Fault diagnosis method and device of multi-split system, storage medium and electronic device

Abstract

The invention provides a fault diagnosis method and device of a multi-split system, a storage medium and an electronic device, and relates to the field of smart home. When fault diagnosis is carried out, a preset fault diagnosis model is used for realizing, and the preset fault diagnosis model comprises a self-adaptive multi-head attention layer, a convolution layer, a full connection layer, a normalization layer and an output layer. In the adaptive multi-head attention layer, the weights of different attention heads are obtained through dynamic learning, so that the adaptive multi-head attention layer can better capture key information in input data, and the accuracy of fault diagnosis is improved. Furthermore, a convolutional layer is configured after the adaptive multi-head attention layer, so that the local features of the data can be better extracted, and the accuracy of fault diagnosis is further improved. Furthermore, in the invention, the influence of the parameters irrelevant to the fault on the fault identification is eliminated, so that the target parameters obtained by screening can better reflect the fault information, and the accuracy of fault diagnosis is further improved.

Description

Technical Field

[0001] The present invention relates to the field of smart home, and more specifically, to a fault diagnosis method, device, storage medium and electronic device for a multi-connected air conditioner system. Background Art

[0002] Due to the characteristics of high efficiency and energy saving, multi-connected air conditioner systems are widely used in large commercial buildings, office buildings and residential places. However, as the complexity of multi-connected air conditioner systems increases, the complexity of fault detection also increases.

[0003] Traditional fault detection methods for multi-connected air conditioner systems often rely on fixed thresholds or rules. However, faults in multi-connected air conditioner systems are complex non-linear problems. Generally, the thresholds or rules set manually are limited by experience, and the setting accuracy is low. Problems such as failure to detect a fault when it occurs and incorrect fault detection often occur. Therefore, how to improve the accuracy of fault detection is a technical problem that needs to be solved urgently by those skilled in the art. Summary of the Invention

[0004] In view of this, the present invention provides a fault diagnosis method, device, storage medium and electronic device for a multi-connected air conditioner system to solve the problem of urgently improving the accuracy of fault detection.

[0005] To solve the above technical problems, the present invention adopts the following technical solutions:

[0006] A fault diagnosis method for a multi-connected air conditioner system includes:

[0007] Obtaining the parameter values of target parameters in the multi-connected air conditioner system; the target parameters include: the sensor aging time, and reference parameters screened from multiple parameters and satisfying the correlation condition;

[0008] Configuring the position coding identifier of the parameter values of the target parameters;

[0009] Taking the parameter values of the target parameters configured with the position coding identifier as input data and inputting them into a preset fault diagnosis model to obtain a fault diagnosis result; the preset fault diagnosis model includes an adaptive multi-head attention layer, a convolutional layer, a fully connected layer, a normalization layer and an output layer. In the preset fault diagnosis model, the input data sequentially passes through the adaptive multi-head attention layer, the convolutional layer, and the fully connected layer and then performs a normalization operation with the input data in the normalization layer. The data obtained from the normalization operation is used as the input of the output layer, and the output of the output layer is the fault diagnosis result; in the adaptive multi-head attention layer, the weights of different attention heads are obtained through dynamic learning.

[0010] Optionally, the adaptive multi-head attention layer includes:

[0011] A data processing module, a splicing layer, an activation function layer, and a linear transformation layer connected in sequence;

[0012] The data processing module includes a linear transformation sub-module, a matrix multiplication sub-module, a matrix transformation sub-module, and an activation layer sub-module. In the data processing module, among the three input sub-data obtained by splitting the input data, after two input sub-data pass through the linear transformation sub-module, the matrix multiplication sub-module, the matrix transformation sub-module, and the activation function module in sequence, a matrix multiplication operation is performed with the linear transformation result of the other input sub-data.

[0013] Optionally, the calculation formula configured in the activation function layer is:

[0014] , where is the output of multiple attention heads, is the total number of attention heads, is the intermediate weight matrix, is the weight matrix of the attention head;

[0015] The calculation formula configured in the linear transformation layer is:

[0016]

[0017] where is the output of the attention head , ranges from 1 to , is the weight of the attention head , obtained from , is the weighted output result.

[0018] Optionally, the determination process of the reference parameter includes:

[0019] Obtain multiple parameters and multiple preset fault types;

[0020] Calculate the correlation between each preset fault type and each parameter;

[0021] Screen out the reference parameters whose correlation with any preset fault type is greater than the correlation threshold.

[0022] Optionally, obtaining the parameter value of the target parameter in the multi-connected air-conditioning system includes:

[0023] Obtain the data uploaded by the configured data acquisition module in the multi-connected system according to the data upload period; the data acquisition module includes a signal sampling module, a filter circuit, a processor, and a communication module; the sensor data collected by the signal sampling module is sequentially processed by the filter circuit and the noise reduction algorithm in the processor, and then uploaded through the communication module;

[0024] Screen the parameter values of the reference parameters from the data;

[0025] Calculate the sensor aging time based on the time tag of the data uploaded by the data acquisition module or the time tag of receiving the data;

[0026] Use the parameter values of the reference parameters and the sensor aging time as the parameter values of the target parameters.

[0027] Optionally, configure the position encoding identifier of the parameter value of the target parameter, including:

[0028] Perform a position encoding operation on the parameter value of the target parameter to obtain the position encoding identifier of the parameter value of the target parameter.

[0029] Optionally, the training process of the preset fault diagnosis model includes:

[0030] Obtain data samples; the data samples include the sample values of the target parameters and the fault identifiers of the target parameters;

[0031] Use the data samples to train the fault diagnosis model until the model training stop condition is met and then stop;

[0032] Use the confusion matrix to evaluate the performance of the trained fault diagnosis model. If the performance meets the performance requirements, use the trained fault diagnosis model as the preset fault diagnosis model; the confusion matrix includes accuracy, recall, and F1 score.

