Dynamic prediction method for refrigeration demand of refrigerator

Abstract

The invention provides a refrigerator refrigeration demand dynamic prediction method. The method comprises the following steps: acquiring working condition data and equipment report data stored in a refrigerator big data networking platform; cleaning the working condition data and the equipment report data to obtain cleaned data; extracting characteristic data representing refrigeration demand quantity and starting point initial data from the cleaned data to establish a characteristic library; performing nonlinear model training based on a regression algorithm on the data of the feature library to generate a refrigeration demand prediction model; and regulating and controlling refrigerator parameters based on the calling data of the refrigeration demand prediction model so as to solve the problem that dynamic prediction of the refrigeration demand of the refrigerator is not accurate.

Description

Technical Field

[0001] This application relates to the field of smart refrigerator technology, and in particular to a method for dynamically predicting refrigerator cooling demand. Background Technology

[0002] As a core electrical appliance in the home, the refrigerator must continuously operate to maintain a stable internal temperature, ensuring constant temperature storage of items and effective preservation of various foods. The refrigeration system maintains its cooling effect through the periodic operation of components such as the compressor. The key lies in accurately identifying and responding to changes in actual cooling demand to achieve efficient energy management and system reliability. In this scenario, users need refrigerators that can intelligently adapt to different environmental conditions and operating habits, reducing unnecessary energy consumption while avoiding increased hardware complexity and maintenance burden.

[0003] In relevant energy-saving control strategies, a representative method is based on real-time acquisition of thermal imaging data of food inside the refrigerator and user behavior data. Data preprocessing and feature processing are then performed to generate a sensing data stream. Subsequently, a three-dimensional thermodynamic twin model of the refrigerator and a quantum annealing algorithm are used to calculate the distribution of cooling demand, generating a multi-temperature zone cooling power allocation scheme. Finally, a phase change energy storage unit is used to achieve directional cooling capacity transfer and dynamically allocate cooling capacity to match changes in user demand. This scheme aims to quickly respond to user behavior and optimize cooling strategies, reducing ineffective energy consumption.

[0004] Nevertheless, the aforementioned methods rely on high-cost dedicated imaging sensors (such as millimeter-wave radar and multiple thermal imaging devices), which not only increases equipment purchase and installation costs but also requires frequent maintenance to ensure sensing stability and sensitivity, resulting in significant resource consumption. Meanwhile, refrigerator control is primarily based on preset rules and temperature thresholds, failing to dynamically adapt to changes in actual operating conditions, leading to inaccurate identification of cooling demand and insufficient exploitation of energy-saving potential. Therefore, there is an urgent need for a low-cost, high-precision method to dynamically predict refrigerator cooling demand, providing a reliable basis for energy-saving control, avoiding hardware dependence, and improving system adaptability. Summary of the Invention

[0005] This application provides a method for dynamically predicting the cooling demand of refrigerators to solve the problem of inaccurate dynamic prediction of the cooling demand of refrigerators.

[0006] This application provides a method for dynamically predicting the cooling demand of a refrigerator, the method comprising:

[0007] Obtain operating condition data and equipment-reported data stored on the refrigerator's big data network platform;

[0008] The operating condition data and equipment-reported data are cleaned to obtain cleaned data.

[0009] Extract feature data characterizing cooling demand and initial start-up data from the cleaned data to establish a feature library;

[0010] The data in the feature library are used to train a nonlinear model based on a regression algorithm to generate a cooling demand prediction model;

[0011] The refrigerator parameters are adjusted based on the data retrieved from the cooling demand prediction model.

[0012] The above method acquires and cleans the operating condition data and equipment-reported data from the refrigerator big data network platform, extracts feature data and initial start-up data to build a feature library, and uses a regression algorithm to train a nonlinear model to generate a refrigeration demand prediction model, thereby improving the accuracy of dynamic prediction of refrigerator refrigeration demand and supporting more effective control of refrigerator parameters.

[0013] Optionally, before acquiring the operating condition data and equipment-reported data stored on the refrigerator big data network platform, the method includes:

[0014] Based on the target model refrigerator with an IoT module, continuous time-domain data is collected by internal sensors and reported to the refrigerator big data network platform at a preset frequency.

