Defrosting method of series-parallel connection dual-system air-cooled refrigerator

Abstract

The invention discloses a defrosting method for a series-parallel dual-system air-cooled refrigerator, which is characterized by comprising the following steps of: acquiring multi-source operation data of the refrigerator, the multi-source operation data comprises evaporator load parameters, motor current data, fan rotating speed, environment temperature and humidity information in a box body, user door opening behavior data and system state data from a series-parallel dual system; a corresponding multi-modal operation state vector is constructed according to the multi-source operation data, the multi-modal operation state vector comprises a plurality of time-aligned sub-vectors, each sub-vector corresponds to one modal type, and the modal types comprise a load modal, an environment modal, a behavior modal and a system modal; and inputting the multi-modal operation state vector into a self-learning defrosting scheduling model constructed based on a graph neural network or a reinforcement learning algorithm. The defrosting frequency and mode can be automatically optimized under complex operation conditions, and the problem of excessive defrosting or defrosting delay is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of air-cooled refrigerators, and in particular to a defrosting method for a series-parallel dual-system air-cooled refrigerator. Background Art

[0002] In recent years, the series-parallel dual-system architecture has become widely adopted in multi-temperature zone refrigerator designs in modern air-cooled refrigerators due to its advantages, including independent temperature control and efficient cooling. This architecture typically features dual evaporators and independent air ducts, enabling flexible switching between refrigerator and freezer modes. However, since frost formation affects the system's heat exchange efficiency, implementing efficient and precise defrost control for both systems is crucial for ensuring overall performance.

[0003] The existing defrosting method for air-cooled refrigerators mainly triggers the defrost process by setting a fixed time interval or a single threshold based on the evaporator temperature. This type of strategy fails to fully consider factors such as changes in ambient temperature and humidity, user door-opening behavior, and system operating load. Especially in the series-parallel dual-system structure, the operating states of the two systems are often not synchronized, resulting in the single threshold judgment being easily inaccurate. Especially under complex operating conditions, misjudgment or response delay is prone to occur, resulting in problems such as untimely or frequent defrosting, affecting the system's thermal efficiency and energy consumption performance. Summary of the Invention

[0004] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract and title of this application to avoid obscuring the purpose of this section, the abstract and the title of the invention, and such simplifications or omissions should not be used to limit the scope of the present invention.

[0005] In view of the problems existing in the above-mentioned defrosting method of the existing series-parallel dual-system air-cooling refrigerator, the present invention is proposed.

[0006] Therefore, the purpose of the present invention is to provide a defrosting method for a series-parallel dual-system air-cooled refrigerator, which is suitable for solving the problem that the existing technology is prone to misjudgment or response lag under complex operating conditions, resulting in untimely or frequent defrosting, affecting the thermal efficiency and energy consumption performance of the system.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, an embodiment of the present invention provides a defrosting method for a series-parallel dual-system air-cooled refrigerator, comprising: Acquire multi-source operating data of the refrigerator, including evaporator load parameters, motor current data, fan speed, internal temperature and humidity information, user door opening behavior data, and system status data from the series-parallel dual system; Constructing a corresponding multimodal operation state vector according to the multi-source operation data, the multimodal operation state vector comprising a plurality of time-aligned sub-vectors, each sub-vector corresponding to a mode type, the mode types comprising load mode, environmental mode, behavioral mode, and system mode; Inputting the multimodal operating state vector into a self-learning defrost scheduling model constructed based on a graph neural network or a reinforcement learning algorithm, the self-learning defrost scheduling model is used to output a current optimal defrost control decision from a plurality of defrost control actions, wherein the defrost control actions include series system defrost, parallel system defrost, alternating defrost, and synchronous defrost; Executing corresponding defrost control instructions based on the defrost control decision output by the self-learning defrost scheduling model; Monitor defrost execution process data, calculate reinforcement learning reward values ​​or loss function indicators for model training based on the data, and update the strategy function or weight parameters of the self-learning defrost scheduling model to achieve adaptive optimization of the defrost strategy.

[0008] As a preferred solution of the defrosting method for a series-parallel dual-system air-cooled refrigerator of the present invention, the step of constructing a corresponding multi-modal operation state vector based on the multi-source operation data includes: When the multi-source operating data includes evaporator operating load, motor current data, and fan speed, extract the time series variation characteristics of each parameter and perform normalization processing to form a corresponding load modal sub-vector for characterizing the heating load and cooling operating status of the dual system; When the multi-source operating data includes ambient temperature and humidity information, features such as the temperature change rate, humidity fluctuation amplitude, and dew point proximity are extracted and normalized to form an environmental modal subvector, which is used to reflect the impact of the internal and external environment of the cabinet on frosting and defrosting; When the multi-source operation data includes user door opening behavior data, extracting user door opening frequency, door opening duration, and corresponding time period behavior features, and performing encoding processing to form a behavior modal subvector, which is used to reflect the effect of user usage habits on temperature and humidity disturbances; When the multi-source operating data includes system status data, extracting features such as system operating mode, fault status, or control parameter status, performing encoding processing, and forming a system modal subvector to reflect the current operating mode and health status of the refrigeration system; The load modal subvector, the environment modal subvector, the behavior modal subvector and the system modal subvector are time-aligned and spliced ​​together to form a complete multi-modal operation state vector, which serves as the input of the self-learning defrost scheduling model.

[0009] As a preferred solution of the defrosting method for a series-parallel dual-system air-cooled refrigerator of the present invention, the multimodal operation state vector is input into a self-learning defrosting scheduling model constructed based on a graph neural network or a reinforcement learning algorithm, including: When the defrost scheduling model is constructed based on a reinforcement learning algorithm, the model determines the corresponding state node according to the current multimodal operation state vector, and selects the defrost control action matching the state from a preset action space as the current defrost control decision; When the defrost scheduling model is constructed based on a graph neural network, the model maps the multimodal operating state vector into a graph structure, where each subvector corresponds to a node in the graph, and the association between each modality type constitutes an edge. Through the aggregation and update mechanism of the graph neural network, a defrost control decision corresponding to the current operating state is generated; The action space includes five defrost control actions: controlling only the series system to perform defrost, controlling only the parallel system to perform defrost, alternating between the series and parallel systems to perform defrost, and simultaneously performing defrost and delayed defrost on the series and parallel systems. The action space is suitable for a self-learning defrost scheduling model constructed using a reinforcement learning algorithm and a graph neural network.