[0033] A fault diagnosis device for a multi-connected system, including:

[0034] A data acquisition module for obtaining the parameter values of the target parameters in the multi-connected system; the target parameters include: sensor aging time, and reference parameters whose correlation meets the correlation condition screened from multiple parameters;

[0035] A position encoding module for configuring the position encoding identifier of the parameter value of the target parameter;

[0036] A fault diagnosis module, configured to input the parameter value of a target parameter configured with a position coding identifier as input data into a preset fault diagnosis model to obtain a fault diagnosis result; the preset fault diagnosis model includes an adaptive multi-head attention layer, a convolutional layer, a fully connected layer, a normalization layer, and an output layer. In the preset fault diagnosis model, the input data sequentially passes through the adaptive multi-head attention layer, the convolutional layer, and the fully connected layer, and then performs a normalization operation with the input data in the normalization layer. The data obtained from the normalization operation is used as the input of the output layer, and the output of the output layer is the fault diagnosis result; in the adaptive multi-head attention layer, the weights of different attention heads are obtained through dynamic learning.

[0037] A computer-readable storage medium, the computer-readable storage medium includes a stored program, wherein the program, when running, executes the fault diagnosis method of the above-mentioned multi-connected air-conditioning system.

[0038] An electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to execute the fault diagnosis method of the above-mentioned multi-connected air-conditioning system through the computer program.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention provides a fault diagnosis method, device, storage medium, and electronic device for a multi-connected air-conditioning system. In the present invention, when performing fault diagnosis, it is implemented using a preset fault diagnosis model. The preset fault diagnosis model includes an adaptive multi-head attention layer, a convolutional layer, a fully connected layer, a normalization layer, and an output layer. In the adaptive multi-head attention layer, the weights of different attention heads are obtained through dynamic learning, enabling the adaptive multi-head attention layer to better capture key information in the input data and improve the accuracy of fault diagnosis. Further, a convolutional layer is configured after the adaptive multi-head attention layer, which can better extract local features of the data and further improve the accuracy of fault diagnosis. Further, in the present invention, when obtaining parameters, the target parameters include the sensor aging time and reference parameters whose correlation satisfies the correlation condition selected from multiple parameters, eliminating the influence of parameters irrelevant to the fault on fault identification, so that the selected target parameters can better reflect the fault information and further improve the accuracy of fault diagnosis. Description of the Drawings

[0041] The drawings here are incorporated into the specification and form a part of this specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.

[0042] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0043] Figure 1 is a schematic diagram of the hardware environment of an interaction method for an intelligent device provided by an embodiment of the present invention;

[0044] Figure 2 is a flowchart of a fault diagnosis method for a multi-connected air-conditioning system provided by an embodiment of the present invention;

[0045] Figure 3 is a flowchart of a method for determining reference parameters provided by an embodiment of the present invention;

[0046] Figure 4 is a flowchart of a method for calculating parameter values provided by an embodiment of the present invention;

[0047] Figure 5 is a structural diagram of a preset fault diagnosis model provided by an embodiment of the present invention;

[0048] Figure 6 is a structural diagram of an adaptive multi-head attention layer provided by an embodiment of the present invention;

[0049] Figure 7 is a flowchart of another fault diagnosis method for a multi-connected air-conditioning system provided by an embodiment of the present invention;

[0050] Figure 8 is a schematic structural diagram of a fault diagnosis device for a multi-connected air-conditioning system provided by an embodiment of the present invention. Detailed implementation manners

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0052] Due to the characteristics of high efficiency and energy saving, multi-connected air-conditioning systems are widely used in large commercial buildings, office buildings, and residential places. However, as the complexity of multi-connected air-conditioning systems increases, the complexity of fault detection also increases.

[0053] Traditional fault detection methods for multi-connected air-conditioning systems often rely on fixed thresholds or rules. However, faults in multi-connected air-conditioning systems are complex non-linear problems. Generally, the thresholds or rules set manually are limited by experience and have low setting accuracy, often resulting in problems such as failure to detect a fault when it occurs and misdetection of faults. For example, when there are multiple fault situations in a multi-connected air-conditioning system, such as sensor failure, four-way valve fault, and condenser blockage, conventional fault detection methods often cannot accurately determine the types or locations of multiple faults, and can only target one type of fault, and may even generate false alarms or missed detections. Another example is that for some faults, such as refrigerant reduction, it can only be successfully judged under severe fault conditions, at which time the system performance has been severely affected and losses have been caused.

[0054] In addition, conventional fault detection methods are less efficient when dealing with a large amount of real-time data and cannot meet the requirements of rapid response and real-time detection. Moreover, fault detection of multi-connected air-conditioning systems can also be carried out through manual troubleshooting, but there is also the problem of low efficiency.

[0055] To solve the above problems of low fault detection accuracy and low efficiency, the inventor found that with the rapid development of artificial intelligence and machine learning technologies, neural networks, as a powerful non-linear modeling tool, have been widely used in various fields. Neural networks can automatically extract the internal features and laws of data by learning and training a large amount of data, so as to realize the modeling and prediction of complex systems. The self-attention mechanism plays an important role in the field of deep learning. It captures the dependencies in the sequence by calculating the correlation between each position in the input sequence and other positions. This mechanism enables the model to have a global receptive field and better understand the relationships between different positions in the sequence. Based on this characteristic, applying it to the field of fault detection of multi-connected air-conditioning systems has great potential and advantages. Therefore, the present invention proposes a fault detection method for a multi-connected air-conditioning system based on a neural network, which realizes autonomous learning and feature extraction of the operating data of the multi-connected air-conditioning system by constructing a neural network model, so as to accurately judge the fault type and location of the system.