[0015] The continuous time-domain data includes: environmental data, setting parameters, and sensor temperature data;

[0016] The environmental data includes ambient temperature and humidity in each compartment of the refrigerator; the setting parameters include set temperature values ​​for each compartment, compressor speed, and reversing valve direction; the sensor temperature data includes temperature values ​​for each compartment and evaporator temperature.

[0017] The IoT module collects continuous time-domain data from the target refrigerator and reports it at a preset frequency. This data covers multiple parameters such as ambient temperature and humidity, set temperature of each compartment, compressor speed, reversing valve direction, compartment temperature and evaporator temperature. This provides more complete and timely raw data support for the subsequent establishment of a feature library, which helps to improve the input data quality of the refrigeration demand prediction model.

[0018] Optionally, the step of cleaning the operating data and equipment-reported data includes:

[0019] The operating condition data and equipment-reported data are sorted by time, and data from faulty equipment are excluded.

[0020] The compressor start-up cycle in the data of each refrigerator is filtered.

[0021] By sorting the operating data by time and excluding data from faulty equipment, while filtering compressor start-up cycle data, the temporal integrity and reliability of the original data can be improved, and the interference of abnormal operating data on feature extraction can be reduced, thus providing a more accurate data foundation for subsequent cooling demand prediction models.

[0022] Optionally, the filtering process includes:

[0023] Filter the refrigerator compartment data cycle for refrigerator compartments where door opening events occur;

[0024] Filter compressor operating cycles containing missing values;

[0025] Remove all data from the defrosting heating phase and the preset time period after it ends.

[0026] By filtering the room cycle data containing door opening events, removing compressor operation cycles with missing values, and removing data from the defrosting stage and the preset time period after its end, the interference of specific operating conditions on data continuity can be reduced, the impact of periodic abnormal data on feature extraction can be reduced, and the validity of compressor operation cycle data can be maintained, thereby providing a more reliable data foundation for refrigeration demand feature extraction.

[0027] Optionally, the steps of extracting feature data characterizing cooling demand and initial start-up data from the cleaned data include:

[0028] Randomly select cleaned data for a preset number of days based on the date;

[0029] The data segment from the compressor's start-up time to its shutdown time is identified as a single operating cycle;

[0030] Calculate the cumulative sum of compressor speed over time in each operating cycle, as characteristic data representing the cooling demand;

[0031] During the same compressor operating cycle, the initial data of compressor start-up point related setting parameters and environmental conditions are extracted according to the aforementioned feature data.

[0032] By randomly sampling data from multiple days to ensure sample diversity, and using the complete start-stop cycle of the compressor as the analysis unit, the cumulative sum of speed and time is used as a quantitative feature of cooling demand. Simultaneously, the setting parameters of the start-up point in the same cycle and the initial data of environmental conditions are extracted, which helps to establish a more accurate characterization relationship of cooling demand features and provides a more relevant training data foundation for the prediction model.

[0033] Optionally, the initial data includes the set temperature value of each compartment, the temperature value of each compartment at the start-up time, the direction status of the reversing valve, the evaporator temperature value, and the ambient temperature.

[0034] By collecting multi-dimensional temperature parameters (including set temperature of each compartment, real-time temperature, evaporator temperature and ambient temperature) and reversing valve direction status at startup, the initial operating conditions and environmental conditions of the equipment can be fully reflected, providing key input features for the refrigeration demand prediction model, enhancing the model's responsiveness to real-time operating conditions, and providing a basis for the dynamic adjustment of compressor parameters.

[0035] Optionally, the step of training a nonlinear model based on a regression algorithm on the data in the feature library to generate a cooling demand prediction model includes:

[0036] Based on feature engineering design, XGBoost regressor is selected as the multivariate prediction algorithm; the input of the multivariate prediction algorithm is the initial operating condition data field at the start time of each record in the feature library, and the output is the corresponding cooling demand.

[0037] The feature library dataset is split into a training set and a validation set;

[0038] Perform standardized preprocessing on the input feature fields;

[0039] Gradient boosting decision trees fits the nonlinear mapping relationship between input features and cooling demand.

[0040] Hyperparameters were tuned based on the validation set to generate a refrigeration demand prediction model that can be deployed to a refrigerator control platform.

[0041] By using the XGBoost regressor to construct a multivariate prediction model, the initial operating condition data at the start-up point is mapped to the cooling demand. After dataset splitting and feature field standardization preprocessing, a gradient boosting decision tree is used to fit the nonlinear relationship. Then, hyperparameters are tuned through the validation set, thereby improving the prediction accuracy of cooling demand, enhancing the model's generalization ability, maintaining the comparability of feature data, optimizing the model's deployment applicability on the refrigerator control platform, and providing support for precise regulation.