[0010] As a preferred solution of the defrosting method for a series-parallel dual-system air-cooled refrigerator of the present invention, the defrosting method further comprises: If the multi-source operating data includes evaporator operating load, motor current data, fan speed, and ambient temperature and humidity, identifying the operating condition type of the data; When the operating condition type is high-load operation, the evaporator load peak, motor current fluctuation characteristics and fan speed change trend are extracted and encoded to obtain a high-load operation vector representation; When the operating condition type is low-load operation, extract the ambient temperature and humidity change rate and the fan speed stability characteristics, perform encoding processing, and obtain a low-load operation vector representation; splicing the high-load operation vector representation or the low-load operation vector representation with the behavioral modal subvector and the system modal subvector to determine the multimodal operation state vector; Determining the multi-modal operating state vector includes a behavioral mode, a system mode, a high-load operating mode, and a low-load operating mode.

[0011] As a preferred solution of the defrosting method for a series-parallel dual-system air-cooled refrigerator of the present invention, the defrosting method further comprises: If the multi-source operation data includes user door opening behavior data, identifying the behavior type of the user door opening behavior data; When the behavior type is high-frequency door opening, extract the door opening frequency, door opening duration and time series features of the corresponding time period, perform encoding processing, and obtain a high-frequency behavior vector representation; When the behavior type is low-frequency door opening, the door opening time point and the environmental temperature and humidity change characteristics are extracted and encoded to obtain a low-frequency behavior vector representation; splicing the high-frequency behavior vector representation or the low-frequency behavior vector representation with the operation mode sub-vector and the system mode sub-vector to determine the multi-modal operation state vector; Determining the multi-modal operating state vector includes operating mode, system mode, high-frequency behavior mode and low-frequency behavior mode.

[0012] As a preferred solution of the defrosting method for a series-parallel dual-system air-cooled refrigerator of the present invention, the defrosting method further comprises: In a case where the multi-source operation data includes system status data, identifying a status type of the system status data; When the state type is a fault state, extract the fault code, occurrence time and abnormal characteristics of the operating parameters, perform encoding processing, and obtain a fault state vector representation; splicing the fault state vector representation with the operation mode sub-vector and the behavior mode sub-vector to determine the multi-modal operation state vector; Determining the multi-modal operating state vector includes operating mode, behavioral mode, system mode and fault state mode.

[0013] As a preferred solution of the defrosting method for a series-parallel dual-system air-cooled refrigerator according to the present invention, the defrost control decision based on the output of the self-learning defrost scheduling model and the execution of the corresponding defrost control instruction include: Obtaining strategy scoring values ​​of multiple defrost control actions output by the self-learning defrost scheduling model; sorting the defrost control actions according to their strategy scores, and obtaining the defrost control action with the highest strategy score; If the score of the defrost control action with the highest strategy score is greater than a first preset strategy threshold, the defrost control action is used as a target control instruction, and a corresponding defrost operation is performed.

[0014] As a preferred solution of the defrosting method for a series-parallel dual-system air-cooled refrigerator according to the present invention, if the score of the defrost control action with the highest score is less than or equal to the first preset strategy threshold, the method includes: Determining similarity scores between features of the behavioral mode subvector and the environmental mode subvector in the multimodal operating state vector and training samples of the self-learning defrost scheduling model; If the similarity score is greater than a second preset similarity threshold, generating a compensation defrost control action based on the rule-based defrost strategy and executing a corresponding operation; If the similarity score is less than or equal to a second preset similarity threshold, executing the system's built-in fixed-cycle defrost control strategy or the temperature threshold-triggered defrost control strategy to complete the defrost task; The first preset strategy threshold is greater than the second preset similarity threshold.

[0015] In a second aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the defrosting method for a series-parallel dual-system air-cooled refrigerator as described in the first aspect of the present invention is implemented.

[0016] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the defrosting method for a series-parallel dual-system air-cooled refrigerator as described in the first aspect of the present invention is implemented.

[0017] Beneficial effects of the present invention: The present invention constructs a multimodal operating state vector based on multi-source operating data such as evaporator load, motor current and user behavior, thereby achieving a comprehensive characterization of the current operating state of the refrigerator; by constructing a self-learning defrost scheduling model, it can perceive system load changes and environmental disturbances in real time, and dynamically decide on serial system, parallel system, alternating or synchronous defrost operations, so that the defrost operation more accurately matches the current demand; through the feedback data after defrost execution, the reward value or loss index for model training is generated, and the defrost strategy function or weight parameters are continuously optimized, so that the scheduling model has adaptive learning ability, so that the defrost frequency and method can be automatically optimized under complex operating conditions, avoiding the problems of excessive defrost or defrost delay, significantly reducing the overall energy consumption of the refrigerator, and improving operating efficiency; By integrating the global operating information of the series-parallel dual systems and uniformly scheduling multiple defrost actions, the problem of over-frost or temperature zone fluctuation of individual evaporators caused by load imbalance between systems is solved; the self-learning defrost scheduling model constructed by the present invention can adapt to air-cooled refrigerator products with different structural configurations and different user behavior patterns. Through the pre-training + online fine-tuning mechanism, it can be quickly deployed in dual-system refrigerators of different models without frequent manual adjustment of the defrost strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort. Among them: Figure 1 This is a schematic diagram of the overall process of a defrosting method for a series-parallel dual-system air-cooled refrigerator proposed by the present invention; Figure 2 The present invention provides a schematic diagram of a defrost control instruction execution flow for a defrost method for a series-parallel dual-system air-cooled refrigerator. DETAILED DESCRIPTION

[0019] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0020] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0022] Furthermore, the present invention is described in detail with reference to schematic diagrams. For ease of illustration, when describing the embodiments of the present invention, cross-sectional views illustrating device structures may be partially enlarged and not to scale. Furthermore, the schematic diagrams are merely illustrative and should not limit the scope of protection of the present invention. Furthermore, in actual production, the three-dimensional dimensions of length, width, and depth should be included.

[0023] Example 1 Reference Figure 1-Figure 2 , which is an embodiment of the present invention, provides a defrosting method for a series-parallel dual-system air-cooled refrigerator.

