Fault type detection method and device of heating and ventilation equipment, medium and electronic equipment

Abstract

The embodiment of the invention discloses a fault type detection method and device of heating and ventilation equipment, a medium and electronic equipment, and relates to the technical field of computers.The method comprises the steps that equipment abnormal parameter reconstruction errors are extracted from equipment abnormal detection data; according to the method, a first fault detection model and a second fault detection model are adopted to carry out fault type detection processing on equipment abnormal parameter reconstruction errors, and fault classification results of detection of the two models are obtained. The two fault detection models are obtained through training of a one-to-one classification architecture technology and a one-to-many classification architecture technology; the second fault detection model obtained by training through the one-to-many classification architecture technology can identify unknown fault types existing in the heating and ventilation equipment, the first fault detection model obtained by training through the one-to-one classification architecture technology can identify known fault types existing in the heating and ventilation equipment, and fault classification results of the two models are integrated. The fault type of the heating and ventilation equipment which is detected to be abnormal can be accurately identified.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method, device, medium and electronic equipment for detecting the fault type of a heating and ventilation equipment. Background Art

[0002] With the acceleration of industrialization, refrigeration equipment plays an increasingly important role in all walks of life and is widely used in food storage, air conditioning systems, chemical production, pharmaceutical industry and other fields. Refrigeration equipment usually works under high load conditions, requiring it to maintain high efficiency and stable performance. However, due to long-term use, changes in environmental factors and improper operation, refrigeration equipment often fails during operation, resulting in reduced system efficiency, energy waste, and even production stagnation and economic losses. Therefore, it is particularly important to perform timely and accurate fault diagnosis of refrigeration equipment.

[0003] In the related art, fault diagnosis of refrigeration equipment relies on manual inspection and empirical judgment, which results in the problem that manual fault diagnosis is time-consuming and labor-intensive. In addition, since the physical parameters of refrigeration equipment are complicated and exhibit nonlinear coupling characteristics, it is also very difficult to complete fault diagnosis relying on manpower. Therefore, how to efficiently and accurately detect faults in refrigeration equipment is a technical problem that needs to be solved urgently. Summary of the invention

[0004] The present application provides a method, device, medium and electronic device for detecting the fault type of a heating and ventilation device. The technical solution is as follows: In a first aspect, an embodiment of the present application provides a method for detecting a fault type of a HVAC device, the method comprising: Determine the device abnormality detection data corresponding to the HVAC device, and perform abnormal data extraction processing based on the device abnormality detection data to obtain a device abnormality parameter reconstruction error; A first fault detection model is used to perform fault type detection processing on the equipment abnormal parameter reconstruction error to obtain a first fault classification result, and a second fault detection model is used to perform fault type detection processing on the equipment abnormal parameter reconstruction error to obtain a second fault classification result, wherein the first fault detection model is obtained after model training based on an initial first fault detection model of a one-to-one classification architecture, and the second fault detection model is obtained after model training based on an initial second fault detection model of a one-to-many classification architecture; If the second fault classification result is a known fault type, determining a target fault type of the HVAC equipment based on the first fault classification result; If the second fault classification result is an unknown fault type, equipment fault diagnosis is performed on the HVAC equipment.

[0005] In a second aspect, an embodiment of the present application provides a device for detecting a fault type of a HVAC device, the device comprising: A data extraction module is used to determine the device abnormality detection data corresponding to the HVAC equipment, and perform abnormal data extraction processing based on the device abnormality detection data to obtain the device abnormality parameter reconstruction error; A model identification module, used to use a first fault detection model to perform fault type detection processing on the equipment abnormal parameter reconstruction error to obtain a first fault classification result, and use a second fault detection model to perform fault type detection processing on the equipment abnormal parameter reconstruction error to obtain a second fault classification result, wherein the first fault detection model is obtained after model training based on an initial first fault detection model of a one-to-one classification architecture, and the second fault detection model is obtained after model training based on an initial second fault detection model of a one-to-many classification architecture; a first diagnostic module, configured to determine a target fault type of the HVAC equipment based on the first fault classification result if the second fault classification result is a known fault type; The second diagnostic module is used to perform equipment fault diagnosis on the HVAC equipment if the second fault classification result is an unknown fault type.

[0006] In a third aspect, an embodiment of the present application provides a computer storage medium, wherein the computer storage medium has a plurality of instructions, and the instructions are suitable for being loaded by a processor and executing the above-mentioned method.

[0007] In a fourth aspect, an embodiment of the present application provides an electronic device, which may include: a memory and a processor; wherein the memory stores a computer program, and the computer program is suitable for being loaded by the memory and executing the above method.

[0008] The beneficial effects brought by the technical solution provided by the embodiment of the present application include at least: The fault type detection method for HVAC equipment provided in the embodiment of the present application extracts the equipment abnormality parameter reconstruction error from the equipment abnormality detection data, and respectively uses a first fault detection model and a second fault detection model to perform fault type detection processing on the equipment abnormality parameter reconstruction error to obtain fault classification results detected by the two models. Since the two fault detection models are respectively trained using a one-to-one classification architecture technology and a one-to-many classification architecture technology, the second fault detection model trained using the one-to-many classification architecture technology can identify unknown fault types existing in the HVAC equipment, and the first fault detection model trained using the one-to-one classification architecture technology can identify known fault types existing in the HVAC equipment. By combining the fault classification results of the two models, the fault type of the HVAC equipment detected to have an abnormality can be accurately identified. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.

[0010] Figure 1 It is a flowchart of a method for detecting a fault type of a HVAC device provided in an embodiment of the present application; Figure 2 It is a flow chart of another method for detecting the fault type of HVAC equipment provided in an embodiment of the present application; Figure 3 It is a flowchart of a training method for a first fault detection model provided in an embodiment of the present application; Figure 4 It is a flowchart of a training method for a second fault detection model provided in an embodiment of the present application; Figure 5 It is a structural schematic diagram of a fault type detection device for HVAC equipment provided in an embodiment of the present application; Figure 6 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0011] In order to make the purpose, features, and advantages of the embodiments of the present application more obvious and understandable, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0012] In the description of the present application, it should be understood that the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In the description of the present application, it should be noted that, unless otherwise clearly specified and limited, "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device comprising a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units that are not listed, or optionally also includes other steps or units inherent to these processes, methods, products or devices. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood in specific circumstances. In addition, in the description of the present application, unless otherwise specified, "multiple" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships can exist, for example, A and / or B can be represented: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the associated objects before and after are a kind of "or" relationship.

[0013] In one embodiment, Figure 1 As shown, a method for detecting the fault type of HVAC equipment is proposed, which can be implemented by a computer program and can be run on a fault type detection device for HVAC equipment based on the von Neumann system. The computer program can be integrated into an application or run as an independent tool application.

[0014] Specifically, the fault type detection method of the HVAC equipment includes: S101, determining the equipment abnormality detection data corresponding to the HVAC equipment, and performing abnormal data extraction processing based on the equipment abnormality detection data to obtain equipment abnormality parameter reconstruction error.

[0015] Among them, HVAC equipment refers to equipment used to adjust indoor air temperature, humidity, circulation, air quality and other environmental conditions. Specifically, HVAC equipment can be refrigeration equipment, and refrigeration equipment can include multi-split units and chillers. Among them, multi-split units are mainly connected to multiple indoor units through an outdoor unit. The outdoor unit and the indoor unit are cooled and heated through refrigerant pipes. The indoor unit adjusts the refrigerant flow to achieve temperature control requirements in different rooms. Chillers are equipment used to provide cooling water. They are commonly found in large-scale air-conditioning systems. Chillers cool the water through a refrigerant circulation system, and then send the cooling water to the air handling unit (such as an air-conditioning fan coil) to achieve indoor temperature control.

[0016] The equipment anomaly detection data refers to the parameter data related to the operating parameters of the equipment with anomalies. Specifically, the equipment anomaly detection data may include the anomaly detection results of the HVAC equipment in multiple preset detection cycles, the total anomaly measurement value, and the equipment operating parameter reconstruction error.

[0017] Among them, the abnormal detection result is used to indicate whether the detection result of the HVAC equipment within the preset detection period is a normal result or an abnormal result. For example, when one day is used as the preset detection period, the abnormal detection result can be used to indicate the daily normal result or the daily abnormal result.

[0018] The total abnormality measurement value can be used to represent the overall abnormality level of the HVAC equipment within a preset detection period. For example, when one day is used as the preset detection period, the total abnormality measurement value can be used to measure the overall abnormality level of the equipment on that day.

[0019] The equipment operation parameter reconstruction error may include the reconstruction errors of multiple equipment parameters. The equipment operation parameter reconstruction error may include the reconstruction errors of normal operation parameters and the reconstruction errors of abnormal operation parameters. For example, when one day is used as the preset detection period, the equipment operation parameter reconstruction error may be used to represent the reconstruction errors of multiple equipment parameters of the equipment on that day. The reconstruction error is used to represent the error value obtained by calculating the difference between the reconstruction result of the equipment operation parameter calculated by the autoencoder model and the original data of the equipment operation parameter. In the context of the integration of multi-feature component working condition information, the autoencoder reconstructs the data of the equipment operation parameter by learning the relationship between various parameters of the equipment. The reconstruction error may be calculated using the mean square error or other indicators to measure the difference. The equipment operation parameter reconstruction error may reflect whether the equipment behaves abnormally under certain working conditions. For example, if the reconstruction error of a certain parameter (such as temperature, pressure, etc.) is large, it may indicate that these parameters are abnormal, indicating that the equipment may have some kind of fault.

[0020] The equipment abnormal parameter reconstruction error includes the reconstruction errors of multiple equipment parameters of the HVAC equipment within a preset detection period when the abnormality exists. For example, when one day is used as the preset detection period, the equipment abnormal parameter reconstruction error can be used to represent the reconstruction errors of multiple equipment parameters of the HVAC equipment on the abnormal day.