[0042] Optionally, the feature engineering design includes calculating the difference between the start-up temperature of the refrigerator compartment and the set temperature, the difference between the start-up temperature of the freezer compartment and the set temperature, and the difference between the average ambient temperature and the set temperature, and encoding the equipment number and the direction of the reversing valve.

[0043] By calculating the difference between the start-up temperature and the set temperature of the refrigerator and freezer compartments, as well as the difference between the mean ambient temperature, and by implementing coding processing for equipment numbers and reversing valve directions, the correlation between model input features and refrigeration demand can be enhanced, the physical meaning of feature variables can be distinguished, individual differences of equipment can be differentiated, and the system working status can be characterized, thereby optimizing the predictive model's ability to identify factors affecting refrigeration demand.

[0044] Optionally, the step of adjusting the refrigerator parameters based on the data retrieved from the cooling demand prediction model includes:

[0045] When the compressor is turned on, the initial value of the start-up point is obtained;

[0046] The initial value of the start-up point is input into the cooling demand prediction model to predict the cooling demand.

[0047] Adjust compressor parameters based on the predicted cooling demand.

[0048] The above method calls the initial value of the start-up point into the prediction model in real time when the compressor starts, generates the predicted value of cooling demand, and adjusts the compressor operating parameters accordingly. This can improve the response speed of cooling control, enhance the matching degree between the compressor operating status and the actual cooling demand, and reduce the risk of energy consumption fluctuations caused by demand fluctuations.

[0049] Optionally, the step of adjusting the compressor parameters based on the predicted cooling demand includes:

[0050] Under the condition of prioritizing the achievement of the basic compressor operating minutes, the compressor speed is adjusted and the operating time is extended to meet the predicted cooling demand;

[0051] The compressor parameters are adjusted within the limits of equipment cooling time and temperature difference.

[0052] The above method, while ensuring the basic compressor's operating time, predicts cooling demand by dynamically adjusting the compressor speed and extending the operating time, and implements parameter adjustment within the constraints of equipment cooling time and temperature difference limits. This can improve the matching accuracy between cooling output and demand, maintain system operational stability, and reduce the risk of energy consumption fluctuations caused by excessive or insufficient cooling capacity.

[0053] As can be seen from the above technical solutions, this application provides a method for dynamically predicting refrigerator cooling demand. This method involves acquiring operating condition data and equipment-reported data stored on a refrigerator big data network platform; cleaning the operating condition data and equipment-reported data to obtain cleaned data; extracting feature data representing cooling demand and initial start-up data from the cleaned data to establish a feature library; training a nonlinear model based on a regression algorithm on the data in the feature library to generate a cooling demand prediction model; and adjusting refrigerator parameters based on the data called by the cooling demand prediction model to solve the problem of inaccurate dynamic prediction of refrigerator cooling demand. Attached Figure Description

[0054] 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.

[0055] Figure 1 A flowchart illustrating the method for dynamically predicting refrigerator cooling demand provided in an embodiment of this application;

[0056] Figure 2 A visualization of feature importance in the dynamic prediction method for refrigerator cooling demand provided in this application embodiment. Detailed Implementation

[0057] 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.

[0058] 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.

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

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

[0061] In this embodiment of the application, the refrigerator is provided with a refrigerator compartment and a freezer compartment.

[0062] The primary function of the refrigerator compartment is preservation. It is typically located in the upper part of the refrigerator and is designed to maintain a temperature above 0°C, usually between 2°C and 8°C. This temperature range is sufficient to slow the growth of bacteria in food, thereby extending its shelf life while preserving its freshness and taste. Various perishable foods, such as vegetables, fruits, dairy products, cooked meats, and leftovers, can be stored in the refrigerator compartment.

[0063] The primary function of the freezer compartment is to freeze and store food for extended periods. It is typically located in the lower half of the refrigerator and is designed to operate at temperatures well below 0°C, generally below -18°C. At this extremely low temperature, the moisture in food freezes rapidly, effectively preventing bacterial growth and allowing food to be preserved for a long time without spoiling. Meat, fish, ice cream, and frozen foods can be stored in the freezer compartment for extended periods.