[0024] The existing defrosting method for air-cooled refrigerators mainly triggers the defrost process by setting a fixed time interval or a single threshold based on the evaporator temperature. This type of strategy fails to fully consider factors such as changes in ambient temperature and humidity, user door-opening behavior, and system operating load. Especially in the series-parallel dual-system structure, the operating states of the two systems are often not synchronized, resulting in the single threshold judgment being easily inaccurate. Especially under complex operating conditions, misjudgment or response delay is prone to occur, resulting in problems such as untimely or frequent defrosting, affecting the system's thermal efficiency and energy consumption performance.

[0025] The present application provides a method for effectively solving the above-mentioned problems. Next, a plurality of embodiments will be combined to explain in detail how to implement the defrosting method of a series-parallel dual-system air-cooled refrigerator.

[0026] Figure 1 The overall flow chart of a defrosting method for a series-parallel dual-system air-cooled refrigerator is shown, including: S1: Obtain multi-source operating data of the refrigerator, which includes evaporator load parameters, motor current data, fan speed, ambient temperature and humidity information inside the cabinet, user door opening behavior data, and system status data from the series-parallel dual system.

[0027] It should be noted that multi-source operation data includes evaporator load parameters, motor current data, fan speed, ambient temperature and humidity information inside the cabinet, user door opening behavior data, and system status data from the series-parallel dual system. Specifically, multi-source operation data includes the following categories: First, the evaporator load parameters (evaporator surface temperature) and fan speed are collected from the series-parallel dual system to characterize the evaporator's heat transfer load and frost accumulation trends. The evaporator surface temperature directly reflects the decline in heat transfer efficiency, while fluctuations in fan speed can be used to detect changes in air duct resistance caused by frost formation. These two parameters are combined to construct the load modal subvector. Second, the compressor motor current data is collected to reflect changes in the refrigeration load. When the refrigerator enters the frosting period or experiences a sudden load change, abnormal fluctuations in the compressor current serve as an important reference for determining whether defrosting should be initiated, contributing to the construction of the system modal subvector. Third, the ambient temperature and relative humidity data for each temperature zone within the refrigerator are collected. The ambient temperature reflects the insulation effectiveness, while the humidity is directly related to the frost formation rate. This data is used in the model to comprehensively analyze the current frost growth rate and the necessary defrost timing, forming the environmental modal subvector. Furthermore, user door opening behavior data is recorded, including the frequency, duration, and time period of each door opening. This behavioral characteristic is closely related to the heat load disturbance and is the key basis for constructing the behavioral modal subvector. In addition, system status data is also collected, such as the current operating mode (refrigeration, defrosting, standby, etc.), fault status, and some control parameter setting values. This is used to assist in determining the overall operating stage and safety status of the refrigerator, and participates in strategy constraints during the model reasoning process as part of the system modal subvector.

[0028] The above operating data is acquired in real time through sensors deployed in the refrigerator evaporator, compressor, electronic control module, fan, motor controller, temperature and humidity sensor, door switch and other locations, and timestamp synchronization is performed to construct a structured multimodal operating state vector, which serves as the input basis for the subsequent self-learning defrost scheduling model.

[0029] S2: Construct a corresponding multimodal operation state vector based on multi-source operation data. The multimodal operation state vector includes multiple time-aligned sub-vectors, each of which corresponds to a mode type, including load mode, environmental mode, behavioral mode, and system mode.

[0030] The corresponding multi-modal operation state vector is constructed based on multi-source operation data, including: In the case of multi-source operating data including evaporator operating load, motor current data and fan speed, the time series variation characteristics of each parameter are extracted and normalized to form the corresponding load modal sub-vector, which is used to characterize the heating load and cooling operating status of the dual system; When multi-source operating data includes ambient temperature and humidity information, features such as temperature change rate, humidity fluctuation amplitude, and dew point proximity are extracted and normalized to form environmental modal subvectors to reflect the impact of the internal and external environments on frosting and defrosting. When multi-source operating data includes user door opening behavior data, behavioral features such as user door opening frequency, door opening duration, and corresponding time period are extracted and encoded to form behavioral modal subvectors to reflect the impact of user usage habits on temperature and humidity disturbances. When multi-source operating data includes system status data, features such as system operating mode, fault status, or control parameter status are extracted and encoded to form system modal subvectors to reflect the current operating mode and health status of the refrigeration system. The load modal sub-vector, environmental modal sub-vector, behavioral modal sub-vector and system modal sub-vector are time-aligned and spliced ​​together to form a complete multi-modal operating state vector, which serves as the input of the self-learning defrost scheduling model.

[0031] Furthermore, the load modal subvector, environmental modal subvector, behavioral modal subvector and system modal subvector are time-aligned and spliced ​​together, including: first, taking the data mode with the highest sampling frequency as the time reference, linear interpolation or sliding time window resampling is performed on the modal subvectors with lower sampling frequency or missing values ​​to ensure that the four types of modal subvectors are aligned in the same time series; for data points with abnormal or missing timestamps, corrections are made by nearest neighbor extrapolation or multimodal correlation estimation; second, the four types of time-aligned modal subvectors are spliced ​​together in the feature dimension according to a fixed splicing order. The multimodal operating state vectors are spliced ​​in different degrees, and the splicing order is load mode, environmental mode, behavioral mode and system mode. Before splicing, feature weights can be introduced according to the degree of influence of each mode on the defrost strategy for weighted normalization to eliminate the influence of the dimensional differences of different modal features on the model judgment. Finally, the spliced ​​multimodal operating state vector is mapped to the input layer of the self-learning defrost scheduling model, where the input layer can be a fixed-dimensional input structure or a variable-dimensional input structure based on the attention mechanism, so that the model can adaptively extract key modal features in different operating scenarios and generate corresponding defrost control decisions.

[0032] S3: The multimodal operating state vector is input into a self-learning defrost scheduling model built based on a graph neural network or a reinforcement learning algorithm. The self-learning defrost scheduling model is used to output the current optimal defrost control decision from multiple defrost control actions. The defrost control actions include series system defrost, parallel system defrost, alternating defrost, and synchronous defrost.