[0021] In some embodiments, the equipment operation data of the HVAC equipment can be obtained from the cloud, and then the equipment anomaly detection model is called to perform equipment anomaly detection processing on the equipment operation data to obtain the equipment anomaly detection data. Specifically, the equipment operation data may include data of various equipment parameters during the equipment operation process, and the equipment parameters may include but are not limited to temperature parameters, pressure parameters, flow parameters, electrical parameters, operation mode and control parameters, power and efficiency parameters. The equipment anomaly detection model can perform equipment anomaly detection processing on the equipment operation data to obtain the anomaly detection results, total anomaly measurement values, and equipment operation parameter reconstruction errors in different preset detection cycles. When the total anomaly measurement value of a certain preset detection cycle is greater than the anomaly measurement threshold, the equipment anomaly detection model determines that the preset detection cycle is a preset detection cycle with anomalies. When the number of preset detection cycles with anomalies exceeds the cycle threshold, the equipment anomaly detection model determines that the HVAC equipment has anomalies, and the equipment anomaly detection model outputs equipment anomaly detection data according to the preset detection cycle with anomalies. Further, according to the anomaly detection results in the equipment anomaly detection data, the equipment anomaly parameter reconstruction error indicated by the anomaly detection result is extracted from the equipment operation parameter reconstruction error.

[0022] S102, using a first fault detection model to perform fault type detection processing on an error in reconstructing abnormal parameters of the equipment to obtain a first fault classification result, and using a second fault detection model to perform fault type detection processing on an error in reconstructing abnormal parameters of the equipment to obtain a second fault classification result.

[0023] The first fault detection model is obtained after model training based on an initial first fault detection model of a one-to-one classification architecture, and the second fault detection model is obtained after model training based on an initial second fault detection model of a one-to-many classification architecture.

[0024] The first fault detection model may include multiple binary classification sub-models, each of which may be used to identify two different fault types. The second fault detection model may include multiple single-category classification sub-models, each of which may be used to identify specific fault types and non-specific fault types. For example, the fault types include the first type, the second type, and the third type. The first fault detection model may include three binary classification sub-models, one of which may be used to identify the first type and the second type, another of which may be used to identify the second type and the third type, and the remaining binary classification sub-model may be used to identify the first type and the third type. The second fault detection model may also include three single-category classification sub-models, one of which may identify the first type and the non-first type, another of which may identify the second type and the non-second type, and the remaining of which may identify the third type and the non-third type.

[0025] The first fault classification result includes the probability that the first fault detection model predicts that the HVAC equipment belongs to multiple reference fault types respectively. The second fault classification result includes the probability that the second fault detection model predicts that the HVAC equipment belongs to multiple reference fault types and multiple non-reference fault types respectively. Specifically, the reference fault type can be a possible fault type of the HVAC equipment determined based on expert experience, and the reference fault type is determined before executing the fault type detection method for the HVAC equipment of the embodiment of the present application. For example, taking the multi-split unit of the HVAC equipment as an example, the possible fault types of the multi-split unit determined based on expert experience may include but are not limited to compressor failure, refrigerant leakage, sensor failure, external unit electronic expansion valve failure, external unit heat exchanger dirt and blockage, internal unit electronic expansion valve failure, and internal unit heat exchanger dirt and blockage.

[0026] In some embodiments, the device abnormal parameter reconstruction error can be first input into the first fault detection model, and then the device abnormal parameter reconstruction error can be input into the second fault detection model. Alternatively, the device abnormal parameter reconstruction error can be first input into the first fault detection model, and then the device abnormal parameter reconstruction error can be input into the second fault detection model. Alternatively, the device abnormal parameter reconstruction error can be input into the second fault detection model while the device abnormal parameter reconstruction error is input into the first fault detection model. The first fault classification result can be obtained by predicting the probability that the HVAC equipment belongs to multiple reference fault types respectively through the first fault detection model, and the second fault classification result can be obtained by predicting the probability that the HVAC equipment belongs to the reference fault type and the non-reference fault type respectively through the second fault detection model.

[0027] S103: If the second fault classification result is a known fault type, determine the target fault type of the HVAC equipment based on the first fault classification result.

[0028] In some embodiments, when there is at least one reference fault type in the second fault type result whose probability prediction result is greater than or equal to the corresponding preset probability threshold, the second fault classification result can be determined as a known fault type. In this scenario, the target fault type of the HVAC equipment can be determined based on the first fault classification result. Specifically, the reference fault type with the maximum predicted probability in the first fault classification result can be used as the target fault type of the HVAC equipment.

[0029] S104: If the second fault classification result is an unknown fault type, perform equipment fault diagnosis on the HVAC equipment.

[0030] In some embodiments, when the probability prediction results of all reference fault types in the second fault type result are less than the corresponding preset probability threshold, the second fault classification result can be determined to be an unknown fault type. In this scenario, the HVAC equipment can be diagnosed for equipment faults by starting the expert review process to accurately diagnose the fault type of the HVAC equipment based on expert experience. For example, an expert review notification can be sent to the user terminal to inform the user terminal to start the expert review process; or, the expert review process can be automatically created and sent to the user terminal.

[0031] Optionally, after the fault type of the HVAC equipment is determined through expert review, the equipment abnormal parameter reconstruction error of the fault type can be stored in the model training pool, and the equipment abnormal parameter reconstruction error can be used as subsequent model training data.

[0032] In the fault type detection method for HVAC equipment provided in an embodiment of the present application, by extracting the equipment abnormality parameter reconstruction error from the equipment abnormality detection data, the first fault detection model and the second fault detection model are used to perform fault type detection processing on the equipment abnormality parameter reconstruction error, and the fault classification results of the two model detections are obtained. Since the two fault detection models are trained using the one-to-one classification architecture technology and the one-to-many classification architecture technology respectively, the second fault detection model trained using the one-to-many classification architecture technology can identify unknown fault types in the HVAC equipment, and the first fault detection model trained using the one-to-one classification architecture technology can identify known fault types in the HVAC equipment. By combining the fault classification results of the two models, the fault type of the HVAC equipment detected to have abnormalities can be accurately identified.

[0033] See next Figure 2 , which is a flow chart of another embodiment of a method for detecting a fault of a HVAC device according to an embodiment of the present application.

[0034] Specifically, the fault type detection method of the HVAC equipment includes: S201, acquiring equipment operation data of HVAC equipment, and using an equipment anomaly detection model to perform equipment anomaly detection processing on the equipment operation data to obtain equipment anomaly detection data.

[0035] Among them, the equipment operation data may include data of various equipment parameters during the equipment operation process. The equipment parameters may include but are not limited to temperature parameters, pressure parameters, flow parameters, electrical parameters, operation mode and control parameters, power and efficiency parameters. Among the above different types of parameters, each type of parameter can be further subdivided into specific parameters of different key parts in the HVAC equipment. Taking the HVAC equipment as a multi-split unit as an example, the temperature parameters may include saturated condensing temperature, outdoor heat exchanger outlet temperature, condenser outlet temperature, compressor suction temperature, compressor exhaust temperature, etc. The electrical parameters may include current, voltage, compressor frequency and other parameters. The operation mode and control parameters may include operation mode, expansion valve opening and other parameters.

[0036] In some embodiments, the equipment anomaly detection model can be trained for the equipment anomaly detection scenario, and the equipment operation data of the HVAC equipment can be obtained from the cloud. The pre-trained equipment anomaly detection model is called to perform equipment anomaly detection processing on the equipment operation data to obtain equipment anomaly detection data. The equipment anomaly detection data includes the anomaly detection results, total anomaly measurement values, and equipment operation parameter reconstruction errors of the HVAC equipment within multiple preset detection cycles. Each preset detection cycle corresponds to an anomaly detection result, a total anomaly measurement value, and a set of equipment operation parameter reconstruction errors. The equipment anomaly detection model can integrate the functions of the autoencoder model. In the training stage of the equipment anomaly detection model, by learning the normal operating data of the HVAC equipment under different working conditions, the trained equipment anomaly detection model can reconstruct the target operating data of the HVAC equipment within a preset detection period. The equipment anomaly detection model can calculate the difference between the reconstructed target operating data and the actual operating data of the equipment (the acquired equipment operating data) to obtain the equipment operating parameter reconstruction error; further, the equipment anomaly detection model determines the total anomaly measurement value of the HVAC equipment within the preset detection period based on the equipment operating parameter reconstruction error of the HVAC equipment within the preset detection period; further, the equipment anomaly detection model determines the anomaly detection result of the HVAC equipment within the preset detection period based on the total anomaly measurement value of the HVAC equipment within the preset detection period. The equipment anomaly detection model determines whether the HVAC equipment has a fault based on the number of preset detection periods in which the anomaly detection result is an abnormal result, and outputs the equipment anomaly detection data after determining that the HVAC equipment has a fault.

[0037] S202, determining the device abnormality detection result and the device operating parameter reconstruction error from the device abnormality detection data, performing abnormal data extraction processing on the device operating parameter reconstruction error based on the device abnormality detection result, and obtaining the device abnormality parameter reconstruction error.

[0038] In some embodiments, the device abnormality detection result and the device operation parameter reconstruction error corresponding to each preset detection period are determined from the device abnormality detection data, and when the device abnormality detection result is an abnormal result, the device operation reconstruction error corresponding to the preset detection period where the abnormal result is located is used as the device abnormality parameter reconstruction error. For example, when one day is used as the preset detection period, if the device abnormality detection result corresponding to a certain day is an abnormal result, then the device operation parameter reconstruction error corresponding to the day is the device abnormality parameter reconstruction error.

[0039] S203, using the first fault detection model to perform fault type probability prediction on the equipment abnormal parameter reconstruction error to obtain a multi-classification probability voting matrix.