[0056] However, when applying the self-attention mechanism to the field of multi-connected air-conditioning system fault detection, in order to better improve the accuracy of fault detection, the present invention improves the self-attention mechanism. In the adaptive multi-head attention layer, the weights of different attention heads are obtained through dynamic learning, enabling the adaptive multi-head attention layer to better capture key information in the input data and improve the accuracy of fault diagnosis. In addition, a convolutional layer is configured after the adaptive multi-head attention layer, which can better extract local features of the data and further improve the accuracy of fault diagnosis. Therefore, the present invention constructs a new fault diagnosis model. The preset fault diagnosis model includes an adaptive multi-head attention layer, a convolutional layer, a fully connected layer, a normalization layer, and an output layer to meet the requirements of fault detection for accuracy. Since the present invention uses a model for fault detection and the model is obtained through training with a large number of samples, when using this model for fault detection, compared with the method of using fixed thresholds or rules set manually, the fault detection accuracy of the model can be improved.

[0057] Furthermore, in the present invention, when obtaining parameters during fault detection, the target parameters include the sensor aging time and reference parameters selected from multiple parameters whose correlation meets the correlation condition, eliminating the influence of parameters irrelevant to the fault on fault identification, so that the selected target parameters can better reflect fault information and further improve the accuracy of fault diagnosis.

[0058] Based on the above, an embodiment of the present invention provides a fault diagnosis method for a multi-connected air-conditioning system. The execution subject of the fault diagnosis method is a fault diagnosis device such as a host computer or a server. In the multi-connected air-conditioning system, a data acquisition module is configured, and the data acquisition module is mainly used to collect data of the multi-connected air-conditioning system and then transfer it to the fault diagnosis device.

[0059] Taking the fault diagnosis device as the server and the data acquisition module as the terminal device as an example, the fault diagnosis method for the multi-connected air-conditioning system in the present invention can be applied to the Figure 1 hardware environment composed of a terminal device 102 and a server 104 as shown. As Figure 1 shown, the server 104 is connected to the terminal device 102 through a network and can be used to provide services (such as application services, etc.) for the terminal or the client installed on the terminal. A database can be set on the server or independently of the server to provide data storage services for the server 104. Cloud computing and / or edge computing services can be configured on the server or independently of the server to provide data operation services for the server 104.

[0060] The above network may include, but is not limited to, at least one of the following: a wired network, a wireless network. The above wired network may include, but is not limited to, at least one of the following: a wide area network, a metropolitan area network, a local area network. The above wireless network may include, but is not limited to, at least one of the following: WIFI (Wireless Fidelity), Bluetooth. The terminal device 102 is not limited to a PC, a mobile phone, a tablet computer, etc.

[0061] In practical applications, for a multi-connected air conditioner system, in order to monitor various operating parameters in real time, such as pressure, voltage, current, inlet and outlet pipe temperature, and operating frequency, etc., the system will install a large number of sensors, connect various sensors to a data acquisition module for data acquisition and transmission. The data acquisition module includes a processor (such as a CPU: Central Processing Unit, central processor), a memory module, a communication module (such as a fourth-generation mobile communication technology 4G module, a wireless network communication technology WIFI module, a Bluetooth module, etc.), a signal sampling module, a filtering circuit, etc.

[0062] Among them, the sensor data collected by the signal sampling module is generally an analog signal. The signal sampling module converts the analog signal into a digital signal. Since the data transmission process will be interfered by noise, and transmitting the data affected by noise to the upper computer for processing will increase the error interference. Therefore, signal noise reduction processing is realized through a filtering circuit and a noise reduction algorithm in the processor, and then uploaded to the fault diagnosis device through the communication module. Since there are various communication modules, it can ensure normal data transmission in multiple scenarios.

[0063] In this embodiment, an advanced noise reduction algorithm is integrated on the basis of hardware filtering to effectively process the original data, thereby significantly reducing the burden of the fault diagnosis device for processing data. This measure not only reduces the interference of noise on data quality, but also greatly improves the accuracy of model fault classification, providing a more reliable basis for fault diagnosis.

[0064] Refer to Figure 2 , the fault diagnosis method of the multi-connected air conditioner system may include:

[0065] S11. Obtain the parameter value of the target parameter in the multi-connected air conditioner system.

[0066] In practical applications, the target parameter is based on the actual configuration. For example, it may include the sensor aging time and the reference parameter whose correlation satisfies the correlation condition selected from multiple parameters.

[0067] In this embodiment, the sensor aging time is taken as the target parameter because the accuracy of the collected data will be affected due to sensor aging, and further the accuracy of determining faults based on the collected data will be affected. Therefore, in this embodiment, the sensor aging time is taken as the target parameter.

[0068] In addition, for a multi-connected system, multiple parameters of the multi-connected system, such as pressure, voltage, current, inlet and outlet pipe temperatures, and operating frequency, etc., will be detected by sensors. Among these parameters, some parameters will affect the fault detection accuracy, some parameters have little impact on fault detection and can be dispensable, or have a reverse impact on fault detection, such as reducing the fault detection accuracy.

[0069] Therefore, in this embodiment, the target parameter further includes a reference parameter whose correlation satisfies the correlation condition and is selected from multiple parameters.

[0070] Specifically, referring to Figure 3 , the determination process of the reference parameter may include:

[0071] S21. Obtain multiple parameters and multiple preset fault types.