[0064] A refrigerator's refrigeration system maintains its cooling effect through the periodic operation of components such as the compressor. The key lies in accurately identifying and responding to changes in actual cooling demand to achieve efficient energy management and system reliability. In this scenario, users need refrigerators that can intelligently adapt to different environmental conditions and operating habits, reducing unnecessary energy consumption while avoiding increased hardware complexity and maintenance burden.

[0065] In related embodiments, real-time thermal imaging data of food inside the refrigerator and user behavior data can be collected, and data preprocessing and feature processing can be performed to generate a sensing data stream. Subsequently, the distribution of cooling demand is calculated using a three-dimensional thermodynamic twin model of the refrigerator and a quantum annealing algorithm to generate a multi-temperature zone cooling power allocation scheme. Finally, a phase change energy storage unit is used to achieve directional transmission of cooling capacity and dynamically allocate cooling capacity to match changes in user demand. This scheme aims to quickly respond to user behavior and optimize cooling strategies, reducing ineffective energy consumption. However, the above embodiments rely on high-cost dedicated imaging sensors, which not only increases equipment purchase and installation costs but also requires frequent maintenance to ensure sensing stability and sensitivity, resulting in significant resource consumption. Furthermore, refrigerator control is mainly based on preset rules and temperature thresholds, which cannot dynamically adapt to changes in actual operating conditions, leading to inaccurate identification of cooling demand and insufficient exploitation of energy-saving potential.

[0066] To address the issue of inaccurate dynamic forecasting of refrigerator cooling demand, see [link to relevant documentation]. Figure 1 This application provides a method for dynamically predicting the cooling demand of a refrigerator, the method comprising:

[0067] S100: Obtains operating condition data and equipment-reported data stored on the refrigerator's big data network platform.

[0068] S200: Clean the operating data and the data reported by the equipment to obtain cleaned data.

[0069] S300: Extract feature data characterizing cooling demand and initial start-up data from the cleaned data to establish a feature library.

[0070] S400: Train a nonlinear model based on a regression algorithm on the data in the feature library to generate a cooling demand prediction model.

[0071] S500: Adjust the refrigerator parameters based on the data called from the cooling demand prediction model.

[0072] It should be understood that by using a refrigerator big data networking platform, historical dynamic data of different users' refrigerators can be obtained, showing the relationship between the refrigerator's operating environment and parameter settings and the refrigerator's cooling demand. This allows for a more reasonable and accurate calculation of the refrigerator's cooling demand, maintaining a stable cooling effect in the refrigerator compartments. This avoids the need to add sensing devices to obtain users' actual information data, and enables refrigerator demand prediction and targeted control at a lower cost.

[0073] The above method acquires and cleans the operating condition data and equipment-reported data from the refrigerator big data network platform, extracts feature data and initial start-up data to build a feature library, and uses a regression algorithm to train a nonlinear model to generate a refrigeration demand prediction model, thereby improving the accuracy of dynamic prediction of refrigerator refrigeration demand and supporting more effective control of refrigerator parameters.

[0074] In some embodiments, before acquiring the operating condition data and device-reported data stored on the refrigerator big data networking platform, the method includes:

[0075] Based on the target model refrigerator with an IoT module, continuous time-domain data is collected by internal sensors and reported to the refrigerator big data network platform at a preset frequency.

[0076] The continuous time-domain data includes: environmental data, setting parameters, and sensor temperature data;

[0077] The environmental data includes ambient temperature and humidity in each compartment of the refrigerator; the setting parameters include set temperature values ​​for each compartment, compressor speed, and reversing valve direction; the sensor temperature data includes temperature values ​​for each compartment and evaporator temperature.

[0078] It should be understood that the preset frequency can be set to collect data once per minute. This preset frequency setting ensures the real-time nature and accuracy of the data, enabling the refrigerator big data network platform to promptly acquire and analyze the latest operating condition information and equipment status. In practical applications, the preset frequency can also be flexibly adjusted according to different refrigerator models, usage environments, and user needs. Furthermore, to ensure the comprehensiveness and integrity of the data, the collected continuous time-domain data covers multiple key dimensions, including environmental data, setting parameters, and sensor temperature data. These data collectively constitute a comprehensive profile of the refrigerator's operating status, providing a solid foundation for subsequent feature extraction and model training.

