Air conditioner, model training method and server

Abstract

The embodiment of the invention provides an air conditioner, a model training method and a server, relates to the technical field of air conditioners, and is used for reducing the complexity of a fault detection model of the air conditioner while keeping the high precision of the fault detection model of the air conditioner. The air conditioner comprises the steps that operation information of the air conditioner is input into a second fault detection model matched with the area where the air conditioner is located; the second fault detection model is trained by the following steps: acquiring an original data set; inputting the plurality of operation data into a first fault detection model, and obtaining a first fault feature extracted by the first fault detection model; constructing a reconstruction data set according to the plurality of first fault features and the plurality of operation data; the reconstructed data set is divided according to the area where the air conditioner is located to obtain reconstructed sub-data sets; and an initial second fault detection model is trained according to the reconstructed sub-data set, a trained second fault detection model is obtained, and the parameter quantity of the second fault detection model is smaller than that of the first fault detection model.

Description

Technical Field

[0001] This application relates to the field of air conditioning technology, and in particular to an air conditioner, a model training method, and a server. Background Technology

[0002] With the continuous development of artificial intelligence technology, deep learning and other technologies have become the mainstream solutions in the field of air conditioner fault detection due to their powerful feature learning and pattern recognition capabilities.

[0003] Currently, fault detection in air conditioners typically relies on fault detection models and air conditioner operating data. However, with the increasing volume and complexity of sensor data, the structural and computational complexity of fault detection models also rises, making the training process more complex and significantly increasing the demand for computing resources. This poses challenges to the deployment and application of fault detection models on low-configuration hardware platforms.

[0004] Therefore, how to reduce the complexity of the fault detection model of air conditioners while maintaining its high accuracy has become an urgent technical problem to be solved. Summary of the Invention

[0005] This application provides an air conditioner, a model training method, and a server, which reduces the complexity of the air conditioner's fault detection model while maintaining its high accuracy.

[0006] To achieve the above objectives, this application adopts the following technical solution.

[0007] In a first aspect, embodiments of this application provide an air conditioner, the air conditioner comprising: at least one processor; and at least one computer memory operatively connected to the at least one processor and storing instructions, which, when executed by the at least one processor, perform the following operations:

[0008] The operating information of the air conditioner is input into a second fault detection model that matches the area where the air conditioner is located, and the fault detection result output by the second fault detection model is obtained; wherein, the second fault detection model is trained in the following way:

[0009] Obtain the raw dataset, which includes the operating data of multiple air conditioners in multiple regions, with each data point corresponding to a fault label.

[0010] Multiple operational data are input into the first fault detection model, and the first fault feature extracted by the first fault detection model is obtained; the first fault detection model is used to extract fault features from the operational data, and to determine the fault type corresponding to the operational data based on the extracted first fault features;

[0011] A reconstructed dataset is constructed based on multiple first fault features and multiple operational data. Each first fault feature or operational data in the reconstructed dataset corresponds to a fault label.

[0012] The reconstructed dataset is divided according to the region where the air conditioner is located, resulting in multiple reconstructed sub-datasets, each corresponding to a region;

[0013] The initial second fault detection model is trained based on the reconstructed subset dataset to obtain the trained second fault detection model. The second fault detection model is used to identify the fault type of the air conditioner in the corresponding area of ​​the reconstructed subset dataset. The number of parameters of the second fault detection model is less than that of the first fault detection model.

[0014] The technical solution provided in this application has at least the following beneficial effects: This application provides an air conditioner that combines the first fault features extracted by the first fault detection model before the activation function layer with the operating data in the original dataset to construct an enhanced reconstructed dataset. This reconstructed dataset is then used as training samples to train an initial second fault detection model with fewer parameters. Since the first fault features are extracted before the activation function layer, they can more comprehensively characterize the inherent characteristics of the air conditioner's operating data, enabling the second fault detection model to learn more detailed air conditioner operating status information during training, thereby enhancing the accuracy of the second fault detection model. Consequently, when performing fault detection on the air conditioner based on the second fault detection model that matches the area where the air conditioner is located, the fault detection of the air conditioner can be more accurate.

[0015] In addition, this application can divide the reconstructed dataset based on the region where the air conditioner is located, and use the divided reconstructed subset dataset as training samples. This not only reduces the complexity of a single training dataset (i.e., the reconstructed subset dataset), but also enables the second fault detection model to perform more accurate fault detection based on the characteristics of air conditioners located in specific regions.

[0016] Meanwhile, since the second fault detection model has fewer parameters than the first fault detection model, it can reduce the complexity of the fault detection model while maintaining its high accuracy.

[0017] In some embodiments, training an initial second fault detection model based on a reconstructed subset dataset to obtain a trained second fault detection model includes: using first fault features and running data in the reconstructed subset dataset as first sample data, and determining a first fault feature corresponding to each first sample data; inputting the first sample data into the initial second fault detection model, extracting fault features from the first sample data using the initial second fault detection model, and obtaining a predicted fault type corresponding to the first sample data based on the extracted second fault features; generating a first loss function based on the predicted fault type and fault label corresponding to the first sample data; generating a second loss function based on the second fault features of the first sample data and the first fault features corresponding to the first sample data; and training the initial second fault detection model based on the first loss function and the second loss function to obtain a trained second fault detection model.

[0018] In some embodiments, the instructions further perform the following operations: inputting the air conditioner's operating information into a third fault detection model that matches the type of air conditioner, and obtaining the fault detection result output by the third fault detection model; wherein, the third fault detection model is trained in the following manner: dividing the reconstructed subset according to the type of air conditioner to obtain multiple subsets, each subset corresponding to a type of air conditioner; training the initial third fault detection model based on the subsets to obtain a trained third fault detection model, the third fault detection model being used to identify the fault type of the air conditioner corresponding to the type of the subset, and the number of parameters of the third fault detection model being less than the number of parameters of the second fault detection model.

