Fault diagnosis model training method and fault diagnosis method

Abstract

The invention is suitable for the technical field of air conditioner fault diagnosis, and provides a fault diagnosis model training method and a fault diagnosis method, and the method comprises the steps: obtaining multi-split training data; wherein the multi-split air conditioner training data comprises labeled data and unlabeled data generated in the operation process of a multi-split air conditioner, a data enhancement model is used for conducting data enhancement on the multi-split air conditioner training data to obtain first prediction label data, the first prediction label data and the labeled data are input into a classification model, and the classification model is used for classifying the first prediction label data and the labeled data. And obtaining a first prediction diagnosis result, and training the data enhancement model and the classification model according to the multi-split training data, the first prediction label data and the first prediction diagnosis result to obtain a fault diagnosis model. The fault diagnosis accuracy of the multi-split air conditioner can be improved.

Description

Technical Field

[0001] The present application belongs to the technical field of air conditioner fault diagnosis, and in particular relates to a fault diagnosis model training method, a fault diagnosis method, a device, an electronic device, and a computer program product. Background Art

[0002] With the development of smart homes, the functionality of multi-split air conditioners is becoming increasingly complex. Various problems may arise during long-term operation, affecting their proper function. Furthermore, the structural complexity and diverse operating conditions of multi-split air conditioners make fault diagnosis reliant on manual testing, which is inefficient and relies heavily on the expertise of maintenance personnel for accurate diagnosis.

[0003] Currently, artificial intelligence technology can be used to build a fault diagnosis model to perform intelligent diagnosis on multi-split air conditioners. However, when building the fault diagnosis model, a large amount of fault data is required as training samples. Since the failure probability of multi-split air conditioners is low, the number of samples is very limited. Therefore, the fault diagnosis model cannot fully learn the characteristics of the sample data, resulting in poor diagnostic accuracy of the trained fault diagnosis model. Summary of the Invention

[0004] The embodiments of the present application provide a fault diagnosis model training method, a fault diagnosis method, an apparatus, an electronic device, and a computer program product, which can improve the accuracy of multi-split air conditioner fault diagnosis.

[0005] In a first aspect, an embodiment of the present application provides a fault diagnosis model training method, comprising:

[0006] Acquire multi-split air conditioner training data; wherein the multi-split air conditioner training data includes labeled data and unlabeled data generated during the operation of the multi-split air conditioner, the labeled data includes a fault label representing a fault state of the multi-split air conditioner, and the unlabeled data does not include a fault label representing a fault state of the multi-split air conditioner;

[0007] Inputting the unlabeled data into a data augmentation model to obtain first predicted label data; the data augmentation model is used to learn the distribution pattern of fault labels in the labeled data and output the first predicted label data containing the same distribution pattern of the fault labels;

[0008] Inputting the first predicted label data and the labeled data into a classification model to obtain a first predicted diagnosis result; the classification model is used to learn a mapping relationship between the first predicted label data, the labeled data, and the fault labels in the first predicted label data and the labeled data, and output the first predicted diagnosis result having the same mapping relationship;

[0009] The data enhancement model and the classification model are trained based on the multi-split training data, the first prediction label data and the first prediction diagnosis result to obtain a fault diagnosis model; wherein the fault diagnosis model is used to output the fault diagnosis result during the real-time operation of the multi-split air conditioner.

[0010] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0011] In an embodiment of the present application, multi-split training data is obtained. Since the multi-split training data includes labeled data and unlabeled data generated during the operation process of the multi-split air conditioner, and the above-mentioned labeled data includes a fault label that characterizes the fault state of the multi-split air conditioner, and the above-mentioned unlabeled data does not include a fault label that characterizes the fault state of the multi-split air conditioner, a data enhancement model is used to learn the distribution pattern of the fault labels in the above-mentioned labeled data, and then the data enhancement model is used to perform data enhancement on the unlabeled data. Data expansion can be performed on the actual operation data to obtain the first predicted label data, thereby expanding the data volume of the training sample. At the same time, the first predicted label data and the labeled data are input into the classification model, so that the above-mentioned classification model can learn the mapping relationship between the above-mentioned first predicted label data, the above-mentioned labeled data, and the fault labels therein, and obtain the first predicted diagnosis result, that is, the above-mentioned classification model can classify the labeled data of the operation process and the data-enhanced first predicted label data based on the above-mentioned mapping relationship, which can further increase the number of classifications of training samples. Therefore, using the above-mentioned multi-connected training data, the first predicted label data and the first predicted diagnosis result to train the data enhancement model and the classification model, the data enhancement model and the classification model can learn the characteristics of richer fault data, thereby improving the accuracy of multi-connected air conditioner fault diagnosis using the fault diagnosis model.

[0012] Optionally, the data augmentation model includes a generator, an encoder, and a discriminator; and inputting the unlabeled data into the data augmentation model to obtain first predicted label data includes:

[0013] Inputting the unlabeled data and random noise into the generator to obtain generated label data; wherein the generated label data includes the fault label predicted by the generator;

[0014] Reconstructing the generated label data using the encoder to obtain a latent vector; wherein the latent vector is data with a distribution similar to that of the generated label data;

[0015] Inputting the latent vector and the labeled data into the discriminator to obtain a judgment result; wherein the judgment result is used to indicate an error in the data input to the discriminator;

[0016] The generator is reversely iterated using the judgment result, and all the generated label data generated by the reverse iteration process of the generator are obtained to obtain the first predicted label data.

[0017] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0018] In an embodiment of the present application, by introducing an encoder into the data augmentation model, the encoder can map the generated label data generated by the generator to the latent space through the hidden layer, better extract the main features therein, and reconstruct it into a latent vector with the same distribution, thereby helping the judge to better judge the error of the input data and improve the accuracy of the generation of the first predicted label data.

[0019] Optionally, the discriminator includes a first discriminator and a second discriminator, and the judgment result includes a first judgment result and a second judgment result; and inputting the latent vector and the labeled data into the discriminator to obtain the judgment result includes:

[0020] Inputting the latent vector and random noise into the first discriminator to obtain a first judgment result; wherein the first judgment result is used to indicate an error between the latent vector and the random noise;

[0021] The unlabeled data and the generated label data are input into the second discriminator to obtain a second judgment result; wherein the second judgment result is used to indicate the error between the unlabeled data and the generated label data.

[0022] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0023] In an embodiment of the present application, through the first discriminator and the second discriminator, the generator's reverse iterative process can generate more realistic data, and the output generated label data can have the characteristics of unlabeled data, thereby further improving the data quality of the first predicted label data.

[0024] Optionally, the classification model includes a plurality of base classifiers, and inputting the first predicted label data and the labeled data into the classification model to obtain a first predicted diagnosis result includes:

[0025] Assigning the same initial weight to the first predicted label data and the labeled data, inputting the first predicted label data and the labeled data into the base classifier to obtain a classification result, and updating the initial weight using the classification result to obtain a target weight;

[0026] Inputting the first predicted label data and the labeled data corresponding to the target weight into the next base classifier to obtain a new classification result, and updating the target weight using the new classification result to obtain a new target weight;

[0027] Returning to the step of inputting the first predicted label data and the labeled data corresponding to the target weight into the next base classifier and subsequent steps until all the base classifiers output classification results;

[0028] Assigning weights to each base classifier according to the classification results of each base classifier to obtain base classifier weights corresponding to each base classifier;

[0029] The classification results output by all the base classifiers are weighted averaged according to the base classifier weights to obtain the first predicted diagnosis result.