[0079] The IoT module collects continuous time-domain data from the target refrigerator and reports it at a preset frequency. This data covers multiple parameters such as ambient temperature and humidity, set temperature of each compartment, compressor speed, reversing valve direction, compartment temperature and evaporator temperature. This provides more complete and timely raw data support for the subsequent establishment of a feature library, which helps to improve the input data quality of the refrigeration demand prediction model.

[0080] In some embodiments, the step of cleaning the operating condition data and the equipment-reported data includes:

[0081] The operating condition data and equipment-reported data are sorted by time, and data from faulty equipment are excluded.

[0082] The compressor start-up cycle in the data of each refrigerator is filtered.

[0083] It should be understood that the above steps aim to remove invalid data caused by equipment failure or abnormal operation, ensuring the data quality used in subsequent analysis. Specifically, time sorting ensures that the data is arranged in chronological order, facilitating subsequent time-series analysis and feature extraction. Excluding data from faulty equipment avoids erroneous data interfering with model training. The compressor operating cycle filtering is based on the fact that the compressor is a core component of the refrigerator's refrigeration system, and the length of its operating cycle directly affects cooling demand and energy consumption. Therefore, filtering out atypical or abnormally short operating cycle data can more accurately reflect the actual operating status and cooling demand of the refrigerator. These cleaning steps are based on a deep understanding of the refrigerator's operating mechanism and aim to provide more accurate and effective data support for subsequent feature extraction and model training.

[0084] By sorting the operating data by time and excluding data from faulty equipment, while filtering compressor start-up cycle data, the temporal integrity and reliability of the original data can be improved, and the interference of abnormal operating data on feature extraction can be reduced, thus providing a more accurate data foundation for subsequent cooling demand prediction models.

[0085] In some embodiments, the filtering process includes: filtering refrigerator compartment data cycles with door opening events; filtering compressor operation cycles containing missing values; and removing all data during the defrosting heating phase and a preset time period after its end.

[0086] It should be understood that the preset time period can be selected as 3 minutes. The introduction of these filtering steps is based on an in-depth analysis of the actual operation of the refrigerator. Door opening events may cause fluctuations in the internal temperature of the refrigerator, thus affecting the accurate judgment of cooling demand. Therefore, filtering out these periodic data helps improve the accuracy of the prediction model. Compressor operating cycles containing missing values ​​may be incomplete due to sensor malfunctions or other reasons. If such data is used for model training, it may cause the prediction results to deviate from reality. As for the data within the preset time period after the defrosting heating stage and its end, since the refrigerator is in a non-cooling state during this time, its data has no practical significance for predicting cooling demand, and therefore should also be removed. Through such filtering, we can further purify the data, laying a solid foundation for subsequent feature extraction and model training.

[0087] In some embodiments, the step of extracting feature data characterizing cooling demand and initial start-up data from the cleaned data includes:

[0088] Data is randomly selected from cleaned data over a preset number of days based on the date.

[0089] It should be understood that, specifically, the cleaned data can be processed through individual refrigerator devices to randomly extract data by date. The preset number of days can be flexibly set according to actual needs; for example, data from the most recent 7 days or 14 days can be selected to ensure the timeliness and representativeness of the data.

[0090] The data segment from the moment the compressor starts to the moment it stops is identified as a running cycle.

[0091] Calculate the cumulative sum of compressor speed over time in each operating cycle, which serves as characteristic data representing the cooling demand.

[0092] It should be understood that the cumulative sum specifically represents its cooling integral area, that is, the sum of continuous rotational speed values ​​within a cycle at the same time interval, which serves as the cooling demand for that cycle. The changes in compressor rotational speed values ​​over multiple cycles, obtained from historical data, characterize the current cooling capacity and actual demand of the equipment.

[0093] During the same compressor operating cycle, the initial data of compressor start-up point related setting parameters and environmental conditions are extracted according to the aforementioned feature data.

[0094] It should be understood that each data entry in the feature library contains the initial values ​​of the start-up point for each dimension in the current cycle and the corresponding cooling demand.

[0095] By randomly sampling data from multiple days to ensure sample diversity, and using the complete start-stop cycle of the compressor as the analysis unit, the cumulative sum of speed and time is used as a quantitative feature of cooling demand. Simultaneously, the setting parameters of the start-up point in the same cycle and the initial data of environmental conditions are extracted, which helps to establish a more accurate characterization relationship of cooling demand features and provides a more relevant training data foundation for the prediction model.