[0019] In some embodiments, training an initial third fault detection model based on a subset of data to obtain a trained third fault detection model includes: using a first fault feature and operational data from the subset of data as second sample data, and determining a second fault feature corresponding to each second sample data; the second fault feature is extracted by the second fault detection model; inputting the second sample data into the initial third fault detection model, extracting fault features from the second sample data using the initial third fault detection model, and obtaining a predicted fault type corresponding to the second sample data based on the extracted third fault features; generating a third loss function based on the predicted fault type and fault label corresponding to the second sample data; generating a fourth loss function based on the third fault features of the second sample data and the second fault features corresponding to the second sample data; and training the initial third fault detection model based on the third loss function and the fourth loss function to obtain a trained third fault detection model; the third fault detection model is used to identify the fault type of the air conditioner corresponding to the subset of data, and the number of parameters of the third fault detection model is less than the number of parameters of the second fault detection model.

[0020] In some embodiments, the initial second fault detection model includes a fully connected layer and an activation function layer; the fully connected layer is used to extract fault features from the sample data, and the activation function layer is used to determine the fault type corresponding to the sample data based on the extracted second fault features.

[0021] In some embodiments, training an initial second fault detection model based on a reconstructed subset of data includes: standardizing the reconstructed subset of data; the standardization process is a process of adjusting the order of magnitude of multiple first fault features and multiple running data in the reconstructed subset of data to a preset order of magnitude; and training the initial second fault detection model based on the reconstructed subset of data after standardization.

[0022] Secondly, embodiments of this application provide a model training method, which includes: acquiring an original dataset, the original dataset including operating data of multiple air conditioners located in multiple regions, each operating data corresponding to a fault label; inputting the multiple operating data into a first fault detection model and acquiring first fault features extracted by the first fault detection model; the first fault detection model is used to extract fault features from the operating data and determine the fault type corresponding to the operating data based on the extracted first fault features; constructing a reconstructed dataset based on the multiple first fault features and the multiple operating data, each first fault feature or operating data in the reconstructed dataset corresponding to a fault label; dividing the reconstructed dataset according to the region where the air conditioner is located to obtain multiple reconstructed sub-datasets, each reconstructed sub-dataset corresponding to a region; training an initial second fault detection model based on the reconstructed sub-datasets to obtain a trained second fault detection model, the second fault detection model being used to identify the fault type of the air conditioner in the region corresponding to the reconstructed sub-dataset, the second fault detection model having fewer parameters than the first fault detection model.

[0023] Thirdly, embodiments of this application provide a server comprising: an acquisition unit for acquiring an original dataset, the original dataset including operating data of multiple air conditioners located in multiple regions, each operating data corresponding to a fault label; the acquisition unit further comprising inputting the multiple operating data into a first fault detection model and acquiring first fault features extracted by the first fault detection model; the first fault detection model being used to extract fault features from the operating data and determine the fault type corresponding to the operating data based on the extracted first fault features; a processing unit for constructing a reconstructed dataset based on the multiple first fault features and the multiple operating data, each first fault feature or operating data in the reconstructed dataset corresponding to a fault label; the processing unit further comprising dividing the reconstructed dataset according to the region where the air conditioner is located to obtain multiple reconstructed sub-datasets, each reconstructed sub-dataset corresponding to a region; and the processing unit further comprising training an initial second fault detection model based on the reconstructed sub-datasets to obtain a trained second fault detection model, the second fault detection model being used to identify the fault type of the air conditioner in the region corresponding to the reconstructed sub-dataset, the second fault detection model having fewer parameters than the first fault detection model.

[0024] Fourthly, embodiments of this application provide a controller, including: one or more processors; one or more memories; wherein the one or more memories are used to store computer program code, the computer program code including computer instructions, and when the one or more processors execute the computer instructions, the controller executes any of the model training methods provided in the second aspect.

[0025] Fifthly, embodiments of this application provide a computer-readable storage medium including computer instructions that, when executed on a computer, cause the computer to perform any of the model training methods provided in the second aspect.

[0026] In a sixth aspect, embodiments of this application provide a computer program product that can be directly loaded into a memory and contains software code. After being loaded and executed by a computer, the computer program product can implement any of the model training methods provided in the second aspect.

[0027] It should be noted that the aforementioned computer instructions may be stored, in whole or in part, on a computer-readable storage medium. This computer-readable storage medium may be packaged together with the controller's processor, or it may be packaged separately from the controller's processor; this application does not impose any limitations on this.

[0028] The beneficial effects described in aspects two through six of this application can be referred to the analysis of the beneficial effects of aspect one, and will not be repeated here. Attached Figure Description

[0029] The accompanying drawings are used to provide a further understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0030] Figure 1 This is a schematic diagram of an application scenario provided by an embodiment of this application;

[0031] Figure 2 This application provides a schematic diagram of information interaction between an air conditioner and a cloud device.

[0032] Figure 3 This application provides a schematic diagram of information interaction between an air conditioner and a peripheral device.

[0033] Figure 4 A flowchart illustrating a model training method provided in this application embodiment;

[0034] Figure 5 A schematic diagram of the architecture of a first fault detection model provided in an embodiment of this application;

[0035] Figure 6 A schematic diagram of the architecture of another first fault detection model provided in an embodiment of this application;

[0036] Figure 7 A flowchart illustrating another model training method provided in this application embodiment;

[0037] Figure 8 A schematic diagram illustrating the construction of a first total loss function provided in an embodiment of this application;

[0038] Figure 9 A flowchart illustrating another model training method provided in this application embodiment;

[0039] Figure 10 A schematic diagram illustrating the construction of a second total loss function provided in an embodiment of this application;

[0040] Figure 11 A flowchart illustrating another model training method provided in this application embodiment;

[0041] Figure 12 This is a schematic diagram of the structure of a server provided in an embodiment of this application. Detailed Implementation

[0042] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0043] It should be noted that all directional indicators (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicator will also change accordingly.

[0044] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0045] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "connected" and "linked" should be interpreted broadly, for example, as a fixed connection, a detachable connection, or an integral connection. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances. Furthermore, when describing pipelines, the terms "connected" and "linked" as used in this application have the meaning of establishing electrical connection. The specific meaning needs to be understood in conjunction with the context.

[0046] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0047] To maintain high accuracy while reducing the complexity of air conditioner fault detection models, this application provides a model training method. This method combines the first fault features extracted by the first fault detection model before the activation function layer with operational data from the original dataset to construct an enhanced reconstructed dataset. This reconstructed dataset is then used as training samples to train an initial second fault detection model with fewer parameters. Since the first fault features are extracted before the activation function layer, they more comprehensively characterize the inherent properties of the air conditioner's operational data, enabling the second fault detection model to learn more detailed air conditioner operational status information during training, thereby enhancing the accuracy of the second fault detection model.