[0030] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0031] In an embodiment of the present application, the weights of the first predicted label data and the labeled data are adjusted according to the classification results of the previous base classifier, so that the next base classifier can be trained based on the adjusted weights, thereby continuously improving the accuracy of the next base classifier training.

[0032] Optionally, after training the data enhancement model and the classification model based on the multi-online training data, the first prediction label data, and the first prediction diagnosis result to obtain a fault diagnosis model, the method further includes:

[0033] Using the unlabeled data as input to the trained classification model in the fault diagnosis model to obtain a second prediction diagnosis result output by the trained classification model;

[0034] Constructing second predicted label data based on the second predicted diagnosis result and the unlabeled data; wherein the second predicted label data includes a pseudo label determined based on the second predicted diagnosis result;

[0035] The trained classification model is fine-tuned using one or more of the labeled data, the first predicted label data, and the second predicted label data to obtain a fine-tuned classification model.

[0036] The beneficial effect of the embodiment of the present application compared with the prior art is that in the embodiment of the present application, after model training, the classification model is fine-tuned again using data mixed with different labeled data, which can further improve the accuracy of the classification model output results.

[0037] In a second aspect, an embodiment of the present application provides a fault diagnosis method applied to a multi-split air conditioner, comprising:

[0038] Acquiring real-time operating data of the multi-split air conditioner during operation;

[0039] The real-time operation data is used as input of a fault diagnosis model to obtain a fault diagnosis result output by the fault diagnosis model, wherein the fault diagnosis model is obtained according to the fault diagnosis model training method described in any one of the first aspects.

[0040] In a third aspect, an embodiment of the present application provides a fault diagnosis model training device, comprising:

[0041] A training data acquisition module is used to acquire multi-split training data; wherein the multi-split training data includes labeled data and unlabeled data generated during the operation of the multi-split air conditioner, the labeled data includes a fault label representing the fault state of the multi-split air conditioner, and the unlabeled data does not include a fault label representing the fault state of the multi-split air conditioner;

[0042] A data augmentation module, configured to input the unlabeled data into a data augmentation model to obtain first predicted label data; the data augmentation model is configured to learn the distribution pattern of fault labels in the labeled data and output the first predicted label data containing the same distribution pattern of the fault labels;

[0043] a classification module, configured to input the first predicted label data and the labeled data into a classification model to obtain a first predictive diagnosis result; the classification model is configured to learn a mapping relationship between the first predicted label data, the labeled data, and the fault labels in the first predicted label data and the labeled data, and output the first predictive diagnosis result having the same mapping relationship;

[0044] A training module trains the data enhancement model and the classification model based on the multi-split training data, the first prediction label data and the first prediction diagnosis result to obtain a fault diagnosis model; wherein the fault diagnosis model is used to output the fault diagnosis result during the real-time operation of the multi-split air conditioner.

[0045] In a fourth aspect, an embodiment of the present application provides a fault diagnosis device for a multi-split air conditioner, comprising:

[0046] A real-time data acquisition module, used to acquire real-time operating data during the operation of the multi-split air conditioner;

[0047] A fault diagnosis module is used to use the real-time operation data as input of a fault diagnosis model to obtain a fault diagnosis result output by the fault diagnosis model, wherein the fault diagnosis model is obtained according to the fault diagnosis model training method described in the first aspect.

[0048] In the fifth aspect, an embodiment of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the fault diagnosis model training method described in the first aspect or the fault diagnosis method described in the second aspect.

[0049] In a sixth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps of the fault diagnosis model training method described in the first aspect or the fault diagnosis method described in the second aspect.

[0050] In the seventh aspect, an embodiment of the present application provides a computer program product, which, when running on an electronic device, enables the electronic device to execute the steps of the fault diagnosis model training method described in the first aspect or the fault diagnosis method described in the second aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0052] Figure 1 This is a schematic diagram of an operation scenario between a multi-split air conditioning system and a control device provided in one embodiment of the present application;

[0053] Figure 2 This is a hardware configuration block diagram of a control device provided in one embodiment of the present application;

[0054] Figure 3 This is a flowchart of a fault diagnosis model training method provided by an embodiment of the present application;

[0055] Figure 4 This is a flow chart of the fault diagnosis model training process provided by the embodiment of the present application;

[0056] Figure 5 This is a flow chart of fine-tuning the classification model provided in the embodiment of the present application;

[0057] Figure 6This is a flowchart of a fault diagnosis method provided by an embodiment of the present application;

[0058] Figure 7 Schematic diagram of the structure of the fault diagnosis model training device provided in an embodiment of the present application;

[0059] Figure 8 is a structural diagram of a fault diagnosis device provided in an embodiment of the present application;

[0060] Figure 9 It is a structural diagram of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0061] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0062] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0063] It will also be understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0064] As used in this specification and the appended claims, the term "if" can be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0065] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0066] References to "one embodiment" or "some embodiments" in this specification mean that a particular feature, structure, or characteristic described in conjunction with that embodiment is included in one or more embodiments of the present application. Thus, phrases such as "in one embodiment," "in some embodiments," "in other embodiments," and "in other embodiments" appearing in various places in this specification do not necessarily refer to the same embodiment, but rather mean "one or more but not all embodiments," unless otherwise specifically emphasized. The terms "including," "comprising," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0067] Figure 1 This is a schematic diagram of the operation scenario between the multi-split air conditioning system and the control device provided in some embodiments of the present application. Figure 1 As shown in FIG, a user can operate the multi-split air conditioning system 200 through touch operation, the mobile terminal 300 and the control device 100. For example, the control device 100 can be a remote controller, a stylus pen or a handle.

[0068] In some embodiments, the control device 100 may be a remote control or a smart home controller. For example, if the control device is a remote control, the remote control and the multi-split air conditioning system may communicate using, but is not limited to, infrared protocol communication, Bluetooth protocol communication, or other short-range communication methods, to control the multi-split air conditioning system 200 wirelessly or wired. The user may control the multi-split air conditioning system 200 by inputting user commands through buttons on the remote control, voice input, or control panel input.

[0069] It should be understood that the multi-split air conditioning system 200 includes a one-to-many air conditioning system, a natural fluorine ground water heat pump system, etc.

[0070] In some embodiments, a mobile terminal 300 (such as a tablet, computer, or mobile phone) can also be used to control the multi-split air conditioning system 200. For example, an application running on the mobile terminal 300 can be used to control the multi-split air conditioning system 200.

[0071] In some embodiments, the multi-split air conditioning system may not use the above-mentioned mobile terminal 300 or control device 100 to receive instructions, but may receive user control through buttons set on the multi-split air conditioning system.

[0072] Figure 2 Some embodiments of this application provide Figure 1 The hardware configuration diagram of the control device in the figure. Figure 2 As shown, the control device 100 may include: a controller 110, a communication interface 130, a user input / output interface, a memory, and a power supply.

[0073] The control device 100 is configured to control the multi-split air-conditioning system 200, and can receive the user's input operation instructions, and convert the operation instructions into instructions that the multi-split air-conditioning system 200 can recognize and respond to, playing the role of an interactive intermediary between the user and the multi-split air-conditioning system 200.

[0074] In some embodiments, the control device 100 may be a smart device. For example, the control device 100 may be installed with various applications for controlling the multi-split air conditioning system 200 according to user needs.

[0075] In some embodiments, as Figure 1 As shown, the mobile terminal 300 or other intelligent electronic devices can play a similar function as the control device 100 after installing the application for controlling the multi-split air-conditioning system 200 .