[0096] In some embodiments, the initial data includes the set temperature value of each compartment, the temperature value of each compartment at the start-up time, the directional valve directional status, the evaporator temperature value, and the ambient temperature.

[0097] By collecting multi-dimensional temperature parameters (including set temperature of each compartment, real-time temperature, evaporator temperature and ambient temperature) and reversing valve direction status at startup, the initial operating conditions and environmental conditions of the equipment can be fully reflected, providing key input features for the refrigeration demand prediction model, enhancing the model's responsiveness to real-time operating conditions, and providing a basis for the dynamic adjustment of compressor parameters.

[0098] In some embodiments, the step of training a nonlinear model based on a regression algorithm on the data in the feature library to generate a cooling demand prediction model includes:

[0099] Based on feature engineering design, the XGBoost regressor was selected as the multivariate prediction algorithm; the input of the multivariate prediction algorithm is the initial operating condition data field at the start-up time of each record in the feature library, and the output is the corresponding cooling demand.

[0100] It should be understood that the XGBoost (Extreme Gradient Boosting Regressor) is a machine learning algorithm based on the gradient boosting framework, specifically designed to solve regression prediction problems. Its core mechanism involves iteratively training multiple weak learners (usually decision trees) and combining their predictions to progressively correct the residuals of preceding models, ultimately forming a strong predictive model.

[0101] In some embodiments, the feature engineering design includes calculating the difference between the start-up temperature of the refrigerator compartment and the set temperature, the difference between the start-up temperature of the freezer compartment and the set temperature, and the difference between the average ambient temperature and the set temperature, and encoding the equipment number and the direction of the reversing valve.

[0102] Specifically, based on feature library data, the differences between the start-up temperature and set temperature of the refrigerator compartment, the start-up temperature and set temperature of the freezer compartment, the average ambient temperature of the same period and the set temperature of the refrigerator compartment, and the average ambient temperature of the same period and the set temperature of the freezer compartment are calculated as initial load features. At the same time, the equipment number is binary encoded, and the switching valve direction is categorized, with the refrigerator direction coded as 1 and the freezer direction coded as 2, generating extended category features.

[0103] By calculating the difference between the start-up temperature and the set temperature of the refrigerator and freezer compartments, as well as the difference between the mean ambient temperature, and by implementing coding processing for equipment numbers and reversing valve directions, the correlation between model input features and refrigeration demand can be enhanced, the physical meaning of feature variables can be distinguished, individual differences of equipment can be differentiated, and the system working status can be characterized, thereby optimizing the predictive model's ability to identify factors affecting refrigeration demand.

[0104] The feature library dataset is split into a training set and a validation set.

[0105] Perform standardized preprocessing on the input feature fields.

[0106] It should be understood that, specifically, the XGBoost regressor can be used to pre-train on non-coding class data, with the regression target being the cooling demand of the samples, and the feature importance ranking for each dimension obtained; the top 8 features with the highest feature importance ranking are selected as the target dimensions for subsequent algorithm training. See also Figure 2Where Top Important Features represents feature importance; envsetfreezed represents the frozen state of environmental parameters; envtmp represents the real-time value of the ambient temperature; colddiffo1c2f represents the temperature difference between the start-up point of refrigerator compartment 1 and freezer compartment 2; envcolddiff represents the difference between the ambient temperature and the set temperature of the refrigerator compartment; bwdiffo1c2f represents the temperature difference between the variable temperature zones of refrigerator compartment 1 and freezer compartment 2; olo2cold represents the overshoot at the start-up point of refrigerator compartment 2; setfreeze represents the set temperature value of the freezer compartment; o2colddiff represents the difference between the start-up temperature and the set temperature of refrigerator compartment 2; cold_01 represents the real-time temperature at the start-up point of refrigerator compartment 1; and envcoldsum represents the difference between the ambient temperature and the set temperature of the refrigerator compartment. The sum of the set temperatures of the refrigerator compartment 1 and the freezer compartment 2; o1colddiff is the difference between the start-up temperature and the set temperature of the refrigerator compartment 1; bianwen_cf2 is the temperature / airflow parameter of the air duct from the refrigerator to the freezer compartment 2 in the variable temperature zone; o1bwdiff is the temperature difference between the refrigerator compartment 1 and the variable temperature zone; bianwen_01 is the real-time temperature of the variable temperature zone 1; o2freezediff is the difference between the start-up temperature and the set temperature of the freezer compartment 2; envfreezediff is the difference between the ambient temperature and the set temperature of the freezer compartment; freeze_01 is the real-time temperature of the freezer compartment 1 at the start-up point; cold_c2f is the estimated heat transfer coefficient from the refrigerator compartment to the freezer compartment; envfreezesum is the sum of the ambient temperature and the set temperature of the freezer compartment.