[0040] It can be understood that the first fault detection model is obtained after model training based on the initial first fault detection model of the one-to-one classification architecture. Figure 3 The flowchart of the training method of the first fault detection model shown in FIG. 1 is a flowchart of the training method of the first fault detection model, and the training method may include the following steps: S301, creating an initial first fault detection model for a HVAC equipment fault detection scenario based on a one-to-one classification architecture, wherein the initial first fault detection model includes at least one initial two-classification sub-model; S302, obtaining a first sample abnormal reconstruction error, and marking a first fault type label on the first sample abnormal reconstruction error; S303, determining a first sample input feature corresponding to the first sample abnormal reconstruction error, determining a classification prediction contribution of the first sample input feature to the initial binary classification sub-model, and selecting a second sample input feature for the initial binary classification sub-model from the first sample input feature based on the classification prediction contribution; S304, selecting a second sample abnormal reconstruction error from the first sample abnormal reconstruction error based on the second sample input feature, and obtaining a second fault type label marked by the second sample abnormal reconstruction error; S305, use the second sample abnormal reconstruction error to perform at least one round of model training on the initial first fault detection model. During the model training process, the second sample abnormal reconstruction error is subjected to fault type detection processing based on the initial first fault detection model to obtain a first predicted fault type. Based on the first predicted fault type and the second fault type label, the model loss function and the binary classification average recall rate indicator are used to adjust the model parameters of the initial first fault detection model until the model training end conditions are met to obtain the first fault detection model.

[0041] In step S301, a possible reference fault type of the HVAC equipment is obtained based on expert experience. Specifically, the number of reference fault types is N. A one-to-one classification architecture technology is used to create a The first initial fault detection model is composed of initial binary classification sub-models, each initial binary classification sub-model is configured to identify two different reference fault types, and any two initial binary classification sub-models can be used to identify at most one same reference fault type.

[0042] In step S302, the first sample abnormal reconstruction error may include sample device abnormal parameter reconstruction errors of multiple sample HVAC devices with abnormalities, and these test HVAC devices with abnormalities may include sample HVAC devices whose fault types are reference fault types. In the first sample abnormal reconstruction error, the sample device abnormal parameter reconstruction errors of the sample HVAC devices belonging to the reference fault type are annotated with corresponding reference fault type labels, and the first fault type label may include multiple reference fault type labels.

[0043] In step S303, all sample parameter features in the first sample abnormal reconstruction error are determined, and the sample parameter features are used as the first sample input features. For example, taking the HVAC equipment as a multi-split unit as an example, the first sample abnormal reconstruction error may include reconstruction errors of parameters such as subcooling, superheat, compressor suction temperature, compressor exhaust temperature, compressor frequency, and expansion valve opening. Subcooling, superheat, compressor suction temperature, compressor exhaust temperature, compressor frequency, and expansion valve opening are all sample parameter features. Therefore, subcooling, superheat, compressor suction temperature, compressor exhaust temperature, compressor frequency, and expansion valve opening can all be used as the first sample input feature. After determining the first sample input feature, the classification prediction contribution of the first sample input feature to the initial two-classification sub-model can be determined by the Shapley value corresponding to the first sample input feature. The classification prediction contribution can be sorted in order from high to low, and the first sample input features corresponding to the classification prediction contribution of the preset number of first sample input features are used as the second sample input features for the initial two-classification sub-model.

[0044] Specifically, the implementation method of calculating the Shapley value corresponding to the first sample input feature is as follows: for each initial binary classification sub-model, the sample reconstruction errors corresponding to the two fault types that the initial binary classification sub-model can identify are taken from the first sample abnormal reconstruction error, the sample reconstruction error is used to train the initial binary classification sub-model to obtain a preliminarily trained initial binary classification sub-model, and the contribution of each first sample input feature to the preliminarily trained initial binary classification sub-model is calculated by a preset Shapley value calculation method to obtain a Shapley value matrix, the number of rows of the Shapley value matrix is ​​the same as the number of samples of the sample reconstruction error, the number of columns of the Shapley value matrix is ​​the same as the number of features of the sample reconstruction error, the absolute value processing is performed on the Shapley value matrix to obtain the contribution of each feature of each sample (sample reconstruction error) to the model prediction as a positive class, and further, the average of each column of the Shapley value matrix is ​​obtained to obtain the average contribution of each feature to the model prediction as a positive class under the support of the entire training set samples, and the average contribution is used as the Shapley value corresponding to each first sample input feature. The Shapley value can be used to represent the importance of the feature to the model.

[0045] It can be understood that customizing the input features of the binary classification model one by one based on the Shapley value, rather than using the same set of all input features for all models, can further improve the diagnostic performance of the first fault detection model.

[0046] In step S304, a second sample abnormal reconstruction error corresponding to the second sample input feature is selected from the first sample abnormal reconstruction error, and the second sample abnormal reconstruction error is used as sample input data for subsequent training of the initial binary classification sub-model.

[0047] In step S305, the model parameters of the initial first fault detection model are adjusted based on the first predicted fault type and the second fault type label using the model loss function and the binary classification average recall rate indicator until the model training end condition is met to obtain the first fault detection model. Specifically, it may include: calculating the model loss value based on the first predicted fault type and the second fault type label using the model loss function, adjusting the model parameters of at least one initial binary classification sub-model based on the model loss value to obtain at least one reference binary classification sub-model; determining a first reference hyperparameter set for the reference binary classification sub-model from a first preset hyperparameter search domain, updating the hyperparameters of the reference binary classification sub-model based on the first reference hyperparameter set, and calculating the reference binary classification sub-model for each updated reference binary classification sub-model. The average recall rate of classification; determine whether the number of first hyperparameter updates of the reference binary classification sub-model is equal to the first preset number threshold; if the number of first hyperparameter updates is not equal to the first preset number threshold, determine the first updated hyperparameter set for the reference binary classification sub-model from the first preset hyperparameter search domain, use the first updated hyperparameter set as the first reference hyperparameter set, and execute the step of updating the hyperparameters of the reference binary classification sub-model based on the first reference hyperparameter set; if the number of first hyperparameter updates is equal to the first preset number threshold, determine the first target hyperparameter set for the reference binary classification sub-model from the first preset hyperparameter search domain based on the reference binary classification average recall rate, update the hyperparameters of the reference binary classification sub-model based on the first target hyperparameter set, and obtain a first fault detection model.

[0048] Specifically, the first preset hyperparameter search domain includes at least one hyperparameter set corresponding to each reference binary classification sub-model. The hyperparameters in the hyperparameter set refer to parameters that cannot be directly learned through data. The hyperparameters may include but are not limited to model integration weights corresponding to each reference binary classification sub-model.

[0049] Specifically, the reference binary classification average recall rate is the average of the verified recall rates obtained by using the leave-one-out cross-validation model evaluation method. In the embodiment of the present application, the leave-one-out cross-validation model evaluation method sets aside the sample equipment abnormal parameter reconstruction error of a sample HVAC device as a verification set each time the model is trained, and the sample equipment abnormal parameter reconstruction errors of all remaining sample HVAC devices are used as training sets, and the process is repeated until the sample equipment abnormal parameter reconstruction errors of each sample HVAC device are used as a verification set. The reference binary classification average recall rate is calculated as follows: for each reference binary classification sub-model, assuming that the data set is sample data of N sample HVAC devices (including sample equipment abnormal parameter reconstruction errors of N sample HVAC devices), each time the sample equipment abnormal parameter reconstruction error of one of the sample HVAC devices is used as a verification set, and the remaining The sample equipment abnormal parameter reconstruction error of each sample HVAC equipment is used as a training set, a first reference hyperparameter set corresponding to the reference binary classification sub-model is selected from the first preset hyperparameter search domain, the first reference hyperparameter set is used to update the hyperparameters of the reference binary classification sub-model, the training set is used to train the reference binary classification sub-model with the updated hyperparameters, and the verification set is used to verify, and the recall rate of this verification is calculated, which is the recall rate corresponding to the reference binary classification sub-model after the hyperparameters are updated using the first reference hyperparameter set. The training process of leaving one verification set out each time is repeated N times until the sample equipment abnormal parameter reconstruction error of each sample HVAC equipment is used as a verification set once, and the average recall rate obtained by each verification is calculated to obtain the reference binary classification average recall rate. The above method of calculating the reference binary classification average recall rate can be used to obtain the reference binary classification average recall rate corresponding to the two fault types that can be identified by each reference binary classification sub-model. For a certain fault type, the recall rate is equal to the number of sample HVAC equipment whose fault type is correctly diagnosed divided by the total number of sample HVAC equipment belonging to the fault type.

[0050] Specifically, for each reference binary classification sub-model, a reference hyperparameter set corresponding to the maximum reference binary classification average recall rate is selected as the first target hyperparameter set, and the first target hyperparameter set is the optimal hyperparameter set of the reference binary classification sub-model. The first target hyperparameter set is used to update the hyperparameters of the reference binary classification sub-model to obtain a first fault detection model.

[0051] It can be understood that when the binary sub-model is optimized for hyperparameters, the embodiment of the present application selects the complete time series data of a sample HVAC equipment as a test set, rather than using the conventional test set selection method in the related art. The conventional test set selection method is random division or stratified sampling, that is, a certain proportion of data is randomly selected from the entire data set as test data, or a fixed proportion of samples are stratified as test data. For time series data, there is a large correlation between the previous and next data. In order to ensure that the evaluated model does not learn future information, the conventional test set selection method usually uses the later continuous time period data as the test set. When there are very few samples, the conventional test set selection method may also use the leave-one-out method or the leave-P method, that is, leave 1 or P samples as the test set each time. Compared with the conventional test set selection method, the embodiment of the present application selects the complete time series data of a sample HVAC equipment as a test set, which fully simulates the scenario of large-scale application on new HVAC equipment after the algorithm training is completed, and the test set selection method of the embodiment of the present application is used for the hyperparameter optimization process, so that the trained model has better robustness.

[0052] In step S203, the number of rows of the multi-classification probability voting matrix is ​​equal to the total number of data items of the equipment abnormal parameter reconstruction error, and the number of columns of the multi-classification probability voting matrix is ​​equal to the total number of reference fault types, wherein the equipment abnormal parameter reconstruction error of a HVAC device within a prediction detection week can be called a piece of data.