[0076] The controller 110 includes a processor 112, RAM 113, ROM 114, a communication interface 130, and a communication bus. The controller 110 is used to control the operation and operation of the control device 100, as well as the communication and cooperation between internal components and external and internal data processing functions.

[0077] Under the control of the controller 110, the communication interface 130 communicates control signals and data signals with the multi-split air conditioning system 200. The communication interface 130 may include at least one of a WiFi chip 131, a Bluetooth module 132, an NFC module 133, or other near-field communication modules.

[0078] The user input / output interface 140 includes at least one of a microphone 141 , a touch panel 142 , a sensor 143 , a button 144 and other input interfaces.

[0079] In some embodiments, the control device 100 includes at least one of a communication interface 130 and an input / output interface 140. The control device 100 is configured with the communication interface 130, such as a WiFi, Bluetooth, or NFC module, to encode user input commands via the WiFi, Bluetooth, or NFC protocols and transmit them to the multi-split air conditioning system 200.

[0080] The memory 190 is used to store various operating programs, data and applications for driving and controlling the control device 100 under the control of the controller. The memory 190 can store various control signal instructions input by the user.

[0081] The power supply 180 is used to provide operating power support for each component of the control device 100 under the control of the controller.

[0082] Multi-split air conditioners circulate refrigerant through at least one outdoor unit and indoor unit to form a refrigerant circuit. The entire unit has a complex structure and diverse operating environments. During operation, various faults may occur, such as fan failure, outdoor unit failure, and refrigerant shortage. Currently, artificial intelligence technology can be used to train fault diagnosis models to perform intelligent diagnosis on multi-split air conditioners, thereby improving the efficiency and accuracy of multi-split air conditioner diagnosis. However, multi-split air conditioner failures are low-probability events, and during the fault diagnosis model training process, it is often difficult to obtain a large amount of real fault data as training samples. The above training samples mainly include fault data labeled with whether a fault has occurred and the fault category.

[0083] In some related technologies, fault data is mainly obtained through manual labeling or laboratory simulation methods. Among them, the accuracy of manual labeling depends on manual experience, and the labeling efficiency is not high. The laboratory simulation method requires laboratory simulated fault experiments to obtain corresponding fault data. However, due to the high cost of fault experiments and the difficulty in conducting specific fault experiments and collecting fault data in actual multi-unit units, the actual available fault data is also relatively small. Therefore, the training samples used in the fault diagnosis model training process are actually small sample data, which makes it impossible for the fault diagnosis model to fully learn the characteristics of the sample data, resulting in poor diagnostic accuracy of the trained fault diagnosis model.

[0084] In order to improve the accuracy of multi-split air conditioner fault diagnosis, the present application provides a fault diagnosis model training method, in which multi-split training data is obtained; wherein the multi-split training data includes labeled data and unlabeled data generated during the operation process of the multi-split air conditioner, and the unlabeled data is enhanced using a data enhancement model to obtain first predicted label data, and the first predicted label data and the labeled data are input into a classification model to obtain a first predicted diagnosis result. According to the multi-split operation data, the first predicted label data and the first predicted diagnosis result, the data enhancement model and the classification model are trained to obtain a fault diagnosis model.

[0085] Figure 3 A flow chart of a fault diagnosis model training method provided in an embodiment of the present application is shown, and is described in detail as follows:

[0086] S31. Obtain multi-split training data; wherein the multi-split training data includes labeled data and unlabeled data generated during the operation of the multi-split air conditioner, the labeled data includes a fault label representing a fault state of the multi-split air conditioner, and the unlabeled data does not include a fault label representing a fault state of the multi-split air conditioner.

[0087] It should be understood that the multi-split air conditioner training data refers to operating data generated during the operation of the multi-split air conditioner in a real environment or a simulated environment, for example, one or more of indoor unit operating data, outdoor unit operating data, and refrigerant operating data. The labeled data refers to operating data labeled with fault labels, and the unlabeled data refers to operating data not labeled with fault labels. For example, the labeled data can be {x, y}, where x is the operating data of the multi-split air conditioner and y is the corresponding label.

[0088] It should also be understood that the fault status represented by the above-mentioned fault label includes whether the multi-split air conditioner is faulty and the fault category, etc. For example, in labeled data, for normal operation data, the marked fault label can be 0; for faulty operation data, labels such as 1, 2, 3... can be marked according to actual conditions.

[0089] S32. Input the above-mentioned unlabeled data into the data enhancement model to obtain first predicted label data; the above-mentioned data enhancement model is used to learn the distribution law of the fault labels in the above-mentioned labeled data, and output the above-mentioned first predicted label data containing the same distribution law of the above-mentioned fault labels.

[0090] It should be understood that the data augmentation model can be a neural network model trained based on the labeled data. For example, the model parameters of a pre-built neural network model are randomly initialized, and after batch input of the labeled data for training, a model is obtained that learns the distribution pattern of fault labels in the labeled data. The data augmentation model can include one or more of convolutional neural networks (CNNs), generative adversarial networks (GANs), and variational autoencoders (VAEs). Optionally, the above-mentioned convolutional neural network includes a convolution layer, a pooling layer and a fully connected layer, and data enhancement can be performed by adjusting the size and step size of the convolution kernel to obtain the above-mentioned first predicted label data; the above-mentioned generative adversarial network includes a generator and a discriminator, and the generator generates generated data that is as close to the real data (i.e., multi-online training data) as possible, and then the discriminator distinguishes the difference between the generated data and the real data, and continuously optimizes the generator to generate a large amount of new generated data, i.e., the above-mentioned first predicted label data; the above-mentioned variational autoencoder reconstructs the input real data through the encoding-decoding process, and can introduce noise or perform other transformations in the reconstruction process to increase the diversity of the data to obtain the above-mentioned first predicted label data.

[0091] Specifically, the multi-online training data is used as the input of the data enhancement model. The data enhancement model predicts the fault labels corresponding to the unlabeled data based on the learned distribution law of the fault labels to obtain the first predicted label data, wherein the first predicted label data includes the unlabeled data and its corresponding predicted fault labels.

[0092] In the embodiment of the present application, the above-mentioned data enhancement model can be used to perform data expansion based on real operating data to improve the diversity of training data.

[0093] S33. Input the above-mentioned first predicted label data and the above-mentioned labeled data into the classification model to obtain a first predicted diagnosis result; the above-mentioned classification model is used to learn the mapping relationship between the above-mentioned first predicted label data, the above-mentioned labeled data, and the fault labels in the above-mentioned first predicted label data and the above-mentioned labeled data, and output the above-mentioned first predicted diagnosis result with the same above-mentioned mapping relationship.

[0094] It should be understood that the above-mentioned classification model is a model that can learn the mapping relationship between the above-mentioned first prediction label data and the operating data included in the above-mentioned labeled data and the corresponding fault labels, and output the prediction diagnosis results, wherein the above-mentioned classification model may include one or more machine learning models, deep learning models and other probability models, wherein the above-mentioned machine learning models may include but are not limited to: logistic regression model (Logistic Regression), support vector machine model (SVM), decision tree model (Decision Tree), random forest model (Random Forest), and gradient boosting tree model (XGBoost / LightGBM / CatBoost), etc.; the above-mentioned deep learning models may include but are not limited to: multilayer perceptron (Multilayer Perceptron, MLP), convolutional neural network (CNN) and recurrent neural network (RNN / LSTM / GRU), etc.; the above-mentioned other probability models may include: Bayesian network (Bayesian Network), hidden Markov model (HMM), etc.

