation of the refrigerator, wherein the multi-dimensional time-series data includes the internal temperature, ambient temperature and door open / close status; In response to the compressor switching to the stop state, the rewarming thermal inertia index is calculated based on the time required for a unit temperature rise inside the refrigerator; The frequency of opening and closing of the refrigerator door within a preset time window is statistically analyzed to determine the attention weight of historical information. The attention weight is negatively correlated with the opening and closing frequency and positively correlated with the thermal inertia index of the temperature return. The preset time window starts when the compressor switches to the stop state and ends at the current control moment. Based on the attention weights, the historical information and short-term states at the current control moment in the long short-term memory network are weighted and fused to obtain the fused feature vector at the current control moment. The historical information is the average value of multiple short-term states before the current control moment. The fused feature vector is input into the predictive regression layer to obtain the predicted temperature at future times; the compressor start-stop control is realized based on the predicted temperature and the real-time internal temperature.
2. The refrigerator compressor start / stop control method based on data processing according to claim 1, characterized in that, After collecting multidimensional time-series data during the refrigerator's operation, the control method further includes: performing a moving average filter on the collected internal temperature to remove random noise; recording the change in the refrigerator door state as a door opening / closing state when the duration exceeds a preset threshold; and normalizing the processed internal temperature, ambient temperature, and door opening / closing state to obtain input data for the long short-term memory network.
3. The refrigerator compressor start / stop control method based on data processing according to claim 1, characterized in that, The calculation of the rewarming thermal inertia index based on the time required for a unit temperature rise inside the refrigerator includes: in response to the compressor switching to the stop state, recording the moment when the compressor is in the stop state as the start time, and recording the start time and the start temperature inside the refrigerator; continuously monitoring until the current control time is reached, recording the current control time as the end time, and recording the end time and the end temperature inside the refrigerator; calculating the time difference between the end time and the start time, and taking the absolute value of the difference between the end temperature inside the refrigerator and the start temperature inside the refrigerator as the effective temperature rise; and taking the ratio of the time difference to the effective temperature rise as the rewarming thermal inertia index at the current control time.
4. The refrigerator compressor start / stop control method based on data processing according to claim 3, characterized in that, The effective temperature rise further includes: calculating the difference between the temperature inside the cut-off chamber and the temperature inside the starting chamber as the first temperature difference; calculating the average ambient temperature and the average temperature inside the chamber between the cut-off time and the starting time; using the difference between the average ambient temperature and the average temperature inside the chamber as the second temperature difference; subtracting the heat transfer coefficient ratio and the product of the second temperature difference from the first temperature difference, and using the absolute value of the difference as the effective temperature rise.
5. The refrigerator compressor start / stop control method based on data processing according to claim 4, characterized in that, The method for determining the heat transfer coefficient ratio includes: controlling a refrigerator in an unloaded state to operate sequentially at at least two different preset ambient temperatures; for each preset ambient temperature, obtaining the natural temperature rise value within a preset calibration time window during compressor shutdown, and calculating the average temperature difference between the average ambient temperature and the average internal temperature within the calibration time window; calculating the ratio of the natural temperature rise value to the average temperature difference at each preset ambient temperature, and determining the arithmetic mean of each ratio as the heat transfer coefficient ratio.
6. The refrigerator compressor start / stop control method based on data processing according to claim 1, characterized in that, The method of determining the attention weight of historical information by statistically analyzing the opening and closing frequency of the refrigerator door within a preset time window includes: calculating a confidence coefficient, which is negatively correlated with the opening and closing frequency; calculating the difference between the rewarming thermal inertia index and the reference thermal inertia index, and dividing the difference by the reference thermal inertia index to obtain a normalized difference; and obtaining the attention weight based on the confidence coefficient and the normalized difference.
7. The refrigerator compressor start / stop control method based on data processing according to claim 6, characterized in that, Attention weight Satisfying the relation: ;in, Current control time The thermal inertia index during warming, The baseline thermal inertia index; This is the confidence coefficient.
8. The refrigerator compressor start / stop control method based on data processing according to claim 1, characterized in that, Current control time fused feature vector Satisfying the relation: ； in, Current control time Attention weights Current control time Historical information Current control time A short-term state.
9. The refrigerator compressor start / stop control method based on data processing according to claim 1, characterized in that, The compressor start-stop control based on the predicted temperature and the real-time chamber temperature includes: starting the compressor in response to the difference between the predicted temperature and the set temperature being greater than the maximum allowable temperature.
10. The refrigerator compressor start / stop control method based on data processing according to claim 1, characterized in that, The training process of the Long Short-Term Memory (LSTM) network includes: collecting historical operating data of the refrigerator as a training set; using multi-dimensional time-series data from historical moments as input to the LSM network; and using the actual internal temperature of the refrigerator at the next future moment as a label; calculating the mean squared error loss between the predicted temperature output by the LSM network and the label; and iteratively updating the LSM network using the backpropagation algorithm until the mean squared error loss is less than a preset convergence threshold or the number of iterations is greater than a preset number.