[0072] In this embodiment, the types of parameters can be determined according to the actual configuration. For example, they can be all detectable dimensions, and the parameters can be such as the above-mentioned pressure, voltage, current, inlet and outlet pipe temperatures, and operating frequency, etc.

[0073] The preset fault types can be all detectable faults, such as sensor failure, four-way valve fault, condenser blockage, refrigerant reduction, etc.

[0074] S22. Calculate the correlation between each preset fault type and each parameter.

[0075] In this embodiment, the correlation calculation formula is:

[0076]

[0077] Among them, represents the degree of correlation between C and x, is the redundancy degree between x and s. Among them, x refers to a single feature, which refers to each parameter in this embodiment, C refers to the target variable, which refers to the classification label in the classification task and can be a vector or an array. In this embodiment, it refers to each fault type, and s is the selected feature, which is selected from the set of selected features. The set of selected features refers to the set composed of the parameters selected during the feature selection process.

[0078] Through the correlation calculation formula, the correlation between each preset fault type and each parameter can be calculated.

[0079] S23. Select reference parameters whose correlation with any of the preset fault types is greater than the correlation threshold.

[0080] Specifically, after calculating the correlation, a correlation threshold can be preconfigured, such as 80%. If the calculated correlation is greater than 80%, it indicates that the parameter has a high correlation with fault detection and can be retained. If the calculated correlation is not greater than 80%, it indicates that the parameter has a low correlation with fault detection and can be excluded.

[0081] For example, there are two fault types in total, such as fault type A and B. For fault type A, the parameters with a correlation greater than 80% are A, B, and C. For fault type B, the parameters with a correlation greater than 80% are A, B, and D. Then all the selected reference parameters are A, B, C, and D, which is the union of the parameters with a correlation greater than the correlation threshold for each fault type.

[0082] It should be noted that the process of determining the reference parameters can be preconfigured, that is, before fault detection, the reference parameters are determined first to improve the efficiency of subsequent fault detection.

[0083] In this embodiment, by deeply analyzing the correlation between data, those data that have no direct association with fault diagnosis are excluded through parameter screening operations, and only those data that have a significant impact on the diagnosis result are retained. A more concise and efficient data set is constructed, laying a foundation for subsequent diagnosis work. In addition, by eliminating the influence of irrelevant data on the final judgment result and ensuring that the model focuses on analyzing the information crucial for fault diagnosis, the inherent laws and structures of the data can be better captured, and the generalization ability of the model can be enhanced.

[0084] After the reference parameters are determined in advance, during actual fault detection, the parameter values of the target parameters including the sensor aging time and the reference parameters can be obtained. Refer to Figure 4 , specifically, it may include:

[0085] S31. Obtain the data uploaded by the data acquisition module configured in the multi-connected air-conditioning system according to the data upload period.

[0086] In practical applications, the data acquisition module includes a signal sampling module, a filter circuit, a processor, and a communication module; the sensor data collected by the signal sampling module is processed successively by the filter circuit and the noise reduction algorithm in the processor, and then uploaded through the communication module. For the specific operation logic of the data acquisition module, please refer to the corresponding description above.

[0087] When the data acquisition module uploads data, it can upload data according to the data upload period, which can be a fixed value or a dynamically changing value. When it is a fixed value, it can be 30s. When it is a dynamically changing value, it can be dynamically adjusted according to the actual scenario. For example, when the failure rate of the multi-connected air-conditioning system is relatively low, the probability that the data collected by the data acquisition module is normal data is relatively high, so there is no need to perform frequent fault detection, and the value of the data upload period is relatively large, such as 1 minute. When the failure rate of the multi-connected air-conditioning system is relatively high, the probability that the data collected by the data acquisition module is abnormal data is relatively high, and frequent fault detection is required to detect the change of faults in real time, and the value of the data upload period is relatively small, such as 20s.

[0088] In practical applications, the data uploaded by the data acquisition module can be all the data that can be detected, or only the data of the reference parameters. At this time, the data transmission volume can be reduced, thereby improving the data transmission efficiency.

[0089] The fault diagnosis module device, such as the host computer, after receiving the data, parses the obtained data according to the data parsing library. The data parsing rule library contains parsing rules for different protocols of sensors, numerical conversion, etc., to obtain data that the host computer can recognize.

[0090] S32. Screen out the parameter values of the reference parameters from the data.

[0091] If the data uploaded by the data acquisition module can be all the data that can be detected, then since the irrelevant data in the full amount of data will affect the diagnostic accuracy and speed, after the data acquisition module transmits the obtained full amount of data set to the host computer, it does not directly input it into the model, but screens out the parameter values of the reference parameters and only inputs the parameter values of the reference parameters into the model.

[0092] If the data uploaded by the data acquisition module is only the data of the reference parameters, then step S32 can be omitted, or the screening step of step S32 is to directly use the received data as the parameter values of the reference parameters.

[0093] S33. Calculate the sensor aging time based on the time stamp of the data uploaded by the data acquisition module or the time stamp of receiving the data.

[0094] In practical applications, the data uploaded by the data acquisition module carries a time tag of the data upload time, and the data received by the fault diagnosis device carries a time tag of the reception time. In this embodiment, since the data transmission delay is small, the time difference between the time tag of the data upload time and the time tag of the reception time is small. In this embodiment, either one can be used. For example, the time tag of the data upload time can be used to calculate the sensor aging time, or the principle of proximity can be adopted to use a time tag to calculate the sensor aging time.

[0095] In this embodiment, taking the use of the time tag of the data upload time to calculate the sensor aging time as an example, the sensor aging time is extended based on the first data upload time of the sensor. For example, every 10 days, the sensor ages by 0.01%. Then the sensor aging time = (time tag of the data upload time - first data upload time of the sensor) / 10 * 0.01%.