[0095] The first predictive diagnosis result may include: fault probability, whether there is a fault, fault type, etc.

[0096] Specifically, the first predicted label data and the above-mentioned labeled data are used as inputs of the classification model. The classification model will learn the mapping relationship between the operating data and the corresponding fault labels included therein, and predict the failure probability of the operating data. Then, the above-mentioned first predicted label data and the above-mentioned labeled data are labeled according to the failure probability to obtain the prediction diagnosis results, including labels such as whether there is a fault and the fault category.

[0097] In an embodiment of the present application, by classifying the original labeled data and the labeled data after data enhancement (i.e., the first predicted label data) through a classification model, richer classification categories can be obtained, thereby further improving the quality of training samples.

[0098] S34. Based on the multi-split training data, the first prediction label data and the first prediction diagnosis result, the data enhancement model and the classification model are trained to obtain a fault diagnosis model; wherein the fault diagnosis model is used to output the fault diagnosis results during the real-time operation of the multi-split air conditioner.

[0099] Specifically, the data augmentation model and the classification model are jointly trained using the multi-online training data, the first predicted label data, and the first predicted diagnosis result. If the training conditions are not met, the model parameters of the data augmentation model and the classification model are updated until the training conditions are met, thereby obtaining a trained data augmentation model and a trained classification model. The training conditions may include one or more of a loss value meeting a preset loss value and a number of model iterations meeting a preset number of times.

[0100] It should be understood that the above-mentioned fault diagnosis model at least includes a trained classification model, that is, the trained data enhancement model and the trained classification model can be combined to obtain the above-mentioned fault diagnosis model; or, only the trained classification model can be used as the above-mentioned fault diagnosis model, thereby reducing the occupation of computing resources.

[0101] In an embodiment of the present application, multi-split training data is obtained. Since the multi-split training data includes labeled data and unlabeled data generated during the operation process of the multi-split air conditioner, and the above-mentioned labeled data includes a fault label that characterizes the fault state of the multi-split air conditioner, and the above-mentioned unlabeled data does not include a fault label that characterizes the fault state of the multi-split air conditioner, a data enhancement model is used to learn the distribution pattern of the fault labels in the above-mentioned labeled data, and then the data enhancement model is used to perform data enhancement on the unlabeled data. Data expansion can be performed on the actual operation data to obtain the first predicted label data, thereby expanding the data volume of the training sample. At the same time, the first predicted label data and the labeled data are input into the classification model, so that the above-mentioned classification model can learn the mapping relationship between the above-mentioned first predicted label data, the above-mentioned labeled data, and the fault labels therein, and obtain the first predicted diagnosis result, that is, the above-mentioned classification model can classify the labeled data of the operation process and the data-enhanced first predicted label data based on the above-mentioned mapping relationship, which can further increase the number of classifications of training samples. Therefore, using the above-mentioned multi-connected training data, the first predicted label data and the first predicted diagnosis result to train the data enhancement model and the classification model, the data enhancement model and the classification model can learn the characteristics of richer fault data, thereby improving the accuracy of multi-connected air conditioner fault diagnosis using the fault diagnosis model.

[0102] In some embodiments, the data augmentation model includes a generator, an encoder, and a discriminator; wherein the encoder may be an encoder in a variational autoencoder (VAE), which is used to map the input data to a latent space and learn the probability distribution characteristics of the data.

[0103] The above-mentioned unlabeled data is input into the data enhancement model to obtain the first predicted label data, including:

[0104] Inputting the unlabeled data and random noise into the generator to obtain generated labeled data; wherein the generated labeled data includes the fault label predicted by the generator;

[0105] Reconstructing the generated label data using the encoder to obtain a latent vector; wherein the latent vector is data with a distribution similar to that of the generated label data;

[0106] Inputting the latent vector and the labeled data into the discriminator to obtain a judgment result; wherein the judgment result is used to indicate the error of the data input to the discriminator;

[0107] The generator is reversely iterated using the above judgment result, and all the above generated label data generated by the reverse iteration process of the generator are obtained to obtain the above first predicted label data.

[0108] Among them, the above random noise and latent vector are data that conform to the Gaussian distribution.

[0109] Specifically, the unlabeled data is input into the generator, and random noise is introduced so that the generator generates labeled data based on the unlabeled data and the random noise output, and then the generated labeled data is input into the encoder, and the data is reconstructed by sampling the hidden layer in the encoder to obtain the latent vector; then the latent vector and the labeled data are input into the discriminator, and the error of the input data is calculated by the discriminator. For example, when the input data includes the latent vector and the labeled data, the judgment result can reflect the error between the latent vector and the labeled data; finally, the generator is continuously reversely iterated according to the judgment result, and all generated label data output by the reverse iteration process of the generator are integrated to obtain the first predicted label data.

[0110] In an embodiment of the present application, by introducing an encoder into the data augmentation model, the encoder can map the generated label data generated by the generator to the latent space through the hidden layer, better extract the main features therein, and reconstruct it into a latent vector with the same distribution, thereby helping the judge to better judge the error of the input data and improve the accuracy of the generation of the first predicted label data.

[0111] In some embodiments, the discriminator includes a first discriminator and a second discriminator, and the judgment result includes a first judgment result and a second judgment result;

[0112] The latent vector and the labeled data are input into the discriminator to obtain the judgment result, including:

[0113] Inputting the latent vector and random noise into the first discriminator to obtain the first judgment result; wherein the first judgment result is used to indicate the error between the latent vector and the random noise;

[0114] The unlabeled data and the generated labeled data are input into the second discriminator to obtain the second judgment result; wherein the second judgment result is used to indicate the error between the unlabeled data and the generated labeled data.

[0115] Specifically, the above-mentioned latent vector and random noise are used as inputs of the first discriminator, so that the first discriminator can calculate the error between the random noise (i.e., data that does not obey the latent space distribution) and the latent vector (i.e., effective representation data extracted from the latent space), and obtain a first judgment result, which prompts the generator reverse iteration process to generate more realistic data; the above-mentioned unlabeled data and the above-mentioned generated label data are input into the second discriminator, so that the first discriminator can calculate the error between the unlabeled data (i.e., real unlabeled data) and the generated label data (i.e., predicted labeled data), which prompts the generated label data output by the generator reverse iteration process to have the characteristics of unlabeled data.

[0116] In an embodiment of the present application, through the first discriminator and the second discriminator, the generator's reverse iterative process can generate more realistic data, and the output generated label data can have the characteristics of unlabeled data, thereby further improving the data quality of the first predicted label data.

[0117] In some embodiments, the classification model includes multiple base classifiers, wherein the base classifiers may be fully connected neural networks, and the classification model may be an ensemble learning model, for example, the classification model may be a Boosting ensemble learning model.

[0118] The first predicted label data and the labeled data are input into the classification model to obtain a first predicted diagnosis result, including:

[0119] Assigning the same initial weight to the first predicted label data and the labeled data, inputting the first predicted label data and the labeled data into the base classifier to obtain a classification result, and updating the initial weight using the classification result to obtain a target weight;

[0120] Inputting the first predicted label data and the labeled data corresponding to the target weight into the next base classifier to obtain a new classification result, and updating the target weight using the new classification result to obtain a new target weight;

[0121] Returning to the step of inputting the first predicted label data and the labeled data corresponding to the target weight into the next base classifier and subsequent steps until all the base classifiers output classification results;

[0122] Assigning weights to each of the above base classifiers according to the classification results of each of the above base classifiers to obtain base classifier weights corresponding to each of the above base classifiers;

[0123] The classification results output by all the base classifiers are weighted averaged according to the base classifier weights to obtain the first prediction diagnosis result.