[0033] The multimodal operating state vector is input into the self-learning defrost scheduling model built based on graph neural network or reinforcement learning algorithm, including: When the defrost scheduling model is constructed based on the reinforcement learning algorithm, the multimodal operating state vector is first input into the reinforcement learning environment as the current environmental state. The environmental state and the action space are associated through a policy network. The policy network can adopt a deep neural network structure to realize the mapping of high-dimensional state to action probability distribution; then, in each decision cycle, the model selects a defrost control action in the action space (defrosting of series system, defrosting of parallel system, alternating defrosting of series and parallel systems, simultaneous defrosting of series and parallel systems, and delayed defrosting) according to the state node, and updates the parameters of the policy network through the reward value fed back by the environment (such as the decrease in the amount of frost on the evaporator after defrosting, the rate of change of energy consumption, the temperature fluctuation amplitude in the box, etc.), so that the defrost strategy adaptively approaches the optimal in long-term operation.

[0034] When the defrost scheduling model is constructed based on a graph neural network, the multimodal operating state vector is first mapped into a graph structure, where the load modal sub-vector, environment modal sub-vector, behavior modal sub-vector, and system modal sub-vector correspond to different types of nodes in the graph respectively; the edges between the nodes are used to represent the association between modal features, such as the heat exchange coupling relationship between the load and the environment, the control logic association between the behavior and the system state, etc.; then, through the message passing and aggregation mechanism of the graph neural network, the node features are iteratively updated to generate a global graph representation vector, which is then input into the defrost decision layer, which outputs the optimal defrost control action under the current operating state.

[0035] Through the above methods, graph neural networks can fully capture the nonlinear correlation between multimodal features, and reinforcement learning can realize online optimization of defrost strategies. Both can be combined with the defrost control mode in the action space to achieve precise scheduling of the defrost process of series-parallel dual-system air-cooled refrigerators.

[0036] The action space includes five defrost control actions: controlling only the series system to perform defrost, controlling only the parallel system to perform defrost, alternating defrost between the series and parallel systems, and simultaneous defrost and delayed defrost between the series and parallel systems. The action space is suitable for a self-learning defrost scheduling model constructed using reinforcement learning algorithms and graph neural networks.

[0037] If the multi-source operating data includes evaporator operating load, motor current data, fan speed, and ambient temperature and humidity, identify the operating condition type of the data; When the operating condition type is high-load operation, the evaporator load peak, motor current fluctuation characteristics and fan speed change trend are extracted and encoded to obtain the high-load operation vector representation; When the operating condition type is low-load operation, the environmental temperature and humidity change rate and fan speed stability characteristics are extracted and encoded to obtain the low-load operation vector representation; splicing the high-load operation vector representation or the low-load operation vector representation with the behavioral mode sub-vector and the system mode sub-vector to determine a multi-modal operation state vector; The multi-modal operation state vector is determined to include behavioral mode, system mode, high-load operation mode and low-load operation mode.

[0038] In this embodiment, the operating condition type can be determined by performing time-series sampling and preprocessing on the evaporator operating load, motor current data, fan speed, and ambient temperature and humidity data. Specifically, the average and peak values ​​of the evaporator operating load can be calculated within a preset sampling window to reflect the system's thermal load level before defrosting. The fluctuation amplitude, standard deviation, and frequency distribution of the motor current data can be extracted to characterize the power fluctuation characteristics of the compressor or fan under different operating conditions. Trend analysis of the fan speed data can be performed to determine its upward, downward, or stable trend within the operating cycle. The rate of change and correlation coefficient of the ambient temperature and humidity data can be calculated to quantify the impact of external environmental changes on the frosting rate.

[0039] When it is comprehensively determined that the evaporator operating load is close to the rated upper limit, the motor current fluctuates significantly, and the fan speed shows a continuous upward trend, it can be determined as high-load operation, and the extracted load peak, current fluctuation characteristics, and speed change trend are normalized to generate a high-load operation vector representation; when the evaporator operating load is in a low value range, the ambient temperature and humidity change slowly, and the fan speed fluctuation amplitude is small, it can be determined as low-load operation, and the temperature and humidity change rate and fan speed stability characteristics are normalized to generate a low-load operation vector representation.

[0040] Finally, the high-load operation vector representation or the low-load operation vector representation is vector-concatenated with the behavioral mode sub-vector and the system mode sub-vector. The resulting multi-modal operation state vector will contain behavioral mode, system mode and high / low load operation mode information, thereby providing comprehensive input features for the accurate decision-making of the self-learning defrost scheduling model.

[0041] If the multi-source operation data includes user door opening behavior data, identify the behavior type of the user door opening behavior data; When the behavior type is high-frequency door opening, the door opening frequency, door opening duration and time series features of the corresponding time period are extracted and encoded to obtain a high-frequency behavior vector representation; When the behavior type is low-frequency door opening, the door opening time point and the environmental temperature and humidity change characteristics are extracted and encoded to obtain the low-frequency behavior vector representation; splicing the high-frequency behavior vector representation or the low-frequency behavior vector representation with the operation mode sub-vector and the system mode sub-vector to determine a multi-modal operation state vector; Determining the multi-modal operating state vector includes operating mode, system mode, high-frequency behavioral mode and low-frequency behavioral mode.

[0042] In this embodiment, the behavior type of user door-opening behavior data is determined by sampling and extracting features from time-series data on door-opening frequency, door-opening duration, and corresponding time periods. Specifically, the frequency distribution of door-opening times and the mean and variance of single door-opening durations are calculated within a preset statistical period to reflect the user's activeness in refrigerator usage. Cluster analysis is performed on door-opening time points to identify whether they exhibit regular distribution patterns. Furthermore, correlation analysis is performed between door-opening behavior and the temporal changes in ambient temperature and humidity data to quantify the impact of user behavior on the stability of the refrigerator's internal environment.

[0043] When it is comprehensively determined that the door opening frequency exceeds the preset threshold, the door opening duration increases significantly, and the door opening time points show a dense distribution feature, it can be identified as a high-frequency door opening behavior, and the extracted door opening frequency statistics, duration distribution characteristics, and time point clustering results are encoded to generate a high-frequency behavior vector representation; when the door opening frequency is in a low value range, the door opening time points are scattered and have no significant correlation with changes in ambient temperature and humidity, it can be identified as a low-frequency door opening behavior, and the door opening time point distribution characteristics and ambient temperature and humidity correlation analysis results are encoded to generate a low-frequency behavior vector representation.

[0044] Finally, the high-frequency behavior vector representation or the low-frequency behavior vector representation is vector-concatenated with the operating mode sub-vector and the system mode sub-vector to form a multi-modal operating state vector that contains operating mode, system mode, and high / low-frequency behavior mode information, thereby providing feature input reflecting user usage habits for the self-learning defrost scheduling model.