[0096] For each sensor, its sensor aging time is calculated, and then the average value of all the calculated sensor aging times is calculated, and the average value is used as the final sensor aging time.

[0097] S34. Use the parameter value of the reference parameter and the sensor aging time as the parameter value of the target parameter.

[0098] In this embodiment, the parameter value of the combined reference parameter and the sensor aging time are used to obtain the parameter value of the target parameter to determine the input of the model.

[0099] In this embodiment, data is collected by the data acquisition module and uploaded to the cloud platform through the gateway. Fault diagnosis can be performed in the cloud platform. By separating data collection and fault diagnosis, the data processing volume of the data acquisition module and the cloud platform can be reduced, the diagnostic portability can be improved, and the on-site maintenance cost can be reduced.

[0100] S12. Configure the position encoding identifier of the parameter value of the target parameter.

[0101] Specifically, the number of target parameters is large. When input into the fault diagnosis model, the order will be disrupted. In this embodiment, in order to avoid the disorder of data and to make full use of the position information in the data, position encoding is added to each data point in the data set. The introduction of position encoding enables the model to fully consider the order and position relationship between data when processing data, which is crucial for improving the accuracy of diagnosis. In specific implementation, a position encoding operation is performed on the parameter value of the target parameter to obtain the position encoding identifier of the parameter value of the target parameter. When input into the model later, since the parameter value of the target parameter carries the position encoding identifier, the order of the parameter value of the target parameter can be known.

[0102] S13. Input the parameter value of the target parameter configured with the position encoding identifier into a preset fault diagnosis model to obtain a fault diagnosis result.

[0103] In this embodiment, for the possible fault problems in a multi-connected air-conditioning system, a preset fault diagnosis model is obtained by modeling according to the characteristics of the multi-connected air-conditioning system. This model can distinguish and judge various fault types, and in the case where a fault has just occurred or has occurred but has not caused greater losses, it can give an alarm in time, reduce economic losses, and lower maintenance costs.

[0104] In this embodiment, the preset fault diagnosis model can be a neural network model, specifically an improvement is made on the basis of the self-attention mechanism to form a new model structure.

[0105] Refer to Figure 5 , the structure of the preset fault diagnosis model is specifically as follows:

[0106] The preset fault diagnosis model includes an adaptive multi-head attention layer, a convolutional layer, a fully connected layer, a normalization layer, and an output layer.

[0107] mMIFS-U is used for correlation calculation. Through correlation calculation, reference parameters are obtained, combined with the sensor aging time, and input data is obtained after positional encoding.

[0108] In the preset fault diagnosis model, the input data is split into three sub-data a, b, and c and input into the adaptive multi-head attention layer. The input data passes through the adaptive multi-head attention layer, the convolutional layer, and the fully connected layer in sequence, and then performs a normalization operation with the input data in the normalization layer. The data obtained from the normalization operation is used as the input of the output layer, and the output of the output layer is the fault diagnosis result.

[0109] In this embodiment, the fault diagnosis result can be one kind of fault or multiple kinds of faults, that is, the present invention can detect all the multiple faults that occur when multiple faults occur simultaneously.

[0110] In practical applications, in the adaptive multi-head attention layer, the weights of different attention heads are obtained through dynamic learning. Combining Figure 6 , introduce the internal structure of the adaptive multi-head attention layer.

[0111] The adaptive multi-head attention layer includes:

[0112] A data processing module, a Concat layer, an activation function layer, and a linear transformation layer connected in sequence;

[0113] The data processing module includes a linear transformation sub-module (linear), a matrix multiplication sub-module, a matrix transformation sub-module, and an activation layer sub-module (Leaky ReLU).

[0114] Among them, in the activation layer, the Leaky ReLU activation function is adopted to further prevent the occurrence of the gradient vanishing phenomenon. The Leaky ReLU activation function is:

[0115]

[0116] Where: refers to the matrix output by the matrix transformation sub-module, refers to the hyperparameter. In practical applications, .

[0117] In practical applications, within the data processing module, the three input sub-data a, b, and c obtained by splitting the input data are each configured with a corresponding linear, that is, there are three liners. In addition, in the present invention, the number of matrix multiplication sub-modules is two, that is, two matrix multiplication operations will be performed. For the input sub-data a, the linear transformation matrix in its linear is , the linear transformation matrix in the linear corresponding to the input sub-data b is , and the linear transformation matrix in the linear corresponding to the input sub-data c is .

[0118] Among the three input sub-data, two input sub-data, such as a and b, sequentially pass through the linear transformation sub-module, and after the output result passes through the matrix multiplication sub-module, the matrix transformation sub-module, and the activation functor number module, a matrix multiplication operation is performed with the linear transformation result (linear result) of the other input sub-data c.

[0119] As can be seen from the above, in the data processing module, a, b, and c are respectively multiplied by the linear transformation matrices , , to obtain Q, K, V. The calculation formula in the data processing module is:

[0120]

[0121] Where, is the number of columns of Q and K. In this implementation, The result of, is the output result of the matrix multiplication sub-module that is located before Concat and adjacent to Concat in Figure 6 .

[0122] In this embodiment, the adaptive multi-head attention is improved by introducing an additional network layer in the adaptive multi-head attention to dynamically learn the weights between different attention heads, and these weights can be adaptively adjusted according to the changes in the input data. Therefore, in this embodiment, a concatenation layer (Concat), an activation function layer, and a linear transformation layer are added to the integration of the adaptive attention.