[0124] Among them, the above classification results include predicted classification probability, classification labels, etc.

[0125] Specifically, the first predicted label data and the labeled data are used as training samples, and the same initial weight is assigned to each training sample. The training samples are used as inputs to a first base classifier to obtain a classification result output by the first base classifier. Then, the accuracy of the first base classifier is determined based on the classification result, and the weights of the training samples incorrectly classified by the first base classifier are increased based on the accuracy of the first base classifier to obtain a target weight. The training samples corresponding to the target weight (i.e., the training samples with updated weights) are input into a second base classifier to obtain a new classification result. The accuracy of the second base classifier is again determined based on the new classification result, and the weights of the training samples incorrectly classified by the second base classifier are increased based on the accuracy of the second base classifier to obtain a new target weight. The training samples corresponding to the new target weight are input into the next base classifier, and the above process is repeated until each base classifier outputs a corresponding classification result. For any base classifier, a weight is assigned based on the accuracy calculated from its classification result to obtain the base classifier weight corresponding to each base classifier. Then, the classification results output by each base classifier are weighted averaged based on the base classifier weights to obtain the first predicted diagnosis result. It should be noted that in the above classification model, the base classifier with a higher classification accuracy has a greater corresponding weight.

[0126] It should be understood that the number of the above-mentioned base classifiers can be set according to actual conditions and is not limited here.

[0127] In the embodiment of the present application, the second predictive diagnosis result is outputted by means of ensemble learning, so that the latter base classifier in the classification model will correct the error of the former base classifier, thereby improving the accuracy of the second predictive diagnosis result.

[0128] In some embodiments, the above-mentioned updating of the initial weights using the above-mentioned classification results to obtain the target weights includes:

[0129] Calculate the accuracy of the base classifier for classifying the first predicted label data and the labeled data based on the classification results;

[0130] The target weight is obtained by increasing the weights of the first predicted label data and the misclassified data in the labeled data based on the accuracy.

[0131] It should be understood that the above-mentioned training samples (i.e., the first predicted label data and the above-mentioned labeled data) include operating data and corresponding fault labels. After inputting into the base classifier, the above-mentioned base classifier will output the predicted classification results (including predicted classification probabilities, classification labels, etc.) based on the operating data, and then determine the incorrectly classified sample data based on whether the classification results are consistent with the fault labels, and then calculate the accuracy of the above-mentioned base classifier in classifying the sample data. Finally, when the accuracy is less than or equal to the preset threshold, the weight of the incorrectly classified data in the sample data is increased to obtain the above-mentioned target weight. Of course, when the accuracy is greater than the preset threshold, the weight of the correctly classified data in the sample data can also be reduced or unchanged. It should be noted that the training process of the remaining base classifiers is similar and will not be repeated here.

[0132] In an embodiment of the present application, by increasing the weight of sample data that was misclassified by the previous base classifier, the next base classifier can pay more attention to the misclassified sample data, thereby better correcting the errors of the previous base classifier, continuously reducing the deviation of the classification prediction of the base classifier, and improving the accuracy of the classification model in classification diagnosis prediction.

[0133] Correspondingly, the weights assigned to the base classifiers according to the classification results of the base classifiers include:

[0134] A weight is assigned to each of the base classifiers according to the accuracy of each of the base classifiers.

[0135] Optionally, the following formula may be used to assign weights to the base classifiers of the base classifiers:

[0136]

[0137] Among them, a i is the base classifier weight of the i-th base classifier, e i is the accuracy of the i-th base classifier.

[0138] In some embodiments, the training of the data augmentation model and the classification model based on the multi-online training data, the first predicted label data, and the first predicted diagnosis result includes:

[0139] Calculating a joint loss based on the multi-online training data, the first predicted label data, and the first predicted diagnosis result; wherein the joint loss includes a loss corresponding to the data augmentation model and a loss corresponding to the classification model;

[0140] The joint loss is used to update the model parameters of the data enhancement model and the classification model.

[0141] Specifically, a first loss is calculated based on the first predicted label data and the latent vector, wherein the first loss is the loss corresponding to the encoder; a second loss is calculated based on the latent vector and random noise, wherein the second loss is the loss corresponding to the first discriminator and the generator; a third loss is calculated based on the first predicted label data and the unlabeled data, wherein the third loss is the loss corresponding to the second discriminator and the generator; a fourth loss is calculated based on the first predicted diagnosis result and the labeled data, wherein the fourth loss is the loss corresponding to the classification model; a joint loss is calculated based on the first loss, the second loss, the third loss and the fourth loss, and when the joint loss does not meet the preset loss requirement (for example, the joint loss is less than a preset loss threshold) or the number of iterations does not meet the iteration requirement (for example, the number of iterations is less than a preset iteration threshold), the model parameters of the encoder, generator, discriminator and each base classifier in the classification model are updated until the joint loss meets the loss requirement, thereby obtaining the fault diagnosis model. For example, a gradient optimizer can be assigned to each sub-network (i.e., generator, encoder, discriminator, and base classifier), the entire model network parameters are differentiated according to the joint loss, the updated gradient is calculated, and then each sub-network is updated by the assigned gradient optimizer to obtain a fault diagnosis model.

[0142] Alternatively, the above-mentioned joint loss can be calculated by the following formula:

[0143] min E min G max D1 max D2 min C V(E,G,D1,D2,C)=L1+L2+L3+L4

[0144] Among them, E represents the encoder, G represents the generator, D1 represents the first discriminator, D2 represents the second discriminator, C represents the integrated classifier, L1 represents the first loss, L2 represents the second loss, L3 represents the third loss, and L4 represents the fourth loss;

[0145] L1=D KL (q(z|x)||P(z))-E q(z|x) [logP(x|z)]

[0146] Z=μ+ε×σ

[0147] Where Z represents the latent vector reconstructed by the hidden layer, μ and σ represent the mean and standard deviation of the posterior distribution respectively, ε represents random noise, x represents the input generated label data, q(z|x) represents the posterior distribution, P(z) represents the standard normal distribution, and D KL(q(z|x)||P(z)) represents the relative entropy of the posterior distribution and the standard normal distribution, and P(x|z) represents the similarity between the latent vector generated by hidden layer sampling and the generated label data;

[0148]

[0149]

[0150] Among them, P g is the distribution of latent vectors, represents sampling from the distribution of latent vectors, is the latent vector input to the first discriminator, P r is the distribution of multi-line training data, x r ~P r Represents sampling from the distribution of multi-online training data, x r is the random noise input to the first discriminator, for and x r Random interpolation data, Random interpolation data The distribution of represents sampling from the distribution of interpolated data, Represents the first discriminator for random interpolation data Derivative, ∈ is the random interpolation coefficient, λ is the penalty coefficient;

[0151]

[0152] in, P is the generated label data input to the second discriminator. u is the distribution of unlabeled data, x u is the unlabeled data input to the second discriminator, for and x u Random interpolation data, Random interpolation data The distribution of Represents the second discriminator for random interpolation data Derivative;

[0153]

[0154] in, Indicates generated label data and labeled data, P i represents the predicted classification probability output by the i-th base classifier, Represents the nuclear norm of the output probability matrix corresponding to the unlabeled data of the i-th base classifier.