[0048] In this embodiment, an air conditioner is a device that regulates and controls parameters such as temperature, humidity, and airflow rate of the ambient air inside a building or structure. For example, an air conditioner can be a cabinet air conditioner, a wall-mounted air conditioner, a multi-split air conditioner, etc., and there is no limitation thereto.

[0049] In this application, the air conditioner performs a cooling / heating cycle by using a compressor, condenser, expansion valve, and evaporator. The cooling cycle includes a series of processes involving compression, condensation, expansion, and evaporation, and supplies refrigerant to the conditioned and heat-exchanged air.

[0050] The compressor compresses refrigerant gas under high temperature and pressure and discharges the compressed refrigerant gas. The discharged refrigerant gas flows into the condenser. The condenser condenses the compressed refrigerant into a liquid phase, and the heat is released to the surrounding environment through the condensation process.

[0051] The expansion valve expands the high-temperature, high-pressure liquid refrigerant condensed in the condenser into a low-pressure liquid refrigerant. The evaporator evaporates the expanded refrigerant in the expansion valve and returns the low-temperature, low-pressure refrigerant gas to the compressor. The evaporator achieves its cooling effect by utilizing the latent heat of refrigerant evaporation to exchange heat with the material being cooled. Throughout the cycle, the air conditioner regulates the temperature of the indoor space.

[0052] The outdoor unit of an air conditioner refers to the part of the refrigeration cycle that includes the compressor and the outdoor heat exchanger. The indoor unit of an air conditioner includes the indoor heat exchanger, and an expansion valve can be provided in either the indoor or outdoor unit.

[0053] The indoor and outdoor heat exchangers function as either condensers or evaporators. When the indoor heat exchanger is used as a condenser, the air conditioner functions as a heater in heating mode; when the indoor heat exchanger is used as an evaporator, the air conditioner functions as a cooler in cooling mode.

[0054] An air conditioner according to some embodiments of this application includes an indoor unit installed in an indoor space. The indoor unit, i.e., the aforementioned indoor unit, is connected via pipes to an outdoor unit, i.e., the aforementioned outdoor unit, installed in an outdoor space. The outdoor unit may include a compressor, an outdoor heat exchanger, an outdoor fan, an expander, and similar components for a refrigeration cycle; the indoor unit may also include an indoor heat exchanger and an indoor fan.

[0055] Figure 1 This is a schematic diagram illustrating an application scenario provided by this application according to an exemplary embodiment. For example... Figure 1 As shown, this application scenario can include air conditioner 101. There can be multiple air conditioners 101, which can be distributed in different areas and belong to different types. For example, the multiple air conditioners 101 may include different types of air conditioners such as a three-in-one multi-split air conditioner or a five-in-one multi-split air conditioner.

[0056] In this embodiment, the application scenario may include cloud device 102. Cloud device 102 utilizes advanced large-scale distributed computing architecture and virtualization technology to build a powerful cloud computing platform. This platform integrates high-performance computing nodes, ample storage space, and advanced data analysis tools, and provides stable and reliable internet infrastructure services.

[0057] In this embodiment, the cloud device 102 has a one-to-one correspondence with the area where the air conditioner is located. The cloud device 102 is used to process the operating data of multiple air conditioners 101 located in the area corresponding to the cloud device 102, and applies advanced algorithms such as deep learning and machine learning to perform complex data analysis and model training. For example, cloud device A corresponds to area 1 and is used to process the operating data of the air conditioner located in area 1; cloud device B corresponds to area 2 and is used to process the operating data of the air conditioner located in area 2.

[0058] By analyzing the operating data of the air conditioner, the cloud device 102 can continuously optimize and update the fault detection model deployed on the cloud device 102, improving the accuracy and generalization ability of the fault detection model deployed on the cloud device 102. At the same time, the cloud device 102 also provides remote support and data synchronization services to the edge device 103, ensuring the coordinated operation and efficient response of the system.

[0059] In this embodiment, the application scenario may include an edge device 103. The edge device 103 is deployed at a physical location close to the air conditioner 101. The edge device 103 has a one-to-one correspondence with the type of air conditioner 101; that is, the edge device 103 is used for real-time fault detection of air conditioners of the same type located in the same area. For example, edge device C corresponds one-to-one with type 1 air conditioners located in the same area, and edge device C is used for real-time fault detection of type 1 air conditioners located in the same area. Edge device D corresponds one-to-one with type 2 air conditioners located in the same area, and edge device D is used for real-time fault detection of type 2 air conditioners located in the same area.

[0060] Edge device 103 possesses powerful local computing and storage capabilities, enabling it to rapidly process the operating data of air conditioner 101. The real-time and localized processing capabilities of edge device 103 make fault detection of air conditioner 101 faster and more accurate, effectively reducing the impact of faults on the operation of air conditioner 101. Simultaneously, edge device 103 can also upload local detection data to cloud device 102, providing a rich data source for data analysis and model optimization in cloud device 102.

[0061] In this embodiment, both cloud device 102 and edge device 103 can be servers. The server can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks, and big data servers. This application does not impose any special restrictions on the specific form of the server.

[0062] The model training method provided in this application can be applied to, for example... Figure 1 The cloud device 102 or edge device 103 shown.

[0063] In one example, such as Figure 2 As shown, the processor of the air conditioner 101 can send the operating information of the air conditioner 101 to the cloud device 102, so as to input the operating information of the air conditioner 101 into the second fault detection model deployed on the cloud device 102, obtain the fault detection result output by the second fault detection model deployed on the cloud device 102, and then the cloud device 102 sends the fault detection result to the air conditioner 101 so that the air conditioner 101 can obtain the fault detection result.

[0064] In another example, such as Figure 3As shown, the processor of the air conditioner 101 can send the operating information of the air conditioner 101 to the edge device 103, so as to input the operating information of the air conditioner 101 into the second fault detection model deployed on the edge device 103, obtain the fault detection result output by the second fault detection model deployed on the edge device 103, and then the edge device 103 sends the fault detection result to the air conditioner 101 so that the air conditioner 101 can obtain the fault detection result.