[0123] In the Concat layer, the outputs of each attention head calculated are concatenated together to obtain a vector through a linear transformation. In the activation function layer, the concatenation result is multiplied by the intermediate weight matrix to finally obtain the weights of each head.

[0124] Among them, the calculation formula configured in the activation function layer is:

[0125] .

[0126] Among them, is the output of multiple attention heads, is the total number of attention heads, is the intermediate weight matrix, not the final weights of each head, is the weight matrix of the attention head, which includes the final weights of the attention heads.

[0127] .

[0128] The output of each attention head represents the information of the data in different representation spaces. Through the Concat output, a long vector containing the information of all heads is obtained. Then, through matrix multiplication, the information of each attention head is dot-producted with the corresponding row in the intermediate weight matrix . After the obtained result is normalized by softmax, the weights of each attention head are obtained.

[0129] Through this process, the model can adaptively adjust the weights of different heads according to the characteristics of the input data, so as to adjust the attention degree to different information. The intermediate weight matrix is learned by the model during the training process. Through the backpropagation algorithm, the model will update the value of the weight matrix according to the gradient of the loss function to optimize the performance of the model, and will gradually adjust to the optimal state during the training process.

[0130] After passing through the activation function layer, the weight matrix of the attention head can be obtained, and the weights of each attention head can be queried from it. Taking the attention head as an example, its weight is , and the output is , the final result is obtained through a linear transformation operation. The calculation formula configured in the linear transformation layer is:

[0131]

[0132] where: is the output of the attention head , ranges from 1 to , is the weight of the attention head , obtained from , is the weighted output result, that is, the sum of the products of each attention head and the weight corresponding to its corresponding attention head. That is, in this embodiment, after calculating the output matrix of the nth attention head, it is multiplied by the weight matrix to obtain the final weighted output.

[0133] In this embodiment, the final output is calculated according to the weight. Since the final output contains the attention weighted information extracted from all heads, it helps the model to better capture the key information in the input data.

[0134] The adaptive multi-head attention mechanism ensures that the model can focus on the important parts of the input data, and configuring a convolutional layer after the adaptive multi-head attention can further extract useful features from these important parts, enabling the model to better understand the structure of the input data and have higher accuracy and robustness when dealing with complex tasks.

[0135] After the convolutional layer is the fully connected layer, which can be a feed-forward fully connected layer. The calculation formula is: , where X is the input data including a, b, and c, W1 and W2 are weight matrices, b1 and b2 are bias matrices, and the finally obtained value is the value with the highest predicted probability, and the result is classified.

[0136] After the fully connected layer is the normalization layer. In this embodiment, residual processing is added in the normalization layer. The residual connection can alleviate the problem of gradient disappearance caused by the increase in network depth, and helps the model to maintain high performance in complex fault situations. The formula used in the normalization layer is:

[0137]

[0138] where: is the standard deviation, is the parameter updated with the model, is the mean, is a decimal number to prevent the denominator from being zero, with a value of 0.0000001, is the output of the normalization layer.

[0139] At the output layer, the activation function outputs the data within a specific range by introducing a non - linear transformation, and finally obtains the fault classification through the output layer.

[0140] In practical applications, the preset fault diagnosis model is obtained through a large amount of data training, and its training process includes:

[0141] 1) Obtain data samples.

[0142] Among them, the data samples include the sample values of the target parameters and the fault identifiers of the target parameters.

[0143] Specifically, in practical applications, a large number of parameter values of the target parameters with faults can be obtained as samples. For example, when there is fault A, the sample value of its target parameter is..., when there are faults A and B, the sample value of its target parameter is..., when there are faults A and C, the sample value of its target parameter is..., when there are faults A, B, and C, the sample value of its target parameter is..., to obtain a large - scale dataset.

[0144] 2) Use the data samples to train the fault diagnosis model until the model training stop condition is met and then stop.

[0145] Specifically, the training process of the fault diagnosis model is as follows: After constructing and processing the dataset through the above steps, shuffle the dataset, and then divide the data into a dataset and a test set in a ratio of 7:3. For each iteration of model training, record the model parameters and output results for future analysis and improvement.

[0146] If the model training stop condition is met (such as the number of training times reaches the number of times to stop training), then the training can be stopped. Subsequently, by evaluating the performance, determine whether to retrain or directly put it into use online.

[0147] 3) Use the confusion matrix to evaluate the performance of the trained fault diagnosis model. If the performance meets the performance requirements, use the trained fault diagnosis model as the preset fault diagnosis model.

[0148] Among them, the confusion matrix includes accuracy, recall rate, and F1 - score.

[0149] Specifically, since the model involves multi - classification problems, using only accuracy to verify cannot fully represent the model performance. Therefore, the confusion matrix is used to verify the accuracy of the final model. The confusion matrix incorporates evaluation metrics such as accuracy, recall rate, and F1 - score to comprehensively evaluate the model performance. If the performance evaluated using the confusion matrix meets the performance requirements, it indicates that the model training accuracy meets the requirements, and then the training can be stopped. Subsequently, the model can be directly used for fault diagnosis operations. If the performance requirements are not met, retrain again.

[0150] In this embodiment, the model not only provides accurate fault diagnosis results, but also can generate highly personalized diagnosis reports and repair suggestions according to the specific needs and scenarios of users.

[0151] During the use of the model, real-time data collection, fault detection, and fault location can be achieved. Specifically, referring to Figure 7 , the multi-connected unit device is installed in the actual environment for verification. The data transmitted in real time through various sensors is input into the model for fault diagnosis to judge and classify the faults. If a fault occurs, further fault location is carried out, and the cloud platform issues an operation command to the device controller to control the operating state of the multi-connected air conditioner device by using the preset algorithm. If it is diagnosed that no fault has occurred, no treatment is performed.