[0053] Specifically, executing step S203 may include: determining the target model input features for at least one binary classification sub-model in the first fault detection model from the equipment abnormality parameter reconstruction error, performing fault type detection processing on the target model input features through at least one binary classification sub-model and the model integration weights corresponding to the binary classification sub-model, and obtaining an initial multi-classification probability voting matrix; determining the equipment abnormality measurement value corresponding to the equipment abnormality parameter reconstruction error based on the equipment abnormality detection data, determining the target weight corresponding to the equipment abnormality measurement value, and performing probability weighted processing on the initial multi-classification probability voting matrix based on the target weight to obtain a multi-classification probability voting matrix.

[0054] It can be understood that the target model input feature is the model input feature data corresponding to multiple features that have a large classification prediction contribution to the binary classification sub-model selected from the equipment abnormal parameter reconstruction error, and the target model input feature is input into the first fault detection model. At least one binary classification sub-model in the first fault detection model performs fault detection processing on its own target model input feature. In the process of fault detection processing, each binary classification sub-model performs probability prediction processing on the target model input feature through its own corresponding model integration weights to obtain the first probability prediction value corresponding to each binary classification sub-model and the reference fault type corresponding to the first probability prediction value, and an initial multi-classification probability voting matrix is ​​generated based on the first probability prediction value and the reference fault type corresponding to the first probability prediction value. Furthermore, the device abnormality measurement value corresponding to the preset detection period where the device abnormality parameter reconstruction error is located can be determined from the device abnormality detection data, and the corresponding relationship between the preset measurement value and the weight can be obtained. The target weight corresponding to the device abnormality measurement value is determined from the corresponding relationship between the preset measurement value and the weight. The first probability prediction value corresponding to the device abnormality parameter reconstruction error is weighted using the target weight to obtain a second probability prediction value. The first probability prediction value corresponding to the device abnormality parameter reconstruction error is the first probability prediction value corresponding to the target model input feature that has a corresponding relationship with the device abnormality parameter reconstruction error. In the initial multi-classification probability voting matrix, the second probability prediction value is used to update the first probability prediction value to obtain a multi-classification probability voting matrix.

[0055] It can be understood that the model integration weight corresponding to each of the above-mentioned binary classification sub-models is optimized based on the reference binary classification average recall rate during the model training process, and is the model integration weight obtained under the condition of achieving the maximum reference binary classification average recall rate. It can solve the negative impact of the unbalanced training data set obtained based on the real operating HVAC equipment on the model training, so that the first fault detection model trained on the limited data set achieves better generalization performance.

[0056] The above-mentioned method of weighting the model prediction probability by the target weight corresponding to the equipment anomaly measurement value achieves the further improvement of the accuracy of the first fault detection model in detecting the fault type of HVAC equipment by focusing on the data scatter points with large reconstruction errors during anomaly detection.

[0057] S204, using a preset probability voting mechanism to perform fault type probability statistical processing on the multi-classification probability voting matrix to obtain a first fault classification result.

[0058] In some embodiments, the preset probability voting mechanism can select a total probability voting mechanism and a proportional voting mechanism. The method of using the total probability voting mechanism to obtain the first fault classification result is: for each reference fault type in the multi-classification probability voting matrix, calculate the sum of the predicted probabilities of all binary classification sub-models, select the reference fault type with the largest sum of predicted probabilities as the equipment fault type identified by the first fault detection model, and generate a first fault classification result including the equipment fault type. The method of using the proportional voting mechanism to obtain the first fault classification result is: for each device abnormal parameter reconstruction error, count the number of votes for each reference fault type determined by the device abnormal parameter reconstruction error in all binary classification sub-models, select the reference fault type with the largest number of votes as the equipment fault type identified by the first fault detection model, and generate a first fault classification result including the equipment fault type.

[0059] S205, performing a fault type probability prediction on the equipment abnormal parameter reconstruction error through a second fault detection model to obtain probability prediction results corresponding to at least two reference fault types.

[0060] It can be understood that the second fault detection model is obtained after model training based on the initial second fault detection model of the one-to-many classification architecture. Figure 4 A flow chart of a training method for a second fault detection model is shown, and the training method may include the following steps: S401, creating an initial second fault detection model for a HVAC equipment fault detection scenario based on a one-to-many classification architecture, the initial second fault detection model including at least one initial single-category classification sub-model; S402, obtaining a third sample abnormal reconstruction error, and marking a third fault type label on the third sample abnormal reconstruction error; S403, determining a third sample input feature corresponding to the third sample abnormal reconstruction error, determining a classification prediction contribution of the third input feature to the initial single-class classification sub-model, and determining a fourth sample input feature for the initial single-class classification sub-model from the third sample input feature based on the classification prediction contribution; S404, selecting a fourth sample abnormal reconstruction error from the third sample abnormal reconstruction error based on the fourth sample input feature, and obtaining a fourth fault type label marked by the fourth sample abnormal reconstruction error; S405, use the fourth sample abnormal reconstruction error to perform at least one round of model training on the initial second fault detection model. During the model training process, the fourth sample abnormal reconstruction error is subjected to fault type detection processing based on the initial second fault detection model to obtain a second predicted fault type, and based on the second predicted fault type and the fourth fault type label, the model loss function and the binary classification average recall rate indicator are used to adjust the model parameters of the initial second fault detection model until the model training end conditions are met to obtain the second fault detection model.

[0061] In step S401, possible reference fault types of HVAC equipment based on expert experience are obtained. Specifically, the number of reference fault types is N. A one-to-many classification architecture technology is used to create a second initial fault detection model based on N initial single-category classification sub-models. Each initial single-category classification sub-model is configured to identify reference fault types and non-reference fault types.

[0062] In step S402, the third sample abnormal reconstruction error may include sample device abnormal parameter reconstruction errors of multiple sample HVAC devices with abnormalities, and these test HVAC devices with abnormalities may include sample HVAC devices whose fault types are reference fault types. In the third sample abnormal reconstruction error, the sample device abnormal parameter reconstruction errors of the sample HVAC devices belonging to the reference fault type are annotated with corresponding reference fault type labels, and the third fault type label may include multiple reference fault type labels.

[0063] In step S403, all sample parameter features in the third sample abnormal reconstruction error are determined, and the sample parameter features are used as the third sample input features. For example, taking the HVAC equipment as a multi-split unit as an example, the third sample abnormal reconstruction error may include reconstruction errors of parameters such as subcooling, superheat, compressor suction temperature, compressor exhaust temperature, compressor frequency, and expansion valve opening. Subcooling, superheat, compressor suction temperature, compressor exhaust temperature, compressor frequency, and expansion valve opening are all sample parameter features. Therefore, subcooling, superheat, compressor suction temperature, compressor exhaust temperature, compressor frequency, and expansion valve opening can all be used as the third sample input feature. After determining the third sample input feature, the classification prediction contribution of the third sample input feature to the initial single-class classification sub-model can be determined by the Shapley value corresponding to the third sample input feature. The classification prediction contribution can be sorted in order from high to low, and the third sample input features corresponding to the classification prediction contribution of the preset number of the first arranged are used as the fourth sample input feature for the initial single-class classification sub-model.

[0064] Specifically, the implementation method of calculating the Shapley value corresponding to the third sample input feature is as follows: for each initial single-category classification sub-model, the sample reconstruction error corresponding to the fault type that the initial single-category classification sub-model can identify is taken out from the third sample abnormal reconstruction error, the sample reconstruction error is used to train the initial single-category classification sub-model to obtain a preliminarily trained initial single-category classification sub-model, and the contribution of each third sample input feature to the preliminarily trained initial single-category classification sub-model is calculated by a preset Shapley value calculation method to obtain a Shapley value matrix, the number of rows of the Shapley value matrix is ​​the same as the number of samples of the sample reconstruction error, the number of columns of the Shapley value matrix is ​​the same as the number of features of the sample reconstruction error, the absolute value processing is performed on the Shapley value matrix to obtain the contribution of each feature of each sample (sample reconstruction error) to the model prediction as a positive class, and further, the average of each column of the Shapley value matrix is ​​obtained to obtain the average contribution of each feature to the model prediction as a positive class with the support of the entire training set samples, and the average contribution is used as the Shapley value corresponding to each third sample input feature. The Shapley value can be used to represent the importance of the feature to the model.

[0065] It can be understood that customizing the input features of the single-category classification model one by one based on the Shapley value, rather than using the same set of all input features for all models, can further improve the diagnostic performance of the second fault detection model.

[0066] In step S404, a fourth sample abnormal reconstruction error corresponding to the fourth sample input feature is selected from the third sample abnormal reconstruction error, and the fourth sample abnormal reconstruction error is used as sample input data for subsequent training of the initial single-category classification sub-model.

[0067] In step S405, the model parameters of the initial second fault detection model are adjusted by using the model loss function and the binary classification average recall rate index based on the second predicted fault type and the fourth fault type label until the model training end condition is met to obtain the second fault detection model. Specifically, it may include: calculating the model loss value by using the model loss function based on the second predicted fault type and the fourth fault type label, adjusting the model parameters of at least one initial single-category classification sub-model based on the model loss value to obtain at least one reference single-category classification sub-model; determining a second reference hyperparameter set for the reference single-category classification sub-model from a second preset hyperparameter search domain, updating the hyperparameters of the reference single-category classification sub-model based on the second reference hyperparameter set, and calculating the reference binary classification sub-model for each updated reference single-category classification sub-model. Classification average recall rate; determine whether the second hyperparameter update times of the reference single-category classification sub-model is equal to the second preset times threshold; if the second hyperparameter update times are not equal to the second preset times threshold, determine the second updated hyperparameter set for the reference single-category classification sub-model from the second preset hyperparameter search domain, use the second updated hyperparameter set as the second reference hyperparameter set, and perform the step of updating the hyperparameters of the reference single-category classification sub-model based on the second reference hyperparameter set; if the second hyperparameter update times are equal to the second preset times threshold, determine the second target hyperparameter set for the reference single-category classification sub-model from the second preset hyperparameter search domain based on the reference two-category average recall rate, update the hyperparameters of the reference single-category classification sub-model based on the second target hyperparameter set, and obtain a second fault detection model.