[0065] The second fault detection model can be trained using the following model training method:

[0066] The following is for reference. Figure 4 This application describes a model training method according to embodiments thereof. Figure 4 This is a flowchart of a model training method according to an embodiment of this application. It should be noted that the step numbers in this application, such as S101, S102, S103, S104, and S105, are only for ease of description and should not be construed as limiting the order of the steps. That is, the execution order of steps S101, S102, S103, S104, and S105 can be specifically determined according to actual needs and is not limited to the order of steps in the following embodiments.

[0067] In some embodiments of this application, such as Figure 4 As shown, the model training method includes at least the following steps.

[0068] S101. Obtain the original dataset.

[0069] The original dataset includes operating data from multiple air conditioners in multiple regions, with each data point corresponding to a fault label.

[0070] It should be noted that, since air conditioners are distributed across different regions, and there are significant differences in climate conditions and operating environments between these regions, to improve the accuracy of air conditioner fault detection in different regions, operational data from multiple air conditioners covering multiple regions and containing diverse operating conditions can be used as the original dataset to train the fault detection model. This approach can improve the accuracy of the fault detection model in identifying faults in air conditioners in different regions, thereby enhancing the regional adaptability of the fault detection model.

[0071] S102. Input multiple operating data into the first fault detection model and obtain the first fault feature extracted by the first fault detection model.

[0072] The first fault detection model is used to extract fault features from the operating data and determine the fault type corresponding to the operating data based on the extracted first fault features.

[0073] In some embodiments, the first fault detection model is used to detect the fault type of an air conditioner located in any of multiple zones. For example, the fault type may include refrigerant leakage, four-way valve failure, compressor failure, sensor failure, indoor unit blockage, etc.

[0074] As one possible implementation, the features output by the intermediate layer of the first fault detection model can be used as the first fault features.

[0075] In one example, the features output by the last fully connected layer of the first fault detection model can be used as the first fault feature. For example, such as... Figure 5 As shown, the first fault detection model 1 includes an input layer, a fully connected layer 1, and an activation function layer connected in sequence. The fully connected layer 1 is the last fully connected layer of the first fault detection model 1; that is, the features output by the fully connected layer 1 serve as the first fault feature. For example, as shown... Figure 6 As shown, the first fault detection model 2 includes an input layer, a fully connected layer 2, a fully connected layer 3, and an activation function layer connected in sequence. Among them, the fully connected layer 3 is the last fully connected layer of the first fault detection model 2, that is, the feature output by the fully connected layer 3 serves as the first fault feature.

[0076] S103. Construct a reconstructed dataset based on multiple first fault characteristics and multiple operational data.

[0077] In this reconstructed dataset, each first fault feature or operational data point corresponds to a fault label. It should be understood that the fault label corresponding to each operational data point in the reconstructed dataset is the same as the label for that operational data point in the original dataset.

[0078] For example, the multiple first fault features extracted by the first fault detection model are {D} i1 D i2 D i3 D in}, where i represents the i-th row. Multiple running data sets in the original dataset are {F1, F2, F3, ..., F...}. m The multiple first fault features and multiple operational data are merged to obtain the reconstructed dataset {F1, F2, F3, ..., F}. m D i1 D i2 D i3 D in}

[0079] S104. Divide the reconstructed dataset according to the area where the air conditioner is located to obtain multiple reconstructed sub-datasets.

[0080] Each reconstructed subset of the dataset corresponds to a region.

[0081] As one possible implementation, each running data in the reconstructed dataset includes a field for indicating the geographical location information of the air conditioner (such as latitude and longitude, city, etc.). Each running data in the reconstructed dataset can be traversed, and the running data can be assigned to the corresponding reconstructed subset based on the field for indicating the geographical location information of the air conditioner. In this way, multiple reconstructed subsets can be obtained.

[0082] S105. Train the initial second fault detection model based on the reconstructed subset dataset to obtain the trained second fault detection model.

[0083] The second fault detection model is used to identify the fault type of the air conditioner in the region corresponding to the reconstructed subset dataset. The number of parameters in the second fault detection model is less than that in the first fault detection model.

[0084] As one possible implementation, based on the reconstructed dataset, the knowledge distillation algorithm can be used to train the initial second fault detection model to obtain a trained second fault detection model.

[0085] For example, when training an initial second fault detection model using a knowledge distillation algorithm, the activation function of the output layer of the initial second fault detection model can be incorporated into the compression coefficient. For instance, the activation function of the output layer is shown in Equation (1) below.

[0086]

[0087] Where, q i z is the soft-label probability of the i-th fault type of the air conditioner; i It is the feature information of the i-th fault type output by the intermediate layer of the second fault detection model; T is the compression coefficient, the larger the value of T, the better the q i The smoother the distribution, the better. The value of T ranges from 0 to 1. When T = 0, the activation function of the output layer is the original activation function of the initial second fault detection model. When T = t and t > 0, the output of the softmax activation function of the output layer tends to be smoother, and the greater the entropy of the distribution, the more the information carried by the negative label will be amplified.

[0088] Furthermore, regarding the use of the knowledge distillation algorithm to train the initial second fault detection model, please refer to the specific descriptions in S201-S205 below, which will not be repeated here.

[0089] In some embodiments, when training the initial second fault detection model based on the reconstructed subset dataset, the reconstructed subset dataset may be standardized. Standardization involves adjusting the order of magnitude of multiple first fault features and multiple running data in the reconstructed subset dataset to a preset order of magnitude. Further, the initial second fault detection model is trained based on the standardized reconstructed subset dataset.

[0090] In this embodiment of the application, after standardizing the reconstructed subset dataset, it can be further divided into a training set and a test set, wherein the number of running data in the training set is 80% of the number of running data in the reconstructed subset dataset, and the number of running data in the test set is 20% of the number of running data in the reconstructed subset dataset.

[0091] In this embodiment, the parameters of the initial second fault detection model can also be optimized using a preset model parameter optimization algorithm. For example, the preset model parameter optimization algorithm is a grid search method.

[0092] In some embodiments, the trained second fault detection model can be deployed on cloud devices or edge devices.

[0093] In this embodiment of the application, when the trained second fault detection model is deployed on a cloud device, the computing power resources of the cloud device are sufficient to meet the computing power resources required by the initial second fault detection model.