[0045] In a case where the multi-source operation data includes system status data, identifying a status type of the system status data; When the state type is a fault state, the fault code, occurrence time and abnormal operating parameter features are extracted and coded to obtain a fault state vector representation; The fault state vector representation is concatenated with the operation mode sub-vector and the behavior mode sub-vector to determine a multi-modal operation state vector; Determine the multi-modal operation state vector including operation mode, behavior mode, system mode and fault state mode.

[0046] In this embodiment, the status type of system status data is determined through a comprehensive analysis of the fault code, occurrence time, and abnormal operating parameter characteristics. Specifically, the system monitors the deviation of system operating parameters from preset thresholds in real time, recording the type of fault code and triggering sequence. The system also extracts the amplitude and duration of abnormal fluctuations in key parameters such as compressor current and evaporator temperature. Furthermore, the system analyzes the correlation between the fault occurrence time and the system operating mode and load status to determine the severity and scope of the fault.

[0047] When a fault code is detected to be effectively triggered and key operating parameters continue to deviate from the normal range, it can be identified as a fault state, and the extracted fault code type, abnormal parameter fluctuation characteristics and occurrence time series are uniquely encoded and normalized to generate a fault state vector representation; when the system operating parameters are within the normal fluctuation range and there are no valid fault code records, it is determined to be a normal operating state and is represented by a standard system modal sub-vector.

[0048] Finally, the fault state vector representation or standard system modal subvector is time-aligned and feature-dimensionally spliced ​​with the operating modal subvector and behavioral modal subvector. The resulting multimodal operating state vector will contain complete information on the operating mode, behavioral mode, system mode, and fault state mode, thereby ensuring that the self-learning defrost scheduling model can accurately identify system anomalies and make corresponding defrost strategy adjustments.

[0049] S4: Based on the defrost control decision output by the self-learning defrost scheduling model, execute the corresponding defrost control instruction.

[0050] Preferably, Figure 2 FIG. 1 is a schematic diagram of a defrost control instruction execution flow of a defrost method for a series-parallel dual-system air-cooled refrigerator proposed by the present invention, including: Obtaining strategy scoring values ​​of multiple defrost control actions output by the self-learning defrost scheduling model; The strategy score is calculated by the strategy network or decision layer of the self-learning defrost scheduling model according to a unified scoring function. The scoring function is used to comprehensively evaluate the expected benefits and costs of candidate defrost actions within a future time window. The specific formula is as follows: ; in, is the strategy score, a is the given candidate action, is the predicted value of the future benefit of action a (such as frost reduction, cooling recovery rate, etc.) based on the current multimodal operation state vector and the self-learning defrost scheduling model, which comes from the value network in reinforcement learning or the reward estimation of the self-learning defrost scheduling model. is the normalized value of the estimated energy consumption of action a, is the normalized estimate of the box temperature deviation (adverse effect) caused by action a, and the parameter 、 and is a weight coefficient, which can be determined through cross-validation or online parameter adjustment during the self-learning defrost scheduling model or engineering deployment. The normalization of the above items uses training set statistics (such as minimum-maximum normalization or mean-variance normalization) to ensure consistency in the numerical scale.

[0051] Sort each defrost control action according to its strategy score, and obtain the defrost control action with the highest strategy score; If the score of the defrost control action with the highest strategy score is greater than the first preset strategy threshold, the defrost control action is used as the target control instruction, and the corresponding defrost operation is performed.

[0052] If the score of the defrost control action with the highest score is less than or equal to the first preset strategy threshold, the method includes: Determine the similarity score between the characteristics of the behavioral mode subvector and the environmental mode subvector in the multimodal operation state vector and the training samples of the self-learning defrost scheduling model; It should be noted that the matching degree can be calculated by concatenating the current behavioral mode sub-vector and the environmental mode sub-vector to form a candidate vector u, and a set of representative sample prototypes saved during the training of the self-learning defrosting scheduling model. (Each Calculate the similarity score for the feature center in a certain type of environment-behavior scenario. The specific formula is as follows: ; in, Score the similarity. is the vector norm; If the similarity score is greater than a second preset similarity threshold, a compensation defrost control action is generated based on the rule-based defrost strategy, and a corresponding operation is performed; If the similarity score is less than or equal to the second preset similarity threshold, the system's built-in fixed-cycle defrost control strategy or the temperature threshold-triggered defrost control strategy is executed to complete the defrost task; The first preset strategy threshold is greater than the second preset similarity threshold.

[0053] After the "compensatory defrost control action" and the backup action are confirmed by the decision-making layer, the control execution module maps the action into a specific control instruction sequence and issues it. For example, it sends switch and power settings to the heater, controls the fan speed or temporarily adjusts the compressor operating mode, sets the defrost duration and sequence (which system to defrost first, whether to defrost alternately, or simultaneously), sets the upper limit of the safe temperature and the maximum allowable defrost duration, etc. (Such control instructions are issued by the main control unit in the form of software commands / control register writes and executed by the corresponding execution unit). During the execution of the action, the system continuously records key indicators of the execution process (actual energy consumption, defrost duration, temperature recovery curve after defrost, user interaction impact, etc.), and transmits the recorded data back to the model training module to be used as a reward signal or supervision sample for reinforcement learning to update the policy network or rule base, completing the closed loop of online learning and offline retraining.

[0054] To ensure the reliability of control decisions, the system should also include anomaly detection and safety constraints: when the control instruction may cause the temperature to exceed the safety upper limit or the control unit responds abnormally, the current defrost action is immediately terminated and switched to the safety strategy with the lowest risk (such as shutdown waiting, alarm or short-term insulation mode); at the same time, the model monitors overfitting, distribution drift and other situations during training or online update and triggers model rollback or manual intervention mechanism.