[0068] Specifically, the second preset hyperparameter search domain includes at least one hyperparameter set corresponding to each reference single-category classification sub-model, and the hyperparameters in the hyperparameter set refer to parameters in the reference single-category classification sub-model that cannot be directly learned through data.

[0069] Specifically, the reference binary classification average recall rate is the average of the verified recall rates obtained by using the leave-one-out cross-validation model evaluation method. In the embodiment of the present application, the leave-one-out cross-validation model evaluation method sets aside the sample equipment abnormal parameter reconstruction error of a sample HVAC device as a validation set each time the model is trained, and the sample equipment abnormal parameter reconstruction errors of all remaining sample HVAC devices are used as training sets, and the process is repeated until the sample equipment abnormal parameter reconstruction errors of each sample HVAC device are used as a validation set. The reference binary classification average recall rate is calculated as follows: for each reference single-category classification sub-model, assuming that the data set is sample data of N sample HVAC devices (including sample equipment abnormal parameter reconstruction errors of N sample HVAC devices), each time the sample equipment abnormal parameter reconstruction error of one of the sample HVAC devices is used as a validation set, and the remaining The sample equipment abnormal parameter reconstruction error of each sample HVAC equipment is used as a training set, a second reference hyperparameter set corresponding to the reference single-class classification submodel is selected from the second preset hyperparameter search domain, and the second reference hyperparameter set is used to update the hyperparameters of the reference single-class classification submodel. The training set is used to train the reference single-class classification submodel after the hyperparameter update, and the validation set is used for validation, and the recall rate of this validation is calculated, which is the recall rate corresponding to the reference single-class classification submodel after the hyperparameter update using the second reference hyperparameter set. The training process of leaving one validation set out each time is repeated N times until the sample equipment abnormal parameter reconstruction error of each sample HVAC equipment is used as a validation set once, and the average recall rate obtained by each validation is calculated to obtain the reference binary classification average recall rate. The reference binary classification average recall rate corresponding to the fault type that can be identified by each reference single-class classification submodel can be obtained by the above-mentioned method of calculating the reference binary classification average recall rate. For a certain fault type, the recall rate is equal to the number of sample HVAC equipment whose fault type is correctly diagnosed divided by the total number of sample HVAC equipment belonging to the fault type.

[0070] Specifically, for each reference single-class classification sub-model, a second reference hyperparameter set corresponding to the maximum reference binary classification average recall rate is selected as the second target hyperparameter set, and the second target hyperparameter set is the optimal hyperparameter set of the reference single-class classification sub-model. The second target hyperparameter set is used to update the hyperparameters of the reference single-class classification sub-model to obtain a second fault detection model.

[0071] It is understandable that when performing hyperparameter optimization for the single-category classification submodel, the embodiment of the present application selects the complete time series data of a sample HVAC device as a test set, rather than using the conventional test set selection method in the related art. Compared with the conventional test set selection method, the embodiment of the present application selects the complete time series data of a sample HVAC device as a test set, which fully simulates the scenario of large-scale application on new HVAC equipment after the algorithm is trained, and adopts the test set selection method of the embodiment of the present application to perform the hyperparameter optimization process, so that the second fault detection model obtained by training has better robustness.

[0072] In step S205, the target model input features for at least one single-category classification sub-model in the second fault detection model are determined from the equipment abnormal parameter reconstruction error, and the target model input features are processed for fault type detection by at least one single-category classification sub-model to obtain probability prediction results corresponding to at least two reference fault types identified by all single-category classification sub-models.

[0073] S206: Obtain preset probability thresholds corresponding to at least two reference fault types.

[0074] S207: If each probability prediction result is less than the corresponding preset probability threshold, the second fault classification result is determined to be an unknown fault type.

[0075] S208: If there is at least one probability prediction result that is greater than or equal to the corresponding preset probability threshold, determine that the second fault classification result is a known fault type.

[0076] S206-S208 are explained below.

[0077] Specifically, the preset probability threshold corresponding to each reference fault type is used to detect whether the HVAC equipment belongs to the corresponding reference fault type. When the probability prediction result corresponding to a reference fault type identified by a single-category classification sub-model is greater than or equal to the corresponding preset probability threshold, it can be determined that the single-category classification sub-model identifies that the HVAC equipment belongs to the reference fault type. When the probability prediction result corresponding to a reference fault type identified by a single-category classification sub-model is less than the corresponding preset probability threshold, it can be determined that the single-category classification sub-model identifies that the HVAC equipment belongs to a non-reference fault type.

[0078] When the probability prediction results corresponding to each reference fault type are less than the preset probability threshold corresponding to the reference fault type, it can be determined that the recognition results of all single-category classification sub-models in the second fault detection type are non-reference fault types, and therefore the second fault classification result can be determined to be an unknown fault type. When there is at least one reference fault type corresponding to a probability prediction result greater than or equal to the preset probability threshold corresponding to the reference fault type, it can be determined that the recognition result of at least one single-category classification sub-model in the second fault detection type is a certain reference fault type, and therefore the second fault classification result can be determined to be a known fault type.

[0079] S209: If the second fault classification result is a known fault type, determine the equipment fault type from the first fault classification result, and determine the equipment fault type as the target fault type of the HVAC equipment.

[0080] S210: If the second fault classification result is an unknown fault type, perform equipment fault diagnosis on the HVAC equipment.

[0081] For details, see S209 and S210. Figure 1 The description of the relevant parts in the illustrated embodiment will not be repeated here.

[0082] In the fault detection method for HVAC equipment provided in an embodiment of the present application, equipment operation data of the HVAC equipment is obtained, and an equipment anomaly detection model is used to perform equipment anomaly detection processing on the equipment operation data to obtain equipment anomaly detection data, and an equipment anomaly detection result and an equipment operation parameter reconstruction error are determined from the equipment anomaly detection data. Based on the equipment anomaly detection result, an abnormal data extraction process is performed on the equipment operation parameter reconstruction error to obtain an equipment abnormal parameter reconstruction error, and a fault type probability prediction is performed on the equipment abnormal parameter reconstruction error through a first fault detection model to obtain a multi-classification probability voting matrix, and a preset probability voting mechanism is used to perform fault type probability statistical processing on the multi-classification probability voting matrix to obtain a first fault classification result, and a fault type probability prediction is performed on the equipment abnormal parameter reconstruction error through a second fault detection model to obtain probability prediction results corresponding to at least two reference fault types, and preset probability thresholds corresponding to at least two reference fault types are obtained. If each probability prediction result is less than the corresponding preset probability threshold, the second fault classification result is determined to be an unknown fault type, and if there is at least one probability prediction result greater than or equal to the corresponding preset probability threshold, the second fault classification result is determined to be a known fault type. Therefore, using the equipment abnormal parameter reconstruction error as the model input data is to use the incremental bias information of the data as the model input data, rather than the absolute information of the data itself, so that the model can achieve better fault type diagnosis performance. In the result of the first fault classification obtained by using the first fault detection model, the prediction probability of the fault type is weighted by the integrated weights corresponding to each sub-model in the first fault detection model and the target weight determined based on the abnormal measurement value, which can improve the accuracy of the first fault detection model in identifying the fault type. In addition, the second fault detection model trained by the one-to-many classification architecture technology can identify the unknown fault types of HVAC equipment. Therefore, the fault classification results of the two models are combined to accurately identify the fault type of the HVAC equipment detected to be abnormal.

[0083] See also Figure 5 , is a schematic diagram of a structure of a fault type detection device for HVAC equipment provided in an embodiment of the present application. It should be noted that, Figure 5 The fault type detection device of the HVAC equipment shown is used to implement the present application Figure 1~Figure 4 For the convenience of explanation, only the part related to the embodiment of the present application is shown. For the specific technical details not disclosed, please refer to the present application. Figure 1~Figure 4 The embodiment shown.

[0084] See also Figure 5, which shows a schematic diagram of the structure of the fault type detection device of the HVAC equipment in the embodiment of the present application. The fault type detection device 1 of the HVAC equipment can be implemented as all or part of the device through software, hardware or a combination of both. According to some embodiments, the fault type detection device 1 of the HVAC equipment includes a data extraction module 11, a model recognition module 12, a first diagnosis module 13 and a second diagnosis module 14, which are specifically used to: A data extraction module 11 is used to determine the device abnormality detection data corresponding to the HVAC device, and perform abnormal data extraction processing based on the device abnormality detection data to obtain a device abnormality parameter reconstruction error; A model identification module 12 is used to use a first fault detection model to perform fault type detection processing on the device abnormal parameter reconstruction error to obtain a first fault classification result, and use a second fault detection model to perform fault type detection processing on the device abnormal parameter reconstruction error to obtain a second fault classification result, wherein the first fault detection model is obtained after model training based on an initial first fault detection model of a one-to-one classification architecture, and the second fault detection model is obtained after model training based on an initial second fault detection model of a one-to-many classification architecture; A first diagnostic module 13, configured to determine a target fault type of the HVAC equipment based on the first fault classification result if the second fault classification result is a known fault type; The second diagnosis module 14 is configured to perform equipment fault diagnosis on the HVAC equipment if the second fault classification result is an unknown fault type.

[0085] Optionally, the model identification module 12 includes: A first identification unit is used to perform a fault type probability prediction on the abnormal parameter reconstruction error of the device through a first fault detection model to obtain a multi-classification probability voting matrix; The second identification unit is used to perform fault type probability statistical processing on the multi-classification probability voting matrix by adopting a preset probability voting mechanism to obtain a first fault classification result.

[0086] Optionally, the first identification unit is specifically used to: Determine a target model input feature for at least one binary classification sub-model in a first fault detection model from the equipment abnormal parameter reconstruction error, perform fault type detection processing on the target model input feature through the at least one binary classification sub-model and the model integration weight corresponding to the binary classification sub-model, and obtain an initial multi-classification probability voting matrix; Based on the device abnormality detection data, a device abnormality metric value corresponding to the device abnormality parameter reconstruction error is determined, a target weight corresponding to the device abnormality metric value is determined, and based on the target weight, probability weighted processing is performed on the initial multi-classification probability voting matrix to obtain a multi-classification probability voting matrix.