[0152] In this embodiment, when performing fault diagnosis, it is implemented by using a preset fault diagnosis model. The preset fault diagnosis model includes an adaptive multi-head attention layer, a convolutional layer, a fully connected layer, a normalization layer, and an output layer. In the adaptive multi-head attention layer, the weights of different attention heads are obtained through dynamic learning, so that the adaptive multi-head attention layer can better capture the key information in the input data and improve the accuracy of fault diagnosis. Further, a convolutional layer is configured after the adaptive multi-head attention layer, which can better extract the local features of the data and further improve the accuracy of fault diagnosis. Further, in the present invention, when obtaining the target parameters, the target parameters include the sensor aging time and the reference parameters whose correlation satisfies the correlation condition selected from multiple parameters, eliminating the influence of parameters irrelevant to the fault on fault recognition, so that the selected target parameters can better reflect the fault information and further improve the accuracy of fault diagnosis.

[0153] In addition, the present invention combines the adaptive multi-head attention mechanism with the convolutional layer, which can not only capture the global attention information but also extract local features, realizing the comprehensive recognition of fault patterns. Specifically, the processed data set is input into the adaptive multi-head attention mechanism. This mechanism can automatically adjust the attention weights between different data points, enabling the model to focus on the data points that have a greater impact on the result during the diagnosis process. In this way, the model can more accurately capture the key information of the fault occurrence. Then, the output of the multi-head attention mechanism is sent to the convolutional layer for further processing. The convolutional layer can extract the local features in the data, which is of great significance for identifying fault patterns. Through convolutional operations, richer and deeper feature information can be extracted, providing strong support for subsequent fault classification.

[0154] In addition, in order to improve the training efficiency and stability of the model, the present invention adds a residual connection after the convolutional layer. The residual connection allows the model to skip certain layers during training and directly transmit information, thus effectively alleviating the problem of vanishing gradients. This enables the model to converge to the optimal solution more quickly and enhances its diagnostic ability in complex fault situations.

[0155] Furthermore, the output of the model is non-linearly transformed through an activation function to obtain the final fault diagnosis result. The introduction of the activation function enables the model to learn and approximate complex fault patterns, improving the accuracy and reliability of diagnosis.

[0156] In summary, by modifying the model architecture as described above, the model can make full use of the relevant information in the data and, through the processing of the adaptive multi-head attention mechanism and the convolutional layer, achieve accurate classification and diagnosis of faults.

[0157] Based on the above embodiments of the fault diagnosis method for a multi-connected air-conditioning system, another embodiment of the present invention provides a fault diagnosis device for a multi-connected air-conditioning system. Referring to Figure 8 , it may include:

[0158] A data acquisition module 11 for acquiring the parameter values of target parameters in the multi-connected air-conditioning system; the target parameters include: the sensor aging time and reference parameters whose correlation meets the correlation condition selected from multiple parameters;

[0159] A position encoding module 12 for configuring the position encoding identifiers of the parameter values of the target parameters;

[0160] A fault diagnosis module 13 for using the parameter values of the target parameters configured with position encoding identifiers as input data and inputting them into a preset fault diagnosis model to obtain a fault diagnosis result; the preset fault diagnosis model includes an adaptive multi-head attention layer, a convolutional layer, a fully connected layer, a normalization layer, and an output layer. In the preset fault diagnosis model, the input data sequentially passes through the adaptive multi-head attention layer, the convolutional layer, and the fully connected layer, and then performs a normalization operation with the input data in the normalization layer. The data obtained from the normalization operation is used as the input of the output layer, and the output of the output layer is the fault diagnosis result; in the adaptive multi-head attention layer, the weights of different attention heads are obtained through dynamic learning.

[0161] Further, the adaptive multi-head attention layer includes:

[0162] A data processing module, a splicing layer, an activation function layer, and a linear transformation layer connected in sequence;

[0163] The data processing module includes a linear transformation sub-module, a matrix multiplication sub-module, a matrix transformation sub-module, and an activation layer sub-module. Inside the data processing module, among the three input sub-data obtained by splitting the input data, after two input sub-data sequentially pass through the linear transformation sub-module, the matrix multiplication sub-module, the matrix transformation sub-module, and the activation function module, a matrix multiplication operation is performed with the linear transformation result of the other input sub-data.

[0164] Further, the calculation formula configured in the activation function layer is:

[0165] , where is the output of multiple attention heads, is the total number of attention heads, is the intermediate weight matrix, is the weight matrix of the attention head;

[0166] The calculation formula configured in the linear transformation layer is:

[0167]

[0168] where is the output of the attention head , ranges from 1 to , is the weight of the attention head , obtained from , is the weighted output result.

[0169] Further, it further includes:

[0170] A parameter determination module, configured to obtain multiple parameters and multiple preset fault types, calculate the correlation between each preset fault type and each parameter, and screen out reference parameters whose correlation with any preset fault type is greater than the correlation threshold.

[0171] Further, the data acquisition module 11 includes:

[0172] A data acquisition sub-module, configured to acquire data uploaded by a data acquisition module configured in a multi-connected system according to a data upload period; the data acquisition module includes a signal sampling module, a filter circuit, a processor, and a communication module; the sensor data collected by the signal sampling module is sequentially processed by the noise reduction algorithm in the filter circuit and the processor, and then uploaded through the communication module;

[0173] A data screening sub-module, configured to screen out the parameter values of the reference parameters from the data;

[0174] A time calculation sub-module, configured to calculate the sensor aging time based on the time tag of the data uploaded by the data acquisition module or the time tag of receiving the data;

[0175] A data determination sub-module, configured to use the parameter value of the reference parameter and the sensor aging time as the parameter value of the target parameter.