[0155] In the embodiment of the present application, the above-mentioned joint loss can more accurately update the model parameters of each sub-network and improve the accuracy of model training.

[0156] In order to better illustrate the training process of the fault diagnosis model, the following Figure 4 For explanation, refer to Figure 4 The figure shows a flow chart of fault diagnosis model training. First, unlabeled data and random noise are input into the generator to obtain generated labeled data. The generated labeled data is then input into the encoder to obtain a latent vector. The latent vector and random noise are input into the first discriminator. Finally, the generated labeled data and unlabeled data are input into the second discriminator so that the first and second discriminators optimize the generator. All labeled data and generated labeled data are integrated to perform supervised training on the classification model. The implementation process of each sub-network during training can refer to the above embodiment and will not be repeated here.

[0157] It should be understood that the above-mentioned fault diagnosis model may include a trained data enhancement model (i.e., a generator, an encoder, and a discriminator) and a trained classification model (i.e., a trained base classifier), or may only include a trained classification model (i.e., a trained base classifier). The output of the trained classification model is the prediction result of the final output of the model. In order to further improve the accuracy of the output result of the fault diagnosis model, in another optional embodiment of the present application, the above-mentioned data enhancement model and the above-mentioned classification model are trained based on the above-mentioned multi-line training data, the above-mentioned first prediction label data, and the above-mentioned first prediction diagnosis result. After obtaining the fault diagnosis model, it further includes:

[0158] Using the unlabeled data as input to the trained classification model in the fault diagnosis model to obtain a second prediction diagnosis result output by the trained classification model;

[0159] Constructing second predicted label data based on the second predicted diagnosis result and the unlabeled data; wherein the second predicted label data includes a pseudo label determined based on the second predicted diagnosis result;

[0160] The trained classification model is fine-tuned using one or more of the labeled data, the first predicted label data, and the second predicted label data to obtain a fine-tuned classification model.

[0161] Specifically, the above-mentioned unlabeled data is input into the trained classification model, and the weighted average result of the classification results of each base classifier is output to obtain the above-mentioned second predictive diagnosis result, and then the results in the above-mentioned second predictive label results that are greater than the preset fault prediction threshold are converted into pseudo labels for the unlabeled data to obtain the above-mentioned second predictive label data, and then one or more of the above-mentioned labeled data, the above-mentioned first predictive label data and the above-mentioned second predictive label data are mixed, and the mixed data is input into the classification model to obtain a third predictive diagnosis result, and the error between the third predictive diagnosis result and the mixed data is calculated, and the model parameters of each base classifier are updated according to the error backpropagation, so as to fine-tune the above-mentioned trained classification model to obtain a fine-tuned classification model.

[0162] It should be noted that after fine-tuning, the above-mentioned fault diagnosis model may include the trained data augmentation model (i.e., the generator, encoder, and discriminator) and the fine-tuned classification model (i.e., the fine-tuned base classifier), or only include the fine-tuned classification model (i.e., the fine-tuned base classifier).

[0163] For example, unlabeled data is input into the trained classification model, and the results in the output second predictive diagnosis results that are greater than 0.6 (the preset fault prediction threshold) are converted into one-hot encoding and used as pseudo labels for the corresponding unlabeled data to obtain second predicted label data. After mixing the above second predicted label data with the labeled data, the trained classification model is fine-tuned according to the output third predictive diagnosis results.

[0164] In an embodiment of the present application, after model training, the classification model is fine-tuned again using data mixed with different labeled data to further improve the accuracy of the classification model output results.

[0165] In order to better illustrate the fine-tuning process of the fault classification model, the following Figure 5 For explanation, refer to Figure 5 The figure shows a flow chart of fine-tuning a classification model. First, unlabeled data is used as input to the trained classification model to obtain second predicted label data output by the classification model. The second predicted label data, labeled data, and generated label data are then integrated, and the classification model is fine-tuned using the integrated label data. The implementation process of the classification model during fine-tuning can be referenced to the above embodiments and will not be repeated here.

[0166] Corresponding to the fault diagnosis model training method described in the above embodiment, Figure 6 A flow chart of a fault diagnosis method provided in an embodiment of the present application is shown, which is applied to a multi-split air conditioner and is described in detail as follows:

[0167] S61. Acquire real-time operating data of the multi-split air conditioner during operation.

[0168] Among them, the above-mentioned real-time operation data refers to data reflecting the real-time operation status of the multi-split air conditioner, which may include: real-time operation data of the indoor unit (for example, real-time operation data of the fin heat exchanger, fan operation data, etc.), real-time operation data of the outdoor unit (for example, real-time operation data of the compressor, real-time temperature of the electrical box, etc.), real-time operation data of the refrigerant (for example, real-time flow rate of the refrigerant, real-time flow rate, etc.) and other real-time operation data (for example, operation data of the hydraulic module), etc. One or more.

[0169] S62: Using the above real-time operation data as input to a fault diagnosis model to obtain a fault diagnosis result output by the above fault diagnosis model, wherein the above fault diagnosis model is obtained according to the above fault diagnosis model training method.

[0170] It should be understood that the above-mentioned fault diagnosis model may include a trained data augmentation model (i.e., a generator, an encoder, and a discriminator) and a fine-tuned classification model (i.e., a fine-tuned base classifier), or only include a fine-tuned classification model (i.e., multiple fine-tuned base classifiers).

[0171] Specifically, when the fault diagnosis model includes a trained data enhancement model and a fine-tuned classification model, the above-mentioned real-time operation data is input into the generator in the above-mentioned trained data enhancement model, and then the encoder thereof is used to reconstruct the data, and the discriminator is used to make a judgment, so that the generator generates predicted label data, and then the above-mentioned predicted label data is input into the fine-tuned classification model to obtain the fault diagnosis result output by the fine-tuned classification model; or, when the fault diagnosis model only includes the fine-tuned classification model, the above-mentioned real-time operation data is directly input into the fine-tuned classification model to obtain the fault diagnosis result output by the fine-tuned classification model. It should be noted that when the above-mentioned fault diagnosis model predicts the fault diagnosis result, the operation process of each sub-network (i.e., the generator, encoder, discriminator, and classification model) can refer to the fault diagnosis model training process in the above embodiment, and will not be repeated here.

[0172] In the embodiment of the present application, since the above-mentioned fault diagnosis model training process learns more abundant fault data features, the accuracy of using fault diagnosis to diagnose multi-split air conditioner faults is improved.

[0173] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0174] Corresponding to the fault diagnosis model training method described in the above embodiment, Figure 7A structural diagram of a fault diagnosis model training device provided in an embodiment of the present application is shown. For ease of explanation, only the parts related to the embodiment of the present application are shown.

[0175] Reference Figure 7 The device may be a fault diagnosis model training device 71 , and the fault diagnosis model training device 71 may include a training data acquisition module 711 , a data enhancement module 712 , a classification module 713 and a training module 714 .