[0087] Optionally, the model identification module 12 is specifically used for: Performing a fault type probability prediction on the device abnormal parameter reconstruction error through a second fault detection model to obtain probability prediction results corresponding to at least two reference fault types; Obtaining preset probability thresholds corresponding to the at least two reference fault types; If each of the probability prediction results is less than the corresponding preset probability threshold, determining that the second fault classification result is an unknown fault type; If there is at least one of the probability prediction results that is greater than or equal to the corresponding preset probability threshold, the second fault classification result is determined to be a known fault type.

[0088] Optionally, the first diagnostic module 13 is specifically used for: If the second fault classification result is a known fault type, the equipment fault type is determined from the first fault classification result, and the equipment fault type is determined as the target fault type of the HVAC equipment.

[0089] Optionally, the data extraction module 11 is further used for: An equipment abnormality detection result and an equipment operation parameter reconstruction error are determined from the equipment abnormality detection data, and an abnormal data extraction process is performed on the equipment operation parameter reconstruction error based on the equipment abnormality detection result to obtain an equipment abnormality parameter reconstruction error.

[0090] Optionally, the fault type detection device 1 for HVAC equipment further includes: A first training unit of the first model is used to create an initial first fault detection model for a HVAC equipment fault detection scenario based on a one-to-one classification architecture, wherein the initial first fault detection model includes at least one initial binary classification sub-model; A second training unit of the first model is used to obtain a first sample abnormal reconstruction error and mark the first sample abnormal reconstruction error with a first fault type label; The third training unit of the first model is used to determine a first sample input feature corresponding to the first sample abnormal reconstruction error, determine a classification prediction contribution of the first sample input feature to the initial binary classification sub-model, and select a second sample input feature for the initial binary classification sub-model from the first sample input feature based on the classification prediction contribution; The fourth training unit of the first model is used to select a second sample abnormal reconstruction error from the first sample abnormal reconstruction error based on the second sample input feature, and obtain a second fault type label marked by the second sample abnormal reconstruction error; The fifth training unit of the first model is used to perform at least one round of model training on the initial first fault detection model using the second sample abnormal reconstruction error. During the model training process, fault type detection is performed on the second sample abnormal reconstruction error based on the initial first fault detection model to obtain a first predicted fault type. Based on the first predicted fault type and the second fault type label, the model loss function and the binary classification average recall rate indicator are used to adjust the model parameters of the initial first fault detection model until the model training end condition is met to obtain the first fault detection model.

[0091] Optionally, the fifth training unit of the first model is specifically used for: Calculate a model loss value using a model loss function based on the first predicted fault type and the second fault type label, and adjust model parameters of the at least one initial binary classification sub-model based on the model loss value to obtain at least one reference binary classification sub-model; Determine a first reference hyperparameter set for the reference binary classification sub-model from a first preset hyperparameter search domain, perform hyperparameter update on the reference binary classification sub-model based on the first reference hyperparameter set, and calculate a reference binary classification average recall rate for each updated reference binary classification sub-model; Determine whether the number of updates of a first hyperparameter of the reference binary classification sub-model is equal to a first preset number threshold; If the first hyperparameter update times are not equal to the first preset times threshold, determining a first updated hyperparameter set for the reference binary classification sub-model from the first preset hyperparameter search domain, using the first updated hyperparameter set as the first reference hyperparameter set, and performing the step of updating the hyperparameters of the reference binary classification sub-model based on the first reference hyperparameter set; If the first hyperparameter update number is equal to the first preset number threshold, then based on the reference binary classification average recall rate, a first target hyperparameter set for the reference binary classification sub-model is determined from the first preset hyperparameter search domain, and the reference binary classification sub-model is hyperparameter updated based on the first target hyperparameter set to obtain a first fault detection model.

[0092] Optionally, the fault type detection device 1 for HVAC equipment further includes: A first training unit of the second model is used to create an initial second fault detection model for a HVAC equipment fault detection scenario based on a one-to-many classification architecture, wherein the initial second fault detection model includes at least one initial single-class classification sub-model; A second training unit of the second model is used to obtain a third sample abnormal reconstruction error and mark the third sample abnormal reconstruction error with a third fault type label; The third training unit of the second model is used to determine a third sample input feature corresponding to the third sample abnormal reconstruction error, determine a classification prediction contribution of the third input feature to the initial single-class classification sub-model, and determine a fourth sample input feature for the initial single-class classification sub-model from the third sample input feature based on the classification prediction contribution; A fourth training unit of the second model is used to select a fourth sample abnormal reconstruction error from the third sample abnormal reconstruction error based on the second sample input feature, and obtain a fourth fault type label marked by the fourth sample abnormal reconstruction error; The fifth training unit of the second model is used to perform at least one round of model training on the initial second fault detection model using the fourth sample abnormal reconstruction error. During the model training process, the fourth sample abnormal reconstruction error is subjected to fault type detection processing based on the initial second fault detection model to obtain a second predicted fault type. Based on the second predicted fault type and the fourth fault type label, the model loss function and the binary classification average recall rate indicator are used to adjust the model parameters of the initial second fault detection model until the model training end condition is met to obtain the second fault detection model.

[0093] Optionally, the fifth training unit of the second model is specifically used for: Calculate a model loss value using a model loss function based on the second predicted fault type and the fourth fault type label, and adjust model parameters of the at least one initial single-category classification sub-model based on the model loss value to obtain at least one reference single-category classification sub-model; Determine a second reference hyperparameter set for a reference single-class classification submodel from a second preset hyperparameter search domain, perform hyperparameter updates on the reference single-class classification submodel based on the second reference hyperparameter set, and calculate a reference binary classification average recall rate for each updated reference single-class classification submodel; Determine whether the number of times a second hyperparameter of the reference single-category classification sub-model is updated is equal to a second preset number threshold; If the second hyperparameter update times are not equal to the second preset times threshold, determining a second updated hyperparameter set for the reference single-class classification sub-model from the second preset hyperparameter search domain, using the second updated hyperparameter set as the second reference hyperparameter set, and performing the step of performing hyperparameter updating on the reference single-class classification sub-model based on the second reference hyperparameter set; If the second hyperparameter update number is equal to the second preset number threshold, then based on the reference binary classification average recall rate, a second target hyperparameter set for the reference single-category classification sub-model is determined from the second preset hyperparameter search domain, and the reference single-category classification sub-model is hyperparameter updated based on the second target hyperparameter set to obtain a second fault detection model.

[0094] Please refer to Figure 6 , Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Exemplarily, the electronic device in the embodiment of the present application may include one or more of the following components: a processor 110, a memory 120, an input device 130, an output device 140, and a bus 150. The processor 110, the memory 120, the input device 130, and the output device 140 may be connected via the bus 150.

[0095] The processor 110 may include one or more processing cores. The processor 110 uses various interfaces and lines to connect various parts of the entire electronic device, and executes various functions of the electronic device and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 120, and calling data stored in the memory 120. Optionally, the processor 110 can be implemented in at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), and programmable logic array (PLA). The processor 110 can integrate one or a combination of a central processing unit (CPU), a graphics processing unit (GPU), and a modem. Among them, the CPU mainly processes the operating system, user interface, and application programs; the GPU is responsible for rendering and drawing display content; and the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 110, but may be implemented separately through a communication chip.

[0096] The memory 120 may include a random access memory (RAM) or a read-only memory (ROM). Optionally, the memory 120 includes a non-transitory computer-readable storage medium. The memory 120 may be used to store instructions, programs, codes, code sets or instruction sets. The memory 120 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for implementing at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the following various method embodiments, etc. The operating system may be an Android system, including a system deeply developed based on the Android system, an iOS system developed by Apple, including a system deeply developed based on the iOS system, or other systems.

[0097] In order for the operating system to distinguish the specific application scenarios of third-party applications, it is necessary to open up data communication between third-party applications and the operating system so that the operating system can obtain the current scenario information of third-party applications at any time, and then perform targeted system resource adaptation based on the current scenario.

[0098] The input device 130 is used to receive input commands or data, and includes but is not limited to a keyboard, a mouse, a camera, a microphone, or a touch device. The output device 140 is used to output commands or data, and includes but is not limited to a display device and a speaker. In one example, the input device 130 and the output device 140 can be combined, and the input device 130 and the output device 140 are touch screen displays.

[0099] The touch display screen can be designed as a full screen, a curved screen or a special-shaped screen. The touch display screen can also be designed as a combination of a full screen and a curved screen, or a combination of a special-shaped screen and a curved screen, which is not limited in the embodiments of the present application.

[0100] In addition, those skilled in the art can understand that the structure of the electronic device shown in the above drawings does not constitute a limitation on the electronic device, and the electronic device may include more or fewer components than shown in the drawings, or combine certain components, or arrange the components differently. For example, the electronic device also includes radio frequency circuits, input units, sensors, audio circuits, wireless fidelity (WiFi) modules, power supplies, Bluetooth modules and other components, which will not be described in detail here.

[0101] exist Figure 6In the electronic device shown, the processor 110 can be used to call the program of the fault type detection method of the HVAC equipment stored in the memory 120, and specifically perform the following operations: Determine the device abnormality detection data corresponding to the HVAC device, and perform abnormal data extraction processing based on the device abnormality detection data to obtain a device abnormality parameter reconstruction error; A first fault detection model is used to perform fault type detection processing on the equipment abnormal parameter reconstruction error to obtain a first fault classification result, and a second fault detection model is used to perform fault type detection processing on the equipment abnormal parameter reconstruction error to obtain a second fault classification result, wherein the first fault detection model is obtained after model training based on an initial first fault detection model of a one-to-one classification architecture, and the second fault detection model is obtained after model training based on an initial second fault detection model of a one-to-many classification architecture; If the second fault classification result is a known fault type, determining a target fault type of the HVAC equipment based on the first fault classification result; If the second fault classification result is an unknown fault type, equipment fault diagnosis is performed on the HVAC equipment.