[0094] For example, the initial second fault detection model can be a neural network model. In this case, a three-layer network architecture can be adopted: an input layer, a hidden layer, and an output layer. This architecture contains only fully connected layers and activation function layers, aiming to effectively capture the complex relationship between input and output while maintaining the lightweight and computational efficiency of the second fault detection model. The input layer has m+n nodes, the output layer has n nodes, and the hidden layer has (m+2n) / 2 nodes. Furthermore, the second fault detection model retains only the fully connected layer and the activation function layer as operators.

[0095] In this embodiment of the application, when the trained second fault detection model is deployed on an edge device, the computing power resources of the edge device are higher than those required by the initial second fault detection model.

[0096] For example, the initial second fault detection model is an n-class support vector machine model with m+n input features.

[0097] based on Figure 4The embodiment shown in this application provides a model training method that combines the first fault features extracted by the first fault detection model before the activation function layer with the operational data in the original dataset to construct an enhanced reconstructed dataset. This reconstructed dataset is then used as training samples to train an initial second fault detection model with fewer parameters. Since the first fault features are extracted before the activation function layer, they can more comprehensively characterize the inherent characteristics of the air conditioner's operational data, enabling the second fault detection model to learn more detailed air conditioner operational status information during training. This enhances the accuracy of the second fault detection model, thus making fault detection of the air conditioner more accurate when using a second fault detection model that matches the region where the air conditioner is located.

[0098] In addition, this application can divide the reconstructed dataset based on the region where the air conditioner is located, and use the divided reconstructed subset dataset as training samples. This not only reduces the complexity of a single training dataset (i.e., the reconstructed subset dataset), but also enables the second fault detection model to perform more accurate fault detection based on the characteristics of air conditioners located in specific regions.

[0099] Meanwhile, since the second fault detection model has fewer parameters than the first fault detection model, it can reduce the complexity of the fault detection model while maintaining its high accuracy.

[0100] In some embodiments, when the trained second fault detection model is deployed on a cloud device, such as Figure 7 As shown, the initial second fault detection model can be trained based on the reconstructed subset dataset using the following steps.

[0101] S201. Take the first fault feature and running data in the reconstructed subset as the first sample data, and determine the first fault feature corresponding to each first sample data.

[0102] As one possible implementation, during the execution of S102 above, after each piece of running data is input into the first fault detection model and a first fault feature is obtained, the one-to-one correspondence between the running data and the first fault feature can be marked. For example, this marking can be a simple index, a key-value pair, or a more complex data structure.

[0103] Furthermore, when determining the first sample data, the running data in the reconstructed subset can be traversed, and the first fault feature corresponding to each running data can be determined based on the one-to-one correspondence between the running data and the first fault feature. Further, the first fault feature and the running data in the reconstructed subset are combined to form the first sample data.

[0104] S202. Input the first sample data into the initial second fault detection model, extract fault features from the first sample data through the initial second fault detection model, and obtain the predicted fault type corresponding to the first sample data based on the extracted second fault features.

[0105] Furthermore, regarding the initial second fault detection model extracting fault features from the first sample data, please refer to the specific description of the first fault detection model extracting fault features from the operating data in S102 above, which will not be repeated here.

[0106] S203. Generate a first loss function based on the predicted fault type and fault label corresponding to the first sample data.

[0107] The predicted fault type corresponding to the first sample data is the predicted fault type output by the second fault detection model when the compression coefficient T is 1.

[0108] As one possible implementation, the first loss function can be generated based on the preset type of the first loss function, the predicted fault type corresponding to the first sample data, and the fault label.

[0109] For example, the first loss function can be of the type of multi-class cross-entropy loss. The first loss function can be shown in the following formula (2):

[0110]

[0111] Where N is the number of the first sample data, C is the total number of fault types, and y ij To determine whether the fault label of the i-th first sample data belongs to the j-th fault type. Let be the probability that the predicted fault type corresponding to the i-th first sample data belongs to the j-th fault type.

[0112] S204. Generate a second loss function based on the second fault features of the first sample data and the first fault features corresponding to the first sample data.

[0113] Among them, the second fault feature of the first sample data is the activation function q of the second fault detection model when the compression coefficient T is t. i The output value.

[0114] Similarly, the first fault feature corresponding to the first sample data is the value output by the activation function of the first fault detection model when the compression coefficient T is t.

[0115] As one possible implementation, the second loss function can be generated based on the preset type of the second loss function, the second fault features of the first sample data, and the first fault features corresponding to the first sample data.

[0116] For example, the second loss function can be of the type KL (kullback-leibler divergence) loss. The second loss function can be shown in the following formula (3):

[0117]

[0118] Where L2 is the second loss function, z t z represents the first fault feature corresponding to the first sample data. s Let σ be the second fault feature of the first sample data, and let σ be the softmax function.

[0119] S205. Based on the first loss function and the second loss function, train the initial second fault detection model to obtain the trained second fault detection model.

[0120] One possible implementation is to perform a weighted summation of the first loss function and the second loss function to generate a first total loss function. Based on this first total loss function, the parameters of the initial second fault detection model can be continuously adjusted through iterative training to minimize the loss function, thereby obtaining a well-trained second fault detection model.

[0121] For example, the first total loss function can be expressed as the following formula (4).

[0122] L Total1 =αL1+βL2 (4)

[0123] Among them, L Total1 L1 is the first total loss function; α and β are weighting coefficients; L1 is the first loss function; L2 is the second loss function.

[0124] In some embodiments, when the trained second fault detection model is deployed on an edge device, the initial second fault detection model can also be trained based on the reconstructed subset dataset according to the above S201-S205.

[0125] Based on S201-S205, the second fault detection model with fewer parameters can also learn from the first fault detection model with more parameters. By using the second loss function to train the initial second fault detection model, the initial second fault detection model can fit the soft label probability information of the first fault detection model, thereby enabling the initial second fault detection model to learn some potential empirical knowledge from the first fault detection model.

[0126] The following is combined Figure 8 This example illustrates the generation of the first total loss function.

[0127] like Figure 8 As shown, the reconstructed subset dataset is input into the first fault detection model and the second fault detection model respectively, and the predicted fault type and the second fault feature of the first sample data output by the second fault detection model and the first fault feature of the first sample data output by the first fault detection model are obtained respectively.