[0176] Reference Figure 7 , the above-mentioned fault diagnosis model training device 71 includes:

[0177] The training data acquisition module 711 is configured to acquire multi-split air conditioning training data; wherein the multi-split air conditioning training data includes labeled data and unlabeled data generated during the operation of the multi-split air conditioning system, wherein the labeled data includes a fault label representing a fault state of the multi-split air conditioning system, and the unlabeled data does not include a fault label representing a fault state of the multi-split air conditioning system;

[0178] The data augmentation module 712 is configured to input the unlabeled data into a data augmentation model to obtain first predicted label data; the data augmentation model is configured to learn the distribution pattern of the fault labels in the labeled data and output the first predicted label data containing the same distribution pattern of the fault labels;

[0179] A classification module 713 is configured to input the first predicted label data and the labeled data into a classification model to obtain a first predictive diagnosis result; the classification model is configured to learn a mapping relationship between the first predicted label data, the labeled data, and the fault labels in the first predicted label data and the labeled data, and output the first predictive diagnosis result having the same mapping relationship;

[0180] The training module 714 trains the data enhancement model and the classification model based on the multi-split training data, the first prediction label data and the first prediction diagnosis result to obtain a fault diagnosis model; wherein the fault diagnosis model is used to output the fault diagnosis results during the real-time operation of the multi-split air conditioner.

[0181] In some embodiments, the data augmentation model includes a generator, an encoder, and a discriminator; wherein the encoder may be an encoder in a variational autoencoder (VAE), which is used to map the input data to a latent space and learn the probability distribution characteristics of the data.

[0182] When the data enhancement module 712 inputs the unlabeled data into the data enhancement model to obtain the first predicted label data, the data enhancement module 712 includes:

[0183] Inputting the unlabeled data and random noise into the generator to obtain generated labeled data; wherein the generated labeled data includes the fault label predicted by the generator;

[0184] Reconstructing the generated label data using the encoder to obtain a latent vector; wherein the latent vector is data with a distribution similar to that of the generated label data;

[0185] Inputting the latent vector and the labeled data into the discriminator to obtain a judgment result; wherein the judgment result is used to indicate the error of the data input to the discriminator;

[0186] The generator is reversely iterated using the above judgment result, and all the above generated label data generated by the reverse iteration process of the generator are obtained to obtain the above first predicted label data.

[0187] In some embodiments, the discriminator includes a first discriminator and a second discriminator, and the judgment result includes a first judgment result and a second judgment result;

[0188] When the data enhancement module 712 inputs the latent vector and the labeled data into the discriminator to obtain a judgment result, the data enhancement module 712 includes:

[0189] Inputting the latent vector and random noise into the first discriminator to obtain the first judgment result; wherein the first judgment result is used to indicate the error between the latent vector and the random noise;

[0190] The unlabeled data and the generated labeled data are input into the second discriminator to obtain the second judgment result; wherein the second judgment result is used to indicate the error between the unlabeled data and the generated labeled data.

[0191] In some embodiments, the classification model includes multiple base classifiers, wherein the base classifiers may be fully connected neural networks, and the classification model may be an ensemble learning model, for example, the classification model may be a Boosting ensemble learning model.

[0192] When the classification module 713 inputs the first predicted label data and the labeled data into the classification model to obtain the first predicted diagnosis result, the classification module 713 includes:

[0193] Assigning the same initial weight to the first predicted label data and the labeled data, inputting the first predicted label data and the labeled data into the base classifier to obtain a classification result, and updating the initial weight using the classification result to obtain a target weight;

[0194] Inputting the first predicted label data and the labeled data corresponding to the target weight into the next base classifier to obtain a new classification result, and updating the target weight using the new classification result to obtain a new target weight;

[0195] Returning to the step of inputting the first predicted label data and the labeled data corresponding to the target weight into the next base classifier and subsequent steps until all the base classifiers output classification results;

[0196] Assigning weights to each of the above base classifiers according to the classification results of each of the above base classifiers to obtain base classifier weights corresponding to each of the above base classifiers;

[0197] The classification results output by all the base classifiers are weighted averaged according to the base classifier weights to obtain the first prediction diagnosis result.

[0198] In some embodiments, when the classification module 713 updates the initial weight using the classification result to obtain the target weight, it includes:

[0199] Calculate the accuracy of the base classifier for classifying the first predicted label data and the labeled data based on the classification results;

[0200] The target weight is obtained by increasing the weights of the first predicted label data and the misclassified data in the labeled data based on the accuracy.

[0201] In some embodiments, when the training module 714 trains the data augmentation model and the classification model based on the multi-line training data, the first predicted label data, and the first predicted diagnosis result, the training module 714 includes:

[0202] Calculating a joint loss based on the multi-online training data, the first predicted label data, and the first predicted diagnosis result; wherein the joint loss includes a loss corresponding to the data augmentation model and a loss corresponding to the classification model;

[0203] The joint loss is used to update the model parameters of the data enhancement model and the classification model.

[0204] In another optional embodiment of the present application, the fault diagnosis model training device 71 further includes a fine-tuning module. The fine-tuning module is configured to train the data enhancement model and the classification model based on the multi-line training data, the first predicted label data, and the first predicted diagnosis result to obtain the fault diagnosis model, including:

[0205] Using the unlabeled data as input to the trained classification model in the fault diagnosis model to obtain a second prediction diagnosis result output by the trained classification model;

[0206] Constructing second predicted label data based on the second predicted diagnosis result and the unlabeled data; wherein the second predicted label data includes a pseudo label determined based on the second predicted diagnosis result;

[0207] The trained classification model is fine-tuned using one or more of the labeled data, the first predicted label data, and the second predicted label data to obtain a fine-tuned classification model.

[0208] Corresponding to the fault diagnosis method described in the above embodiment, Figure 8 A structural diagram of a fault diagnosis device provided in an embodiment of the present application is shown. For ease of explanation, only the parts related to the embodiment of the present application are shown.

[0209] Reference Figure 8 The device may be a fault diagnosis device 81 , and the fault diagnosis device 81 may include a real-time data acquisition module 811 and a fault diagnosis module 812 .

[0210] Reference Figure 8 , the above-mentioned fault diagnosis device 81 includes:

[0211] A real-time data acquisition module 811 is used to acquire real-time operation data of the multi-split air conditioner during operation;

[0212] The fault diagnosis module 812 is configured to use the real-time operating data as input to a fault diagnosis model to obtain a fault diagnosis result output by the fault diagnosis model, wherein the fault diagnosis model is obtained according to the above-mentioned fault diagnosis model training method.

[0213] It should be noted that the information interaction, execution process, etc. between the devices / units are based on the same concept as the method embodiments of this application. Their specific functions and technical effects can be found in the method embodiment section and will not be repeated here.

[0214] Figure 9 This is a schematic diagram of the structure of an electronic device provided in one embodiment of the present application. Figure 9As shown, the electronic device 9 of this embodiment includes: at least one processor 90 ( Figure 9 Only one is shown in the figure), a memory 91, and a computer program 92 stored in the memory 91 and executable on the at least one processor 90. When the processor 90 executes the computer program 92, the steps of any of the method embodiments are implemented.

[0215] The electronic device 9 can be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc., or a microcomputer used in industrial equipment, smart home, multi-split air conditioner, etc. The electronic device can include, but is not limited to, a processor 90 and a memory 91. It will be understood by those skilled in the art that Figure 9 It is only an example of the electronic device 9 and does not constitute a limitation of the electronic device 9. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the electronic device may also include an input and sending device, a network access device, a bus, etc.

[0216] The processor 90 may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0217] In some embodiments, the memory 91 may be an internal storage unit of the electronic device 9, such as a hard disk or memory of the electronic device 9. The memory 91 may also be an external storage device of the electronic device 9, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 9. Furthermore, the memory 91 may include both an internal storage unit of the electronic device 9 and an external storage device. The memory 91 is used to store an operating system, application programs, a boot loader, data, and other programs, such as the program code of the computer program. The memory 91 may also be used to temporarily store data that has been sent or is about to be sent.