[0107] Gradient boosting decision trees fits the nonlinear mapping relationship between input features and cooling demand.

[0108] It should be understood that, based on the filtered dataset, the samples are randomly sorted and divided into an 80% training set and a 20% validation set, and supervised training is performed with the cooling demand as the target value; the non-difference class features are standardized, and the algorithm parameters are set to tree depth 8, number of iterations 100, and learning rate 0.3.

[0109] Hyperparameters were tuned based on the validation set to generate a refrigeration demand prediction model that can be deployed to a refrigerator control platform.

[0110] It should be understood that, based on the algorithm training strategy, the dataset is randomly divided 10 times for iterative training. The R2 score of each training iteration supports the model's interpretability; its value ranges from [0, 1], with values ​​closer to 1 indicating better fit. Training epochs with scores exceeding 0.9 are selected, and the accuracy of the cooling demand prediction on the validation set is calculated. The accuracy is denoted as: (Y - |Y - Yp|) / Y * 100%, where Y is the actual cooling demand and Yp is the predicted cooling demand. The model with an accuracy exceeding 90% in training epochs is selected as the target cooling demand prediction model.

[0111] The above method constructs a multivariate prediction model by using an XGBoost regressor to map the initial operating condition data at the start-up point to the cooling demand. After dataset splitting and feature field standardization preprocessing, a gradient boosting decision tree is used to fit the nonlinear relationship. Then, hyperparameters are tuned through a validation set, thereby improving the prediction accuracy of cooling demand, enhancing the model's generalization ability, maintaining the comparability of feature data, optimizing the model's deployment applicability on the refrigerator control platform, and providing support for precise regulation.

[0112] In some embodiments, the step of adjusting refrigerator parameters based on the call data from the cooling demand prediction model includes:

[0113] When the compressor is turned on, the initial value of the start-up point is obtained;

[0114] The initial value of the start-up point is input into the cooling demand prediction model to predict the cooling demand.

[0115] Adjust compressor parameters based on the predicted cooling demand.

[0116] The above method calls the initial value of the start-up point into the prediction model in real time when the compressor starts, generates the predicted value of cooling demand, and adjusts the compressor operating parameters accordingly. This can improve the response speed of cooling control, enhance the matching degree between the compressor operating status and the actual cooling demand, and reduce the risk of energy consumption fluctuations caused by demand fluctuations.

[0117] In some embodiments, the step of adjusting compressor parameters based on the predicted cooling demand includes:

[0118] Under the condition of prioritizing the achievement of the basic compressor operating minutes, the compressor speed is adjusted and the operating time is extended to meet the predicted cooling demand;

[0119] The compressor parameters are adjusted within the limits of equipment cooling time and temperature difference.

[0120] The above method, while ensuring the basic compressor's operating time, predicts cooling demand by dynamically adjusting the compressor speed and extending the operating time, and implements parameter adjustment within the constraints of equipment cooling time and temperature difference limits. This can improve the matching accuracy between cooling output and demand, maintain system operational stability, and reduce the risk of energy consumption fluctuations caused by excessive or insufficient cooling capacity.

[0121] As can be seen from the above technical solutions, the embodiments of this application provide a method for dynamically predicting the cooling demand of a refrigerator. This method involves acquiring operating condition data and equipment-reported data stored on a refrigerator big data network platform; cleaning the operating condition data and equipment-reported data to obtain cleaned data; extracting feature data representing the cooling demand and initial start-up data from the cleaned data to establish a feature library; training a nonlinear model based on a regression algorithm on the data in the feature library to generate a cooling demand prediction model; and adjusting refrigerator parameters based on the data called by the cooling demand prediction model to solve the problem of inaccurate dynamic prediction of refrigerator cooling demand.

[0122] Similar parts between the embodiments provided in this application can be referred to mutually. The specific implementation methods provided above are only a few examples under the overall concept of this application and do not constitute a limitation on the scope of protection of this application. For those skilled in the art, any other implementation methods extended from the solution of this application without creative effort shall fall within the scope of protection of this application.