[0128] Next, a first loss function is generated based on the predicted fault type and fault label corresponding to the first sample data. A second loss function is generated based on the second fault features of the first sample data and the first fault features corresponding to the first sample data. Finally, a first total loss function is generated based on the first loss function and the second loss function.

[0129] It should be noted that, Figure 8 The first fault feature can also be the first fault feature obtained by inputting the original dataset into the first fault detection model in step S102 above. This application does not make specific limitations on this.

[0130] In some embodiments, when training a fault detection model deployed on an edge device, the edge device can also learn from a fault detection model deployed on a cloud device to further compress the edge device's fault detection model. Therefore, as... Figure 9 As shown in the embodiments of this application, a model training method may further include the following steps.

[0131] S301. Divide the reconstructed subset according to the type of air conditioner to obtain multiple subsets.

[0132] Each subset of data corresponds to a type of air conditioner.

[0133] As one possible implementation, each running data in the reconstructed subset includes a field indicating the type of air conditioner. Each running data in the reconstructed subset can be traversed, and the running data can be assigned to the corresponding subset based on the field indicating the type of air conditioner in the running data. In this way, multiple subsets can be obtained, and the initial third fault detection model can be trained based on the subsets to obtain a trained third fault detection model.

[0134] S302. Take the first fault feature and the operating data in the subset as the second sample data, and determine the second fault feature corresponding to each second sample data.

[0135] As one possible implementation, during the process of inputting runtime data from the reconstructed dataset into the second fault detection model, a second fault feature can be obtained for each instance of runtime data input into the second fault detection model. After obtaining a second fault feature, the one-to-one correspondence between the runtime data and the second fault feature can be marked. For example, this marking can be a simple index, key-value pairs, or a more complex data structure.

[0136] Similarly, referring to the description in S201 above, we can obtain the one-to-one correspondence between the operating data and the first fault characteristics.

[0137] Thus, when determining the second sample data, the running data in the subset can be traversed, and based on the one-to-one correspondence between the running data and the first fault feature, the first fault feature and the running data can be combined as the second sample data. Based on the one-to-one correspondence between the running data and the second fault feature, the second fault feature corresponding to the running data can be used as the second fault feature corresponding to the second sample data.

[0138] S303. Input the second sample data into the initial third fault detection model, extract fault features from the second sample data through the initial third fault detection model, and obtain the predicted fault type corresponding to the second sample data based on the extracted third fault features.

[0139] S304. Generate a third loss function based on the predicted fault type and fault label corresponding to the second sample data.

[0140] Understandably, the first loss function is used to characterize the difference between the predicted fault type and the fault label corresponding to the second sample data.

[0141] S305. Generate a fourth loss function based on the third fault feature of the second sample data and the second fault feature corresponding to the second sample data.

[0142] Understandably, the fourth loss function is used to characterize the difference between the third fault feature of the second sample data and the second fault feature corresponding to the second sample data. By using the fourth loss function to train the initial third fault detection model, the initial third fault detection model can fit the soft label probability information of the second fault detection model, thereby enabling the initial third fault detection model to learn some potential empirical knowledge from the second fault detection model.

[0143] S306. Based on the third and fourth loss functions, train the initial second fault detection model to obtain the trained second fault detection model.

[0144] The third fault detection model is used to identify the fault type of the air conditioner corresponding to the subset of data. The number of parameters in the third fault detection model is less than that in the second fault detection model.

[0145] One possible implementation is to perform a weighted summation of the first loss function and the second loss function to generate a first total loss function. Based on this first total loss function, the parameters of the initial second fault detection model can be continuously adjusted through iterative training to minimize the loss function, thereby obtaining a well-trained second fault detection model.

[0146] For example, the second total loss function can be expressed as the following formula (5).

[0147] L Total2 =γL3+δL4 (5)

[0148] Among them, L total2 L3 is the second total loss function; γ and δ are weighting coefficients; L4 is the third loss function; and L5 is the fourth loss function.

[0149] Furthermore, for S303, please refer to the specific description of S102 above; for S304, please refer to the specific description of S203 above; for S305, please refer to the specific description of S204 above; and for S306, please refer to the specific description of S205 above. This application will not elaborate further on these points.

[0150] Based on S301-S306, the fault detection model deployed on the edge device can also learn from the fault detection model deployed on the cloud device. By using the fourth loss function to train the initial third fault detection model, the initial third fault detection model can fit the soft label probability information of the second fault detection model, thereby enabling the initial third fault detection model to learn some potential empirical knowledge from the second fault detection model.

[0151] The following is combined Figure 10 This example illustrates the generation of the second total loss function.

[0152] like Figure 10 As shown, the subsets are input into the second fault detection model and the third fault detection model respectively, and the predicted fault type and the third fault feature of the second sample data output by the third fault detection model and the second fault feature of the second sample data output by the second fault detection model are obtained respectively.

[0153] Next, a third loss function is generated based on the predicted fault type and fault label corresponding to the second sample data. A fourth loss function is generated based on the third fault features of the second sample data and the second fault features corresponding to the second sample data. Finally, a second total loss function is generated based on the third and fourth loss functions.

[0154] In some embodiments, when training a second fault detection model deployed on an edge device, the second fault detection model deployed on the edge device can also learn from a fault detection model deployed on a cloud device. Therefore, a model training method provided in this application embodiment may further include the following steps. It should be understood that in this application embodiment, the first fault detection model is a fault detection model deployed on a cloud device.

[0155] It should be noted that, Figure 10 The second fault feature can also be the second fault feature obtained by inputting the reconstructed subset dataset into the second fault detection model in step S202 above. This application does not make specific limitations on this.

[0156] The following is combined Figure 11 This example illustrates the complete process of training a fault detection model deployed on a cloud device and a fault detection model deployed on an edge device, according to embodiments of this application.

[0157] like Figure 11 As shown, this complex n-class fault detection model includes an input layer, a fully connected layer, and an activation function layer, among which... Figure 11 The fully connected layer shown is the last fully connected layer in the complex n-class fault detection model.

[0158] First, the original dataset {F1, F2, F3, ..., Fm} is input into a complex n-class classification fault detection model, and multiple first fault features {D1, D2, D3, ..., Dn} are obtained from the output of the fully connected layer. Then, the original dataset and the multiple first fault features are used to reconstruct the dataset, resulting in the reconstructed dataset {F1, F2, F3, ..., Fm, D1, D2, D3, ..., Dn}.