[0218] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the functional units and modules is used as an example for illustration. In actual applications, the function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0219] An embodiment of the present application also provides a network device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor implements the steps of any of the method embodiments when executing the computer program.

[0220] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the various method embodiments can be implemented.

[0221] An embodiment of the present application provides a computer program product. When the computer program product is run on an electronic device, the electronic device can implement the steps in the various method embodiments when executing the computer program product.

[0222] If the integrated unit is implemented in the form of 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 present application implements all or part of the process in the embodiment method by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of each method embodiment. The computer program includes computer program code, which can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the camera / electronic device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, mobile hard disk, magnetic disk or optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0223] In the embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0224] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0225] In the embodiments provided in this application, it should be understood that the disclosed devices / network equipment and methods can be implemented in other ways. For example, the device / network equipment embodiments described above are merely illustrative. For example, the division of the modules or units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0226] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0227] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. A fault diagnosis model training method, characterized in that: include: Acquire multi-split air conditioner training data; wherein the multi-split air conditioner training data includes labeled data and unlabeled data generated during the operation of the multi-split air conditioner, the labeled data includes a fault label representing a fault state of the multi-split air conditioner, and the unlabeled data does not include a fault label representing a fault state of the multi-split air conditioner; Inputting the unlabeled data into a data augmentation model to obtain first predicted label data; the data augmentation model is used to learn the distribution pattern of fault labels in the labeled data and output the first predicted label data containing the same distribution pattern of the fault labels; Inputting the first predicted label data and the labeled data into a classification model to obtain a first predicted diagnosis result; the classification model is used to learn a mapping relationship between the first predicted label data, the labeled data, and the fault labels in the first predicted label data and the labeled data, and output the first predicted diagnosis result having the same mapping relationship; The data enhancement model and the classification model are trained based on the multi-split training data, the first prediction label data and the first prediction diagnosis result to obtain a fault diagnosis model; wherein the fault diagnosis model is used to output the fault diagnosis result during the real-time operation of the multi-split air conditioner.

2. The fault diagnosis model training method according to claim 1, wherein: The data augmentation model includes a generator, an encoder, and a discriminator; the step of inputting the unlabeled data into the data augmentation model to obtain first predicted label data includes: Inputting the unlabeled data and random noise into the generator to obtain generated label data; wherein the generated label data includes the fault label predicted by the generator; Reconstructing the generated label data using the encoder to obtain a latent vector; wherein the latent vector is data with a distribution similar to that of the generated label data; Inputting the latent vector and the labeled data into the discriminator to obtain a judgment result; wherein the judgment result is used to indicate an error in the data input to the discriminator; The generator is reversely iterated using the judgment result, and all the generated label data generated by the reverse iteration process of the generator are obtained to obtain the first predicted label data.

3. The fault diagnosis model training method according to claim 2, wherein: The discriminator includes a first discriminator and a second discriminator, and the judgment result includes a first judgment result and a second judgment result; Inputting the latent vector and the labeled data into the discriminator to obtain a judgment result includes: Inputting the latent vector and random noise into the first discriminator to obtain a first judgment result; wherein the first judgment result is used to indicate an error between the latent vector and the random noise; The unlabeled data and the generated label data are input into the second discriminator to obtain a second judgment result; wherein the second judgment result is used to indicate the error between the unlabeled data and the generated label data.

4. The fault diagnosis model training method according to any one of claims 1 to 3, characterized in that: The classification model includes a plurality of base classifiers, and the first predicted label data and the labeled data are input into the classification model to obtain a first predicted diagnosis result, including: Assigning the same initial weight to the first predicted label data and the labeled data, inputting the first predicted label data and the labeled data into the base classifier to obtain a classification result, and updating the initial weight using the classification result to obtain a target weight; Inputting the first predicted label data and the labeled data corresponding to the target weight into the next base classifier to obtain a new classification result, and updating the target weight using the new classification result to obtain a new target weight; Returning to the step of inputting the first predicted label data and the labeled data corresponding to the target weight into the next base classifier and subsequent steps until all the base classifiers output classification results; Assigning weights to each base classifier according to the classification results of each base classifier to obtain base classifier weights corresponding to each base classifier; The classification results output by all the base classifiers are weighted averaged according to the base classifier weights to obtain the first predicted diagnosis result.

5. The fault diagnosis model training method according to claim 4, wherein: The step of updating the initial weight using the classification result to obtain the target weight includes: Calculating the accuracy of the base classifier in classifying the first predicted label data and the labeled data according to the classification result; Increasing the weights of the first predicted label data and misclassified data in the labeled data based on the accuracy to obtain the target weight; Correspondingly, allocating weights to the base classifiers according to the classification results of the base classifiers includes: A weight is assigned to each of the base classifiers according to the accuracy of each of the base classifiers.

6. The fault diagnosis model training method according to claim 4, characterized in that: The training of the data enhancement model and the classification model according to the multi-online training data, the first prediction label data, and the first prediction diagnosis result includes: Calculating a joint loss based on the multi-online training data, the first predicted label data, and the first predicted diagnosis result; wherein the joint loss includes a loss corresponding to the data augmentation model and a loss corresponding to the classification model; The joint loss is used to update model parameters of the data augmentation model and the classification model.

7. The fault diagnosis model training method according to claim 4, characterized in that: After training the data enhancement model and the classification model based on the multi-online training data, the first prediction label data, and the first prediction diagnosis result to obtain a fault diagnosis model, the method further includes: Using the unlabeled data as input to the trained classification model in the fault diagnosis model to obtain a second prediction diagnosis result output by the trained classification model; Constructing second predicted label data based on the second predicted diagnosis result and the unlabeled data; wherein the second predicted label data includes a pseudo label determined based on the second predicted diagnosis result; The trained classification model is fine-tuned using one or more of the labeled data, the first predicted label data, and the second predicted label data to obtain a fine-tuned classification model.

8. A fault diagnosis method, characterized in that: Applied to multi-split air conditioners, including: Acquiring real-time operating data of the multi-split air conditioner during operation; The real-time operation data is used as input of a fault diagnosis model to obtain a fault diagnosis result output by the fault diagnosis model, wherein the fault diagnosis model is obtained according to the fault diagnosis model training method according to any one of claims 1 to 7.

9. A fault diagnosis model training device, characterized in that: include: A training data acquisition module is used to acquire multi-split training data; wherein the multi-split training data includes labeled data and unlabeled data generated during the operation of the multi-split air conditioner, the labeled data includes a fault label representing the fault state of the multi-split air conditioner, and the unlabeled data does not include a fault label representing the fault state of the multi-split air conditioner; A data augmentation module, configured to input the unlabeled data into a data augmentation model to obtain first predicted label data; the data augmentation model is configured to learn the distribution pattern of fault labels in the labeled data and output the first predicted label data containing the same distribution pattern of the fault labels; a classification module, configured to input the first predicted label data and the labeled data into a classification model to obtain a first predictive diagnosis result; the classification model is configured to learn a mapping relationship between the first predicted label data, the labeled data, and the fault labels in the first predicted label data and the labeled data, and output the first predictive diagnosis result having the same mapping relationship; A training module is used to train the data enhancement model and the classification model based on the multi-split training data, the first prediction label data and the first prediction diagnosis result to obtain a fault diagnosis model; wherein the fault diagnosis model is used to output the fault diagnosis result during the real-time operation of the multi-split air conditioner.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 7 or claim 8 is implemented.

Images