[0055] In the embodiment of the present application, the weight coefficient 、 and The determination of can be determined in the following manner: using a gradient descent method or a Bayesian optimization algorithm, with the goal of minimizing the long-term energy consumption and temperature fluctuation of the defrost strategy, automatically adjusting the weight coefficient on the training set to ensure that the sum of the weighted contributions of each feature (benefit, energy consumption, temperature deviation) is 1, thereby balancing the defrost effect and system stability, or presetting an initial value based on engineering experience, and then adapting to the operating characteristics of a specific refrigerator model through online fine-tuning; the first preset strategy threshold and the second preset similarity threshold can be determined by calculating the distribution of the score values ​​of all candidate action strategies in the validation set during the model training phase, selecting a specific percentile as the first preset strategy threshold to ensure the execution of high-confidence actions, or clustering the behavior-environment modal features of the training samples, and using the median of the inter-cluster distance as the second preset similarity threshold to ensure that the compensation strategy is triggered only when it highly matches the historical scenario; the method for determining the weight coefficient and threshold in this embodiment is not limited to the above method, and those skilled in the art may adopt other equivalent technical means to achieve this according to actual needs.

[0056] S5: Monitor the defrost execution process data, calculate the reinforcement learning reward value or loss function indicator for model training based on the data, and update the strategy function or weight parameters of the self-learning defrost scheduling model to achieve adaptive optimization of the defrost strategy.

[0057] Preferably, the defrost execution process data is measured, and based on the data, a reinforcement learning reward value or loss function indicator for model training is calculated, and the strategy function or weight parameter of the self-learning defrost scheduling model is updated to achieve adaptive optimization of the defrost strategy, including: S5.1. Monitoring defrost execution process data: After executing the defrost control instruction, focus on detecting operating data directly related to the defrost effect (including the evaporator surface temperature change rate, which is used to reflect the completion rate of the defrost process, the total power consumption of the electric heater and fan during the defrost process, which is used to evaluate energy efficiency, the temperature fluctuation amplitude of the refrigerator and freezer compartments after defrosting, which reflects the impact of defrosting on the temperature inside the box, and the time it takes for the evaporator load to return to normal operating conditions after defrosting, which is used to evaluate system stability). These operating data are pre-processed and used for subsequent analysis.

[0058] S5.2-1. Calculate the reinforcement learning reward value or loss function index: Based on the monitored defrost execution data, calculate the reinforcement learning reward value or loss function index used for self-learning defrost scheduling model training to evaluate the effectiveness of the defrost control action. The specific calculation method is as follows: For the self-learning defrost scheduling model based on the reinforcement learning algorithm, the system quantifies the defrost effect through a reward function. The reward function comprehensively considers defrost efficiency, energy consumption, and temperature stability. The specific formula is as follows: ; in, Indicates the reward value at time t, which represents the comprehensive effect of the defrosting action. is the defrost efficiency, which is defined as the ratio of the evaporator surface temperature change rate to the target rate and is dimensionless after normalization; is the normalized energy consumption, which is defined as the ratio of the actual energy consumption to the maximum expected energy consumption during the defrosting process, dimensionless; is the temperature fluctuation amplitude, defined as the ratio of the temperature fluctuation in the refrigerator and freezer compartments after defrosting to the maximum allowable fluctuation. It is dimensionless and ranges from [0, 1]. K1, K2, and K3 are weight coefficients, representing the relative importance of defrosting efficiency, energy consumption, and temperature stability, respectively. K1 + K2 + K3 = 1, which can be set through historical data optimization or expert experience. S5.2-2. Loss function indicator calculation: For the self-learning defrost scheduling model based on graph neural network (GNN), the system uses the loss function to evaluate the deviation between the predicted defrost control decision and the actual effect. The specific formula is as follows: ; in, is the loss function value at the tth moment, reflecting the prediction deviation, is the evaporator surface temperature when defrosting ends. is the target defrost termination temperature, is the actual energy consumption during the defrosting process, is the temperature fluctuation amplitude of the refrigerator and freezer after defrosting. m1, m2 and m3 are weight coefficients, which are used to balance the contribution of temperature deviation, energy consumption and temperature fluctuation respectively. The three can be optimized by gradient descent method.

[0059] S5.3. Update the self-learning defrost scheduling model: Based on the calculated reward value (reinforcement learning model) or loss function indicator (graph neural network model), the system updates the model's policy function or weight parameters to optimize the defrost strategy; S5.3-1: In the reinforcement learning model, the system uses the policy gradient algorithm to update the policy function. The specific contents are as follows: Based on reward value , calculate the cumulative expected reward of the strategy, the specific formula is: ; in, is the cumulative return at time t, is the discount factor, T is the end time of the defrost cycle, and k is the time step index, which represents each discrete time point from the current time t to the end time T; By maximizing cumulative returns , the policy gradient method is used to update the parameters of the policy function, so that the self-learning defrost scheduling model tends to select the defrost control action with higher reward. The update formula is as follows: ; in, are the parameters of the policy function, is the learning rate, is the objective function of the strategy, calculated based on the cumulative return, is the objective function gradient; S5.3-2: In the graph neural network model, the system updates the weight parameters of the graph neural network through the backpropagation algorithm. The specific steps include: Based on the loss function , calculate the error of the current defrost control strategy; update the weight parameters of the graph neural network through the gradient descent method. The specific formula is: ; in, is the weight parameter of the graph neural network, is the gradient of the loss function; S5.3-3: To ensure the adaptability of the model, the system supports both online fine-tuning and offline training modes; Online fine-tuning: During refrigerator operation, defrost execution data is collected in real time and model parameters are updated in small batches to adapt to changes in the current operating environment and user behavior patterns. Offline training: Use historical defrost data (for example, operating data from the past 30 days) for batch training to optimize the model's initial parameters and improve the model's generalization ability under different operating conditions.

[0060] S5.3: Through the above steps, the system can dynamically adjust the defrost strategy based on the defrost execution effect. For example, if low defrost efficiency is detected, the system will increase the score of the alternating defrost or synchronous defrost strategies, giving priority to the more efficient defrost action. If energy consumption is too high, the system will reduce the score of high-energy-consuming actions (such as synchronous defrost) and tend to choose delayed defrost or single-system defrost. If frequent user door openings are detected, resulting in temperature and humidity fluctuations, the system will adjust the weight of the behavioral mode subvector, prioritizing the impact of environmental modes on defrost decisions. Through continuous monitoring and updating, the model can adaptively optimize the defrost frequency and method, avoiding excessive or insufficient defrost, thereby significantly reducing energy consumption under complex operating conditions.