[0159] Next, the reconstructed dataset is sent to the cloud and the edge respectively, so as to train the fault detection model deployed on the cloud device and the fault detection model deployed on the edge device based on the reconstructed dataset.

[0160] In the cloud, there can be multiple cloud devices, each responsible for fault detection of various types of air conditioners within a specific area, such as... Figure 11As shown, fault detection model 1 is used to detect faults in various types of air conditioners located in area 1. Fault detection model 2 is used to detect faults in various types of air conditioners located in area 2. Fault detection model 3 is used to detect faults in various types of air conditioners located in area 3.

[0161] At the edge, there can be multiple edge devices, each responsible for fault detection of a specific type of air conditioner. Furthermore, there is a one-to-one or one-to-many correspondence between cloud devices and edge devices. For example... Figure 11 As shown, fault detection models 11, 12, 13, and 14 belong to fault detection model 1 in the cloud. Fault detection model 11 is used to detect faults in air conditioners of type 11 located in region 1. Fault detection model 12 is used to detect faults in air conditioners of type 12 located in region 1. Fault detection model 13 is used to detect faults in air conditioners of type 13 located in region 1. Fault detection model 14 is used to detect faults in air conditioners of type 14 located in region 1.

[0162] Similarly, fault detection models 21, 22, 23, 24, 31, 32, 33, and 34 can be referred to the above description, and will not be repeated here.

[0163] like Figure 12 As shown, this application embodiment provides a server for performing the above-described... Figure 2 The model training method shown includes a server 2000 comprising an acquisition unit 2001 and a processing unit 2002.

[0164] In this embodiment of the application, the acquisition unit 2001 is used to acquire the original dataset, which includes the operating data of multiple air conditioners in multiple regions, and each operating data corresponds to a fault label.

[0165] In this embodiment of the application, the acquisition unit 2001 is further configured to input multiple operating data into the first fault detection model and acquire the first fault feature extracted by the first fault detection model; the first fault detection model is configured to extract fault features from the operating data and determine the fault type corresponding to the operating data based on the extracted first fault feature.

[0166] In this embodiment of the application, the processing unit 2002 is used to construct a reconstruction dataset based on multiple first fault features and multiple operating data, wherein each first fault feature or operating data in the reconstruction dataset corresponds to a fault label.

[0167] In this embodiment of the application, the processing unit 2002 is further configured to divide the reconstructed dataset according to the region where the air conditioner is located, to obtain multiple reconstructed sub-datasets, each of which corresponds to a region;

[0168] In this embodiment of the application, the processing unit 2002 is further configured to train the initial second fault detection model based on the reconstructed subset dataset to obtain the trained second fault detection model. The second fault detection model is used to identify the fault type of the air conditioner in the area corresponding to the reconstructed subset dataset. The number of parameters of the second fault detection model is less than the number of parameters of the first fault detection model.

[0169] In some embodiments, the processing unit 2002 is specifically configured to: take the first fault features and running data in the reconstructed subset as first sample data, and determine the first fault features corresponding to each first sample data; input the first sample data into an initial second fault detection model, extract fault features from the first sample data using the initial second fault detection model, and obtain the predicted fault type corresponding to the first sample data based on the extracted second fault features; generate a first loss function based on the predicted fault type and fault label corresponding to the first sample data; generate a second loss function based on the second fault features of the first sample data and the first fault features corresponding to the first sample data; and train the initial second fault detection model based on the first loss function and the second loss function to obtain a trained second fault detection model.

[0170] In some embodiments, the processing unit 2002 is further configured to: divide the reconstructed subset according to the type of air conditioner to obtain multiple subsets, each subset corresponding to a type of air conditioner; train the initial third fault detection model based on the subsets to obtain a trained third fault detection model, the third fault detection model being used to identify the fault type of the air conditioner corresponding to the subset, the third fault detection model having fewer parameters than the second fault detection model.

[0171] Figure 12 The units in the process can also be called modules; for example, a processing unit can be called a processing module.

[0172] Figure 12If the various units in the process are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this application. Storage media for storing computer software products include: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0173] This application also provides a controller, including a processor and a memory; the memory stores instructions executable by the processor; when the processor is configured to execute the instructions, the controller performs any of the methods provided in the above embodiments.

[0174] This application also provides a computer-readable storage medium including computer-executable instructions that, when run on a computer, cause the computer to perform any of the methods provided in the above embodiments.

[0175] This application also provides a computer program product containing computer execution instructions, which, when run on a computer, causes the computer to perform any of the methods provided in the above embodiments.

[0176] This application also provides a chip, including a processor and an interface. The processor is coupled to a memory through the interface. When the processor executes a computer program in the memory or computer execution instructions, any of the methods provided in the above embodiments are executed.

[0177] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software programs, implementation can be, in whole or in part, in the form of a computer program product. This computer program product includes one or more computer-executable instructions. When these computer-executable instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer-executable instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer-executable instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device containing one or more servers, data centers, etc., that can be integrated with the medium. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state disks, SSDs).

[0178] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, the disclosure, and the appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple instances. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0179] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of this application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from the spirit and scope of this application. Thus, if such modifications and modifications of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and modifications.

[0180] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An air conditioner, characterized in that, include: At least one processor; and at least one computer memory operatively connected to the at least one processor and storing instructions that, when executed by the at least one processor, perform the following operations: The operating information of the air conditioner is input into a second fault detection model that matches the area where the air conditioner is located, and the fault detection result output by the second fault detection model is obtained; wherein, the second fault detection model is trained in the following way: Obtain the raw dataset, which includes the operating data of multiple air conditioners in multiple regions, with each data point corresponding to a fault label; Multiple sets of operational data are input into a first fault detection model, and a first fault feature is obtained from the first fault detection model; the first fault detection model is used to extract fault features from the operational data, and to determine the fault type corresponding to the operational data based on the extracted first fault features; A reconstructed dataset is constructed based on multiple first fault features and multiple operational data, wherein each first fault feature or operational data in the reconstructed dataset corresponds to a fault label; The reconstructed dataset is divided according to the region where the air conditioner is located, resulting in multiple reconstructed sub-datasets, each corresponding to a region; The initial second fault detection model is trained based on the reconstructed subset dataset to obtain a trained second fault detection model. The second fault detection model is used to identify the fault type of the air conditioner in the region corresponding to the reconstructed subset dataset. The number of parameters of the second fault detection model is less than that of the first fault detection model.