[0102] In one embodiment, when the processor 110 executes the step of using the first fault detection model to perform fault type detection processing on the device abnormal parameter reconstruction error to obtain the first fault classification result, the following operations are specifically performed: Performing a fault type probability prediction on the abnormal parameter reconstruction error of the equipment through a first fault detection model to obtain a multi-classification probability voting matrix; A preset probability voting mechanism is used to perform fault type probability statistical processing on the multi-classification probability voting matrix to obtain a first fault classification result.

[0103] In one embodiment, when the processor 110 performs the step of predicting the probability of the fault type of the device abnormal parameter reconstruction error by using the first fault detection model to obtain a multi-classification probability voting matrix, the following operations are specifically performed: Determine a target model input feature for at least one binary classification sub-model in a first fault detection model from the equipment abnormal parameter reconstruction error, perform fault type detection processing on the target model input feature through the at least one binary classification sub-model and the model integration weight corresponding to the binary classification sub-model, and obtain an initial multi-classification probability voting matrix; Based on the device abnormality detection data, a device abnormality metric value corresponding to the device abnormality parameter reconstruction error is determined, a target weight corresponding to the device abnormality metric value is determined, and based on the target weight, probability weighted processing is performed on the initial multi-classification probability voting matrix to obtain a multi-classification probability voting matrix.

[0104] In one embodiment, when the processor 110 executes the step of using the second fault detection model to perform fault type detection processing on the device abnormal parameter reconstruction error to obtain the second fault classification result, the following operations are specifically performed: Performing a fault type probability prediction on the device abnormal parameter reconstruction error through a second fault detection model to obtain probability prediction results corresponding to at least two reference fault types; Obtaining preset probability thresholds corresponding to the at least two reference fault types; If each of the probability prediction results is less than the corresponding preset probability threshold, determining that the second fault classification result is an unknown fault type; If there is at least one of the probability prediction results that is greater than or equal to the corresponding preset probability threshold, the second fault classification result is determined to be a known fault type.

[0105] In one embodiment, when the processor 110 executes the step of determining the target fault type of the HVAC equipment based on the first fault classification result if the second fault classification result is a known fault type, the processor 110 specifically performs the following operations: If the second fault classification result is a known fault type, the equipment fault type is determined from the first fault classification result, and the equipment fault type is determined as the target fault type of the HVAC equipment.

[0106] In one embodiment, when the processor 110 performs the step of performing abnormal data extraction processing based on the device abnormality detection data to obtain a device abnormality parameter reconstruction error, the following operations are specifically performed: An equipment abnormality detection result and an equipment operation parameter reconstruction error are determined from the equipment abnormality detection data, and an abnormal data extraction process is performed on the equipment operation parameter reconstruction error based on the equipment abnormality detection result to obtain an equipment abnormality parameter reconstruction error.

[0107] In one embodiment, the processor 110 further performs the following operations: Creating an initial first fault detection model for a HVAC equipment fault detection scenario based on a one-to-one classification architecture, wherein the initial first fault detection model includes at least one initial two-classification sub-model; Obtaining a first sample abnormal reconstruction error, and marking the first sample abnormal reconstruction error with a first fault type label; Determine a first sample input feature corresponding to the first sample abnormal reconstruction error, determine a classification prediction contribution of the first sample input feature to the initial binary classification sub-model, and select a second sample input feature for the initial binary classification sub-model from the first sample input feature based on the classification prediction contribution; Selecting a second sample abnormal reconstruction error from the first sample abnormal reconstruction error based on the second sample input feature, and obtaining a second fault type label marked by the second sample abnormal reconstruction error; The second sample abnormal reconstruction error is used to perform at least one round of model training on the initial first fault detection model. During the model training process, the second sample abnormal reconstruction error is subjected to fault type detection processing based on the initial first fault detection model to obtain a first predicted fault type. Based on the first predicted fault type and the second fault type label, the model loss function and the binary classification average recall rate indicator are used to adjust the model parameters of the initial first fault detection model until the model training end conditions are met to obtain the first fault detection model.

[0108] In one embodiment, when the processor 110 performs the step of adjusting the model parameters of the initial first fault detection model based on the first predicted fault type and the second fault type label using the model loss function and the binary classification average recall rate index until the model training end condition is met to obtain the first fault detection model, the following operations are specifically performed: Calculate a model loss value using a model loss function based on the first predicted fault type and the second fault type label, and adjust model parameters of the at least one initial binary classification sub-model based on the model loss value to obtain at least one reference binary classification sub-model; Determine a first reference hyperparameter set for the reference binary classification sub-model from a first preset hyperparameter search domain, perform hyperparameter update on the reference binary classification sub-model based on the first reference hyperparameter set, and calculate a reference binary classification average recall rate for each updated reference binary classification sub-model; Determine whether the number of updates of a first hyperparameter of the reference binary classification sub-model is equal to a first preset number threshold; If the first hyperparameter update times are not equal to the first preset times threshold, determining a first updated hyperparameter set for the reference binary classification sub-model from the first preset hyperparameter search domain, using the first updated hyperparameter set as the first reference hyperparameter set, and performing the step of updating the hyperparameters of the reference binary classification sub-model based on the first reference hyperparameter set; If the first hyperparameter update number is equal to the first preset number threshold, then based on the reference binary classification average recall rate, a first target hyperparameter set for the reference binary classification sub-model is determined from the first preset hyperparameter search domain, and the reference binary classification sub-model is hyperparameter updated based on the first target hyperparameter set to obtain a first fault detection model.

[0109] In one embodiment, the processor 110 further performs the following operations: Creating an initial second fault detection model for a HVAC equipment fault detection scenario based on a one-to-many classification architecture, wherein the initial second fault detection model includes at least one initial single-category classification sub-model; Obtaining a third sample abnormal reconstruction error, and marking the third sample abnormal reconstruction error with a third fault type label; Determine a third sample input feature corresponding to the third sample abnormal reconstruction error, determine a classification prediction contribution of the third input feature to the initial single-class classification sub-model, and determine a fourth sample input feature for the initial single-class classification sub-model from the third sample input feature based on the classification prediction contribution; Selecting a fourth sample abnormal reconstruction error from the third sample abnormal reconstruction error based on the second sample input feature, and obtaining a fourth fault type label marked by the fourth sample abnormal reconstruction error; The fourth sample abnormal reconstruction error is used to perform at least one round of model training on the initial second fault detection model. During the model training process, the fourth sample abnormal reconstruction error is subjected to fault type detection processing based on the initial second fault detection model to obtain a second predicted fault type. Based on the second predicted fault type and the fourth fault type label, the model loss function and the binary classification average recall rate indicator are used to adjust the model parameters of the initial second fault detection model until the model training end conditions are met to obtain the second fault detection model.

[0110] In one embodiment, when the processor 110 performs the step of adjusting the model parameters of the initial second fault detection model based on the second predicted fault type and the fourth fault type label using the model loss function and the binary classification average recall rate index until the model training end condition is met to obtain the second fault detection model, the following operations are specifically performed: Calculate a model loss value using a model loss function based on the second predicted fault type and the fourth fault type label, and adjust model parameters of the at least one initial single-category classification sub-model based on the model loss value to obtain at least one reference single-category classification sub-model; Determine a second reference hyperparameter set for a reference single-class classification submodel from a second preset hyperparameter search domain, perform hyperparameter updates on the reference single-class classification submodel based on the second reference hyperparameter set, and calculate a reference binary classification average recall rate for each updated reference single-class classification submodel; Determine whether the number of times a second hyperparameter of the reference single-category classification sub-model is updated is equal to a second preset number threshold; If the second hyperparameter update times are not equal to the second preset times threshold, determining a second updated hyperparameter set for the reference single-class classification sub-model from the second preset hyperparameter search domain, using the second updated hyperparameter set as the second reference hyperparameter set, and performing the step of performing hyperparameter updating on the reference single-class classification sub-model based on the second reference hyperparameter set; If the second hyperparameter update number is equal to the second preset number threshold, then based on the reference binary classification average recall rate, a second target hyperparameter set for the reference single-category classification sub-model is determined from the second preset hyperparameter search domain, and the reference single-category classification sub-model is hyperparameter updated based on the second target hyperparameter set to obtain a second fault detection model.

[0111] In one embodiment, when the processor 110 executes the step of determining the device abnormality detection data corresponding to the HVAC device, the processor 110 specifically performs the following operations: The equipment operation data of the HVAC equipment is obtained, and the equipment anomaly detection model is used to perform equipment anomaly detection processing on the equipment operation data to obtain equipment anomaly detection data.

[0112] In addition, those skilled in the art can understand that the structure of the electronic device shown in the above drawings does not constitute a limitation on the electronic device, and the washing device may include more or fewer components than shown, or combine certain components, or arrange the components differently. For example, the electronic device also includes components such as radio frequency circuits, audio circuits, WiFi components, power supplies, and Bluetooth components, which will not be described in detail here.

[0113] An embodiment of the present application further provides a computer-readable storage medium storing at least one instruction, wherein the at least one instruction is used to be executed by a processor to implement the method for detecting the fault type of HVAC equipment as described in the above embodiments.

[0114] An embodiment of the present application further provides a computer program product, which stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the fault type detection method of the HVAC equipment as described in the above embodiments.

[0115] Those skilled in the art should be aware that in one or more of the above examples, the functions described in the embodiments of the present application can be implemented with hardware, software, firmware, or any combination thereof. When implemented using software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein the communication media include any media that facilitates the transmission of a computer program from one place to another. The storage medium can be any available medium that a general or special-purpose computer can access.