[0061] For example, suppose a series-parallel dual-system air-cooled refrigerator used in a household is operated in a high temperature environment in summer (ambient temperature 30°C, relative humidity 80%), and the system detects that the load of the freezer evaporator continues to increase (the load value reaches 85%), and the user opens the door frequently (4 times per hour, each door opening lasts 15 seconds). The self-learning defrost scheduling model selects alternating defrost control actions based on the multi-modal operating state vector (including load mode, environmental mode, behavioral mode and system mode). During the defrost execution process, the system monitors that the freezer temperature fluctuation is controlled at ±0.4°C, the defrost energy consumption is 0.12 kWh, and the defrost efficiency (evaporator surface temperature recovery rate) reaches 0.85 (normalized value). The reward value is calculated through reinforcement learning reward function. The score was 0.78, indicating that this defrost action was effective. The system then updated the policy function, increasing the priority of alternating defrost. However, during subsequent operation, if energy consumption increased to 0.18 kWh (exceeding 115% of the expected value), the self-learning defrost scheduling model lowered the score of the synchronous defrost strategy and switched to a delayed defrost strategy, thereby optimizing energy consumption and maintaining temperature stability in the refrigerator and freezer compartments. Ultimately, temperature fluctuations were controlled to ±0.3°C, and energy consumption was reduced to 0.10 kWh. This process was achieved by online fine-tuning of model parameters to ensure that the defrost strategy adapted to user behavior and environmental changes.

[0062] In the embodiment of the present application, the initial values ​​of the weight coefficients K1, K2, and K3 can be determined by statistically analyzing a large amount of operating data of the refrigerator under various operating conditions (such as high load, low load, high frequency door opening, and low frequency door opening), combined with the optimization goals of energy efficiency and temperature stability. In actual application, these weights can be dynamically optimized according to the specific operating environment through online fine-tuning (based on the gradient descent method); m1, m2, and m3 correspond to temperature deviation, energy consumption, and temperature fluctuation, respectively. Their initial values ​​are determined through simulation experiments; the discount factor Can be determined by analyzing the long-term impact of rewards during the defrost cycle; learning rate Verified and confirmed through experiments. In laboratory tests, the convergence speed and stability of the model parameters after updating are observed through the small batch gradient descent method, and the values ​​that can quickly adapt to changes in operating status and avoid overfitting are selected. In practical applications, the adaptive learning rate algorithm and the policy function parameters can be used to The initial value is determined by a pre-trained model, and a reinforcement learning model (such as the PPO algorithm) is trained based on a large amount of simulated data (covering different loads, environments, and user behaviors) to generate an initial strategy. The initial value of the graph neural network parameter W is determined by offline training of the graph neural network. The training data includes multimodal operating state vectors (load, environment, behavior, system mode) and their corresponding defrost effects. In addition, the initial values ​​of the above parameters can be preliminarily set through small-scale test data, and then optimized through continuous monitoring and feedback.

[0063] It should be noted that the defrosting method of traditional air-cooled refrigerators usually uses a fixed time interval or a single temperature threshold trigger, which is difficult to adapt to changes in ambient temperature and humidity, user door opening behavior and system load differences. In particular, in a series-parallel dual-system structure, the operating states of the two systems are not synchronized, which can easily lead to untimely or frequent defrosting, affecting thermal efficiency and energy consumption. This embodiment introduces a self-learning defrost scheduling model based on a multimodal operating state vector, combined with reinforcement learning or graph neural networks, to achieve dynamic defrosting decisions. Taking into account multi-dimensional factors such as load, environment, behavior and system status, a refined control system for series, parallel, alternating, synchronous and delayed defrosting is constructed. By monitoring defrost execution data in real time and updating model parameters, the system can adaptively optimize the defrost frequency and method, overcoming the static and single nature of traditional methods and significantly improving the accuracy and efficiency of defrosting. It effectively avoids the decline in refrigeration efficiency due to frost accumulation or the waste of energy due to frequent defrosting. The present invention solves the problem of dual-system load imbalance through global operation information fusion and online fine-tuning mechanism, adapts to refrigerator products with different structural configurations and user behavior patterns, greatly improves operation efficiency and energy utilization, and reduces maintenance costs.

[0064] In summary, the present invention constructs a multimodal operating state vector based on multi-source operating data such as evaporator load, motor current and user behavior, thereby achieving a comprehensive characterization of the current operating state of the refrigerator; by constructing a self-learning defrost scheduling model, it can perceive system load changes and environmental disturbances in real time, and dynamically decide on serial system, parallel system, alternating or synchronous defrost operations, so that the defrost operation can more accurately match the current demand; through the feedback data after defrost execution, the reward value or loss index for model training is generated, and the defrost strategy function or weight parameters are continuously optimized, so that the scheduling model has adaptive learning ability, so that in complex operation The defrost frequency and mode can be automatically optimized under any conditions to avoid excessive defrosting or defrosting delay, significantly reduce the overall energy consumption of the refrigerator, and improve operating efficiency; by integrating the global operating information of the series-parallel dual systems, multiple defrost actions are uniformly scheduled, which solves the problems of over-frost or temperature zone fluctuation of individual evaporators caused by load imbalance between systems; the self-learning defrost scheduling model constructed by the present invention can adapt to air-cooled refrigerator products with different structural configurations and different user behavior patterns. Through pre-training + online fine-tuning mechanism, it can be quickly deployed in dual-system refrigerators of different models without frequent manual adjustment of the defrost strategy.

[0065] Example 2 This is an embodiment of the present invention, which is different from the previous embodiment in that: If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the method of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0066] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0067] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting, or processing it in another suitable manner as necessary, and then storing it in a computer memory.

[0068] It should be understood that various components of the present invention may be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods may be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof may be used: a discrete logic circuit having logic gate circuits for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc.

[0069] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A defrosting method for a series-parallel dual-system air-cooled refrigerator, characterized in that: include: Acquire multi-source operating data of the refrigerator, including evaporator load parameters, motor current data, fan speed, internal temperature and humidity information, user door opening behavior data, and system status data from the series-parallel dual system; Constructing a corresponding multimodal operation state vector according to the multi-source operation data, the multimodal operation state vector comprising a plurality of time-aligned sub-vectors, each sub-vector corresponding to a mode type, the mode types comprising load mode, environmental mode, behavioral mode, and system mode; Inputting the multimodal operating state vector into a self-learning defrost scheduling model constructed based on a graph neural network or a reinforcement learning algorithm, the self-learning defrost scheduling model is used to output a current optimal defrost control decision from a plurality of defrost control actions, wherein the defrost control actions include series system defrost, parallel system defrost, alternating defrost, and synchronous defrost; Executing corresponding defrost control instructions based on the defrost control decision output by the self-learning defrost scheduling model; Monitor defrost execution process data, calculate reinforcement learning reward values ​​or loss function indicators for model training based on the data, and update the strategy function or weight parameters of the self-learning defrost scheduling model to achieve adaptive optimization of the defrost strategy.