Claims

1. A method of dynamically predicting a refrigeration demand of a refrigerator, characterized in that, The method comprises: obtaining working condition data and device reported data stored in a refrigerator big data networking platform; cleaning the working condition data and device reported data to obtain cleaned data; extracting feature data representing refrigeration demand and initial data of starting point from the cleaned data to establish a feature library; training a nonlinear model based on a regression algorithm on the data of the feature library to generate a refrigeration demand prediction model; controlling the parameters of the refrigerator based on the calling data of the refrigeration demand prediction model.

2. The method of claim 1, wherein, Before obtaining the working condition data and device reported data stored in the refrigerator big data networking platform, the method comprises: based on a target model refrigerator configured with an Internet of Things module, collecting continuous time domain data through internal sensors of the refrigerator and reporting to the refrigerator big data networking platform at a preset frequency; the continuous time domain data includes environmental data, setting parameters and sensor temperature data; the environmental data includes environmental temperature and humidity of each compartment of the refrigerator; the setting parameters include set temperature values of each compartment, compressor speed and directional valve direction; the sensor temperature data includes temperature values of each compartment and evaporator temperature values.

3. The method of claim 1, wherein, The step of cleaning the working condition data and device reported data comprises: performing time sorting processing on the working condition data and device reported data, and excluding fault device data in the data; filtering the compressor start period in each refrigerator data.

4. The method of claim 3, wherein, The filtering processing comprises: filtering the compartment data period of the refrigerator with door opening events; filtering the compressor operation period with missing values; removing all data within a preset period after the defrosting and heating stage. 5.The dynamic prediction method of a refrigerator's cooling demand according to claim 2, characterized in that, The step of extracting feature data representing refrigeration demand and initial data of starting point from the cleaned data comprises: randomly extracting a preset number of days of cleaned data by date; identifying the data segment between the start time and the shutdown time of the compressor state as an operation period; calculating the cumulative sum of the compressor speed in the time dimension in each operation period as the feature data representing the refrigeration demand; under the same compressor operation period, extracting initial data of setting parameters and environmental conditions corresponding to the feature data of the compressor starting point. 6.The dynamic prediction method of a refrigerator's cooling demand according to claim 5, characterized in that, The initial data includes set temperature values of each compartment, temperature values of each compartment at the starting time, directional valve direction state, evaporator temperature values and environmental temperature.

7. The method of claim 1, wherein the dynamic prediction of the refrigerator cooling demand is performed based on a user's schedule. The step of training a nonlinear model based on a regression algorithm on the data of the feature library to generate a refrigeration demand prediction model comprises: selecting an XGBoost regressor as a multivariate prediction algorithm based on feature engineering design; the input of the multivariate prediction algorithm is the initial working condition data field of each record in the feature library, and the output is the corresponding refrigeration demand; splitting the data set of the feature library into a training set and a validation set; performing standardization preprocessing on the input feature field; fitting the nonlinear mapping relationship between the input features and the refrigeration demand through gradient boosting decision trees; based on the validation set, the hyperparameters are optimized to generate a refrigeration demand prediction model that can be deployed to a refrigerator control platform. 8.The dynamic prediction method of a refrigerator's cooling demand according to claim 7, characterized in that, The feature engineering includes calculating the difference between the starting temperature of the refrigeration compartment and the set temperature, the difference between the starting temperature of the freezing compartment and the set temperature, the difference between the average ambient temperature and the set temperature, and encoding the equipment number and the reversing valve direction. 9.The dynamic prediction method of a refrigerator's cooling demand according to claim 1, characterized in that, The step of regulating the parameters of the refrigerator based on the calling data of the refrigeration demand prediction model includes: When the compressor is turned on, the initial value of the starting point is obtained; The initial value of the starting point is input into the refrigeration demand prediction model to predict the refrigeration demand; Adjust the compressor parameters based on the predicted refrigeration demand. 10.The dynamic prediction method of a refrigerator's cooling demand according to claim 9, characterized in that, The step of adjusting the compressor parameters based on the predicted refrigeration demand includes: Under the condition of priority to reach the basic compressor starting minutes, adjust the compressor speed and extend the starting time to meet the predicted refrigeration demand; The adjustment of the compressor parameters is carried out without exceeding the equipment refrigeration time limit and the temperature difference limit.

Images