2. The air conditioner according to claim 1, characterized in that, The step of training the initial second fault detection model based on the reconstructed subset dataset to obtain the trained second fault detection model includes: The first fault feature and the running data in the reconstructed subset are used as the first sample data, and the first fault feature corresponding to each first sample data is determined. The first sample data is input into the initial second fault detection model, and the fault features of the first sample data are extracted through the initial second fault detection model. Based on the extracted second fault features, the predicted fault type corresponding to the first sample data is obtained. A first loss function is generated based on the predicted fault type and fault label corresponding to the first sample data; A second loss function is generated based on the second fault features of the first sample data and the first fault features corresponding to the first sample data. The initial second fault detection model is trained based on the first loss function and the second loss function to obtain the trained second fault detection model.

3. The air conditioner according to claim 1, characterized in that, The instruction also performs the following operations: The operating information of the air conditioner is input into a third fault detection model that matches the type of the air conditioner, and the fault detection result output by the third fault detection model is obtained; wherein, the third fault detection model is trained in the following way: The reconstructed sub-dataset is divided according to the type of air conditioner to obtain multiple sub-datasets, each of which corresponds to a type of air conditioner; The initial third fault detection model is trained based on the subset dataset to obtain a trained third fault detection model. The third fault detection model is used to identify the fault type of the air conditioner corresponding to the subset dataset. The number of parameters of the third fault detection model is less than that of the second fault detection model.

4. The air conditioner according to claim 3, characterized in that, The step of training the initial third fault detection model based on the subset dataset to obtain the trained third fault detection model includes: The first fault feature and the running data in the subset are used as the second sample data, and the second fault feature corresponding to each second sample data is determined; the second fault feature is extracted by the second fault detection model; the second sample data is input into the initial third fault detection model, and the fault feature is extracted from the second sample data through the initial third fault detection model, and the predicted fault type corresponding to the second sample data is obtained according to the extracted third fault feature; A third loss function is generated based on the predicted fault type and fault label corresponding to the second sample data; A fourth loss function is generated based on the third fault feature of the second sample data and the second fault feature corresponding to the second sample data. The initial third fault detection model is trained based on the third loss function and the fourth loss function to obtain the trained third fault detection model.

5. The air conditioner according to any one of claims 2-4, characterized in that, The initial second fault detection model includes a fully connected layer and an activation function layer; the fully connected layer is used to extract fault features from the sample data, and the activation function layer is used to determine the fault type corresponding to the sample data based on the extracted second fault features.

6. The air conditioner according to any one of claims 1-4, characterized in that, The step of training the initial second fault detection model based on the reconstructed subset dataset includes: The reconstructed subset of data is standardized; the standardization process is used to adjust the order of magnitude of multiple first fault features and multiple operational data in the reconstructed subset of data to a preset order of magnitude. The initial second fault detection model is trained based on the reconstructed subset of data after standardization.

7. A model training method, characterized in that, include: Obtain the raw dataset, which includes the operating data of multiple air conditioners in multiple regions, with each data point corresponding to a fault label; Multiple sets of operational data are input into the first fault detection model, and the first fault feature extracted by the first fault detection model is obtained. The first fault detection model is used to extract fault features from the operating data and determine the fault type corresponding to the operating data based on the extracted first fault features. A reconstructed dataset is constructed based on multiple first fault features and multiple operational data, wherein each first fault feature or operational data in the reconstructed dataset corresponds to a fault label; The reconstructed dataset is divided according to the region where the air conditioner is located, resulting in multiple reconstructed sub-datasets, each corresponding to a region; The initial second fault detection model is trained based on the reconstructed subset dataset to obtain a trained second fault detection model. The second fault detection model is used to identify the fault type of the air conditioner in the region corresponding to the reconstructed subset dataset. The number of parameters of the second fault detection model is less than that of the first fault detection model.

8. A server, characterized in that, include: An acquisition unit is used to acquire a raw dataset, which includes operating data of multiple air conditioners in multiple regions, and each data point corresponds to a fault label. The acquisition unit is further configured to input multiple sets of the running data into the first fault detection model and acquire the first fault feature extracted by the first fault detection model; The first fault detection model is used to extract fault features from the operating data and determine the fault type corresponding to the operating data based on the extracted first fault features. The processing unit is configured to construct a reconstruction dataset based on a plurality of first fault features and a plurality of operational data, wherein each first fault feature or operational data in the reconstruction dataset corresponds to a fault label; The processing unit is further configured to divide the reconstructed dataset according to the region where the air conditioner is located, to obtain multiple reconstructed sub-datasets, each of which corresponds to a region; The processing unit is further configured to train the initial second fault detection model based on the reconstructed subset dataset to obtain a trained second fault detection model. The second fault detection model is used to identify the fault type of the air conditioner in the region corresponding to the reconstructed subset dataset. The number of parameters of the second fault detection model is less than the number of parameters of the first fault detection model.

9. The server according to claim 8, characterized in that, The processing unit is specifically used for: The first fault feature and the running data in the reconstructed subset are used as the first sample data, and the first fault feature corresponding to each first sample data is determined. The first sample data is input into the initial second fault detection model, and the fault features of the first sample data are extracted through the initial second fault detection model. Based on the extracted second fault features, the predicted fault type corresponding to the first sample data is obtained. A first loss function is generated based on the predicted fault type and fault label corresponding to the first sample data; A second loss function is generated based on the second fault features of the first sample data and the first fault features corresponding to the first sample data. The initial second fault detection model is trained based on the first loss function and the second loss function to obtain the trained second fault detection model.

10. The server according to claim 8, characterized in that, The processing unit is further configured to: The reconstructed sub-dataset is divided according to the type of air conditioner to obtain multiple sub-datasets, each of which corresponds to a type of air conditioner; The initial third fault detection model is trained based on the subset dataset to obtain a trained third fault detection model. The third fault detection model is used to identify the fault type of the air conditioner corresponding to the subset dataset. The number of parameters of the third fault detection model is less than that of the second fault detection model.

Images