2. The defrosting method for a series-parallel dual-system air-cooled refrigerator according to claim 1, characterized in that: The step of constructing a corresponding multi-modal operation state vector according to the multi-source operation data includes: When the multi-source operating data includes evaporator operating load, motor current data, and fan speed, extracting time series variation characteristics of each parameter and performing normalization processing to form a corresponding load modal sub-vector; When the multi-source operating data includes ambient temperature and humidity information, extracting features such as the temperature change rate, humidity fluctuation amplitude, and dew point proximity, performing normalization processing, and forming an environmental modal subvector; In the case where the multi-source operation data includes user door opening behavior data, extracting user door opening frequency, door opening duration, and corresponding time period and other behavioral features, and performing encoding processing to form a behavioral modal sub-vector; In the case where the multi-source operation data includes system state data, extracting features such as system operation mode, fault state or control parameter state, performing encoding processing, and forming a system modal subvector; The load modal subvector, the environment modal subvector, the behavior modal subvector and the system modal subvector are time-aligned and spliced ​​together to form a complete multi-modal operation state vector, which serves as the input of the self-learning defrost scheduling model.

3. The defrosting method for a series-parallel dual-system air-cooled refrigerator according to claim 1, characterized in that: The multimodal operation state vector is input into a self-learning defrost scheduling model based on a graph neural network or a reinforcement learning algorithm, including: When the defrost scheduling model is constructed based on a reinforcement learning algorithm, the model determines the corresponding state node according to the current multimodal operation state vector, and selects the defrost control action matching the state from a preset action space as the current defrost control decision; When the defrost scheduling model is constructed based on a graph neural network, the model maps the multimodal operating state vector into a graph structure, where each subvector corresponds to a node in the graph, and the association between each modality type constitutes an edge. Through the aggregation and update mechanism of the graph neural network, a defrost control decision corresponding to the current operating state is generated; The action space includes five defrost control actions: controlling only the series system to perform defrost, controlling only the parallel system to perform defrost, alternating between the series and parallel systems to perform defrost, and simultaneously performing defrost and delayed defrost on the series and parallel systems. The action space is suitable for a self-learning defrost scheduling model constructed using a reinforcement learning algorithm and a graph neural network.

4. The defrosting method for a series-parallel dual-system air-cooled refrigerator according to claim 3, characterized in that: The defrosting method further comprises: If the multi-source operating data includes evaporator operating load, motor current data, fan speed, and ambient temperature and humidity, identifying the operating condition type of the data; When the operating condition type is high-load operation, the evaporator load peak, motor current fluctuation characteristics and fan speed change trend are extracted and encoded to obtain a high-load operation vector representation; When the operating condition type is low-load operation, extract the ambient temperature and humidity change rate and the fan speed stability characteristics, perform encoding processing, and obtain a low-load operation vector representation; splicing the high-load operation vector representation or the low-load operation vector representation with the behavioral modal subvector and the system modal subvector to determine the multimodal operation state vector; Determining the multi-modal operating state vector includes a behavioral mode, a system mode, a high-load operating mode, and a low-load operating mode.

5. The defrosting method for a series-parallel dual-system air-cooled refrigerator according to claim 3, characterized in that: The defrosting method further comprises: If the multi-source operation data includes user door opening behavior data, identifying the behavior type of the user door opening behavior data; When the behavior type is high-frequency door opening, extract the door opening frequency, door opening duration and time series features of the corresponding time period, perform encoding processing, and obtain a high-frequency behavior vector representation; When the behavior type is low-frequency door opening, the door opening time point and the environmental temperature and humidity change characteristics are extracted and encoded to obtain a low-frequency behavior vector representation; splicing the high-frequency behavior vector representation or the low-frequency behavior vector representation with the operation mode sub-vector and the system mode sub-vector to determine the multi-modal operation state vector; Determining the multi-modal operating state vector includes operating mode, system mode, high-frequency behavior mode and low-frequency behavior mode.

6. The defrosting method for a series-parallel dual-system air-cooled refrigerator according to claim 3, characterized in that: The defrosting method further comprises: In a case where the multi-source operation data includes system status data, identifying a status type of the system status data; When the state type is a fault state, extract the fault code, occurrence time and abnormal characteristics of the operating parameters, perform encoding processing, and obtain a fault state vector representation; splicing the fault state vector representation with the operation mode sub-vector and the behavior mode sub-vector to determine the multi-modal operation state vector; Determining the multi-modal operating state vector includes operating mode, behavioral mode, system mode and fault state mode.

7. The defrosting method for a series-parallel dual-system air-cooled refrigerator according to claim 1, characterized in that: The defrost control decision based on the output of the self-learning defrost scheduling model and executing the corresponding defrost control instruction include: Obtaining strategy scoring values ​​of multiple defrost control actions output by the self-learning defrost scheduling model; sorting the defrost control actions according to their strategy scores, and obtaining the defrost control action with the highest strategy score; If the score of the defrost control action with the highest strategy score is greater than a first preset strategy threshold, the defrost control action is used as a target control instruction, and a corresponding defrost operation is performed.

8. The defrosting method for a series-parallel dual-system air-cooled refrigerator according to claim 7, characterized in that: If the score of the defrost control action with the highest score is less than or equal to the first preset strategy threshold, the method includes: Determining similarity scores between features of the behavioral mode subvector and the environmental mode subvector in the multimodal operating state vector and training samples of the self-learning defrost scheduling model; If the similarity score is greater than a second preset similarity threshold, generating a compensation defrost control action based on the rule-based defrost strategy and executing a corresponding operation; If the similarity score is less than or equal to a second preset similarity threshold, executing the system's built-in fixed-cycle defrost control strategy or the temperature threshold-triggered defrost control strategy to complete the defrost task; The first preset strategy threshold is greater than the second preset similarity threshold.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the defrosting method for the series-parallel dual-system air-cooled refrigerator according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the defrosting method for a series-parallel dual-system air-cooled refrigerator according to any one of claims 1 to 8 are implemented.

Images