[0116] The above description is only an optional embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for detecting the fault type of HVAC equipment, characterized in that: The method comprises: Determine the device abnormality detection data corresponding to the HVAC device, and perform abnormal data extraction processing based on the device abnormality detection data to obtain a device abnormality parameter reconstruction error; A first fault detection model is used to perform fault type detection processing on the equipment abnormal parameter reconstruction error to obtain a first fault classification result, and a second fault detection model is used to perform fault type detection processing on the equipment abnormal parameter reconstruction error to obtain a second fault classification result, wherein the first fault detection model is obtained after model training based on an initial first fault detection model of a one-to-one classification architecture, and the second fault detection model is obtained after model training based on an initial second fault detection model of a one-to-many classification architecture; If the second fault classification result is a known fault type, determining a target fault type of the HVAC equipment based on the first fault classification result; If the second fault classification result is an unknown fault type, equipment fault diagnosis is performed on the HVAC equipment.

2. The method according to claim 1, characterized in that The adopting the first fault detection model to perform fault type detection processing on the equipment abnormal parameter reconstruction error to obtain a first fault classification result includes: Performing a fault type probability prediction on the abnormal parameter reconstruction error of the equipment through a first fault detection model to obtain a multi-classification probability voting matrix; A preset probability voting mechanism is used to perform fault type probability statistical processing on the multi-classification probability voting matrix to obtain a first fault classification result.

3. The method according to claim 2, characterized in that The method of performing a fault type probability prediction on the device abnormal parameter reconstruction error by using the first fault detection model to obtain a multi-classification probability voting matrix includes: Determine a target model input feature for at least one binary classification sub-model in a first fault detection model from the equipment abnormal parameter reconstruction error, perform fault type detection processing on the target model input feature through the at least one binary classification sub-model and the model integration weight corresponding to the binary classification sub-model, and obtain an initial multi-classification probability voting matrix; Based on the device abnormality detection data, a device abnormality metric value corresponding to the device abnormality parameter reconstruction error is determined, a target weight corresponding to the device abnormality metric value is determined, and based on the target weight, probability weighted processing is performed on the initial multi-classification probability voting matrix to obtain a multi-classification probability voting matrix.

4. The method according to claim 1, characterized in that: The method of using the second fault detection model to perform fault type detection processing on the device abnormal parameter reconstruction error to obtain a second fault classification result includes: Performing a fault type probability prediction on the device abnormal parameter reconstruction error through a second fault detection model to obtain probability prediction results corresponding to at least two reference fault types; Obtaining preset probability thresholds corresponding to the at least two reference fault types; If each of the probability prediction results is less than the corresponding preset probability threshold, determining that the second fault classification result is an unknown fault type; If there is at least one of the probability prediction results that is greater than or equal to the corresponding preset probability threshold, the second fault classification result is determined to be a known fault type.

5. The method according to claim 1, characterized in that If the second fault classification result is a known fault type, determining the target fault type of the HVAC equipment based on the first fault classification result includes: If the second fault classification result is a known fault type, the equipment fault type is determined from the first fault classification result, and the equipment fault type is determined as the target fault type of the HVAC equipment.

6. The method according to claim 1, characterized in that Performing abnormal data extraction processing based on the device abnormality detection data to obtain a device abnormality parameter reconstruction error includes: An equipment abnormality detection result and an equipment operation parameter reconstruction error are determined from the equipment abnormality detection data, and an abnormal data extraction process is performed on the equipment operation parameter reconstruction error based on the equipment abnormality detection result to obtain an equipment abnormality parameter reconstruction error.

7. The method according to claim 1, characterized in that The method further comprises: Creating an initial first fault detection model for a HVAC equipment fault detection scenario based on a one-to-one classification architecture, wherein the initial first fault detection model includes at least one initial two-classification sub-model; Obtaining a first sample abnormal reconstruction error, and marking the first sample abnormal reconstruction error with a first fault type label; Determine a first sample input feature corresponding to the first sample abnormal reconstruction error, determine a classification prediction contribution of the first sample input feature to the initial binary classification sub-model, and select a second sample input feature for the initial binary classification sub-model from the first sample input feature based on the classification prediction contribution; Selecting a second sample abnormal reconstruction error from the first sample abnormal reconstruction error based on the second sample input feature, and obtaining a second fault type label marked by the second sample abnormal reconstruction error; The second sample abnormal reconstruction error is used to perform at least one round of model training on the initial first fault detection model. During the model training process, the second sample abnormal reconstruction error is subjected to fault type detection processing based on the initial first fault detection model to obtain a first predicted fault type. Based on the first predicted fault type and the second fault type label, the model loss function and the binary classification average recall rate indicator are used to adjust the model parameters of the initial first fault detection model until the model training end conditions are met to obtain the first fault detection model.

8. The method according to claim 7, characterized in that The method of adjusting the model parameters of the initial first fault detection model by using a model loss function and a binary classification average recall rate index based on the first predicted fault type and the second fault type label until a model training end condition is met to obtain a first fault detection model includes: Calculate a model loss value using a model loss function based on the first predicted fault type and the second fault type label, and adjust model parameters of the at least one initial binary classification sub-model based on the model loss value to obtain at least one reference binary classification sub-model; Determine a first reference hyperparameter set for the reference binary classification sub-model from a first preset hyperparameter search domain, perform hyperparameter update on the reference binary classification sub-model based on the first reference hyperparameter set, and calculate a reference binary classification average recall rate for each updated reference binary classification sub-model; Determine whether the number of updates of a first hyperparameter of the reference binary classification sub-model is equal to a first preset number threshold; If the first hyperparameter update times are not equal to the first preset times threshold, determining a first updated hyperparameter set for the reference binary classification sub-model from the first preset hyperparameter search domain, using the first updated hyperparameter set as the first reference hyperparameter set, and performing the step of updating the hyperparameters of the reference binary classification sub-model based on the first reference hyperparameter set; If the first hyperparameter update number is equal to the first preset number threshold, then based on the reference binary classification average recall rate, a first target hyperparameter set for the reference binary classification sub-model is determined from the first preset hyperparameter search domain, and the reference binary classification sub-model is hyperparameter updated based on the first target hyperparameter set to obtain a first fault detection model.

9. The method according to claim 1, characterized in that: The method further comprises: Creating an initial second fault detection model for a HVAC equipment fault detection scenario based on a one-to-many classification architecture, wherein the initial second fault detection model includes at least one initial single-category classification sub-model; Obtaining a third sample abnormal reconstruction error, and marking the third sample abnormal reconstruction error with a third fault type label; Determine a third sample input feature corresponding to the third sample abnormal reconstruction error, determine a classification prediction contribution of the third input feature to the initial single-class classification sub-model, and determine a fourth sample input feature for the initial single-class classification sub-model from the third sample input feature based on the classification prediction contribution; Selecting a fourth sample abnormal reconstruction error from the third sample abnormal reconstruction error based on the fourth sample input feature, and obtaining a fourth fault type label marked by the fourth sample abnormal reconstruction error; The fourth sample abnormal reconstruction error is used to perform at least one round of model training on the initial second fault detection model. During the model training process, the fourth sample abnormal reconstruction error is subjected to fault type detection processing based on the initial second fault detection model to obtain a second predicted fault type. Based on the second predicted fault type and the fourth fault type label, the model loss function and the binary classification average recall rate indicator are used to adjust the model parameters of the initial second fault detection model until the model training end conditions are met to obtain the second fault detection model.

10. The method according to claim 9, characterized in that The method of adjusting the model parameters of the initial second fault detection model by using a model loss function and a binary classification average recall rate index based on the second predicted fault type and the fourth fault type label until a model training end condition is met to obtain a second fault detection model includes: Calculate a model loss value using a model loss function based on the second predicted fault type and the fourth fault type label, and adjust model parameters of the at least one initial single-category classification sub-model based on the model loss value to obtain at least one reference single-category classification sub-model; Determine a second reference hyperparameter set for a reference single-class classification submodel from a second preset hyperparameter search domain, perform hyperparameter updates on the reference single-class classification submodel based on the second reference hyperparameter set, and calculate a reference binary classification average recall rate for each updated reference single-class classification submodel; Determine whether the number of times a second hyperparameter of the reference single-category classification sub-model is updated is equal to a second preset number threshold; If the second hyperparameter update times are not equal to the second preset times threshold, determining a second updated hyperparameter set for the reference single-class classification sub-model from the second preset hyperparameter search domain, using the second updated hyperparameter set as the second reference hyperparameter set, and performing the step of updating the hyperparameters of the reference single-class classification sub-model based on the second reference hyperparameter set; If the second hyperparameter update number is equal to the second preset number threshold, then based on the reference binary classification average recall rate, a second target hyperparameter set for the reference single-category classification sub-model is determined from the second preset hyperparameter search domain, and the reference single-category classification sub-model is hyperparameter updated based on the second target hyperparameter set to obtain a second fault detection model.

11. The method according to claim 1, characterized in that: The determining of the equipment abnormality detection data corresponding to the HVAC equipment includes: The equipment operation data of the HVAC equipment is obtained, and the equipment anomaly detection model is used to perform equipment anomaly detection processing on the equipment operation data to obtain equipment anomaly detection data.

12. A fault type detection device for HVAC equipment, characterized in that: The device comprises: A data extraction module is used to determine the device abnormality detection data corresponding to the HVAC equipment, and perform abnormal data extraction processing based on the device abnormality detection data to obtain the device abnormality parameter reconstruction error; A model identification module, used to use a first fault detection model to perform fault type detection processing on the equipment abnormal parameter reconstruction error to obtain a first fault classification result, and use a second fault detection model to perform fault type detection processing on the equipment abnormal parameter reconstruction error to obtain a second fault classification result, wherein the first fault detection model is obtained after model training based on an initial first fault detection model of a one-to-one classification architecture, and the second fault detection model is obtained after model training based on an initial second fault detection model of a one-to-many classification architecture; a first diagnostic module, configured to determine a target fault type of the HVAC equipment based on the first fault classification result if the second fault classification result is a known fault type; The second diagnostic module is used to perform equipment fault diagnosis on the HVAC equipment if the second fault classification result is an unknown fault type.

13. A computer storage medium, characterized in that: The computer storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor and executing the method according to any one of claims 1 to 11.

14. An electronic device, characterized in that: include: A processor and a memory; wherein the memory stores a computer program, and the computer program is suitable for being loaded by the processor and executing the method according to any one of claims 1 to 11.

Images