[0176] Furthermore, the position encoding module 12 is specifically configured to:

[0177] Perform a position encoding operation on the parameter value of the target parameter to obtain a position encoding identifier of the parameter value of the target parameter.

[0178] Furthermore, it further includes a model training module, including:

[0179] A sample acquisition sub-module, configured to acquire data samples; the data samples include sample values of the target parameter and fault identifiers of the target parameter;

[0180] A model training sub-module, configured to use the data samples to train a fault diagnosis model until the model training stop condition is met and then stop; use a confusion matrix to evaluate the performance of the trained fault diagnosis model, and if the performance meets the performance requirements, use the trained fault diagnosis model as a preset fault diagnosis model; the confusion matrix includes accuracy, recall, and F1 score.

[0181] In this embodiment, during fault diagnosis, it is implemented using a preset fault diagnosis model. The preset fault diagnosis model includes an adaptive multi-head attention layer, a convolutional layer, a fully connected layer, a normalization layer, and an output layer. In the adaptive multi-head attention layer, the weights of different attention heads are obtained through dynamic learning, enabling the adaptive multi-head attention layer to better capture key information in the input data and improve the accuracy of fault diagnosis. Furthermore, configuring a convolutional layer after the adaptive multi-head attention layer can better extract local features of the data and further improve the accuracy of fault diagnosis. Further, in the present invention, when obtaining parameters, the target parameters include the sensor aging time and reference parameters whose correlation satisfies the correlation condition selected from multiple parameters, eliminating the influence of parameters irrelevant to the fault on fault identification, so that the selected target parameters can better reflect fault information and further improve the accuracy of fault diagnosis.

[0182] It should be noted that for the working processes of each module and sub-module in this embodiment, please refer to the corresponding descriptions in the above embodiments and will not be elaborated here.

[0183] Based on the above embodiments of the fault diagnosis method and device for a multi-connected air conditioner system, another embodiment of the present invention provides a computer-readable storage medium, which includes a stored program. When the program runs, it executes the above-mentioned fault diagnosis method for the multi-connected air conditioner system.

[0184] Based on the above embodiments of the fault diagnosis method and device for a multi-connected air conditioner system, another embodiment of the present invention provides an electronic device, which can be the above-mentioned fault diagnosis device, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to execute the above-mentioned fault diagnosis method for the multi-connected air conditioner system through the computer program.

[0185] In this embodiment, when performing fault diagnosis, it is implemented using a preset fault diagnosis model, which includes an adaptive multi-head attention layer, a convolutional layer, a fully connected layer, a normalization layer, and an output layer. In the adaptive multi-head attention layer, the weights of different attention heads are obtained through dynamic learning, enabling the adaptive multi-head attention layer to better capture key information in the input data and improve the accuracy of fault diagnosis. Further, a convolutional layer is configured after the adaptive multi-head attention layer to better extract local features of the data and further improve the accuracy of fault diagnosis. Further, in the present invention, when obtaining the target parameters, the target parameters include the sensor aging time and reference parameters that are screened from multiple parameters and whose correlation satisfies the correlation condition, eliminating the influence of parameters irrelevant to the fault on fault identification, so that the screened target parameters can better reflect the fault information and further improve the accuracy of fault diagnosis.

[0186] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A fault diagnosis method for a multi-connected unit system, characterized in that Including: Obtaining the parameter values of target parameters in a multi-connected air-conditioning system; The target parameters include: the sensor aging time and reference parameters whose correlation satisfies the correlation condition selected from multiple parameters; Configuring the position coding identifier of the parameter value of the target parameter; Taking the parameter values of the target parameter configured with the position coding identifier as input data and inputting them into a preset fault diagnosis model to obtain a fault diagnosis result; the preset fault diagnosis model includes an adaptive multi-head attention layer, a convolutional layer, a fully connected layer, a normalization layer, and an output layer. In the preset fault diagnosis model, the input data sequentially passes through the adaptive multi-head attention layer, the convolutional layer, and the fully connected layer, and then performs a normalization operation with the input data in the normalization layer. The data obtained from the normalization operation is used as the input of the output layer, and the output of the output layer is the fault diagnosis result; in the adaptive multi-head attention layer, the weights of different attention heads are obtained through dynamic learning.

2. The fault diagnosis method of the multi-connected air conditioner system according to claim 1, characterized in that The adaptive multi-head attention layer includes: A data processing module, a splicing layer, an activation function layer, and a linear transformation layer connected in sequence; The data processing module includes a linear transformation sub-module, a matrix multiplication sub-module, a matrix transformation sub-module, and an activation layer sub-module. In the data processing module, among the three input sub-data obtained by splitting the input data, two input sub-data sequentially pass through the linear transformation sub-module, the matrix multiplication sub-module, the matrix transformation sub-module, and the activation function module, and then perform a matrix multiplication operation with the linear transformation result of the other input sub-data.

3. The fault diagnosis method of the multi-connected air conditioner system according to claim 2, wherein The calculation formula configured in the activation function layer is: , where is the output of multiple attention heads, is the total number of attention heads, is the intermediate weight matrix, is the weight matrix of the attention head; The calculation formula configured in the linear transformation layer is: Among them, is the output of the attention head , and the value range of is from 1 to . is the weight of the attention head , obtained from , and is the weighted output result.

4. The fault diagnosis method of the multi-connected air conditioner system according to claim 1, characterized in that The determination process of the reference parameter includes: Obtaining multiple parameters and multiple preset fault types; Calculating the correlation between each preset fault type and each parameter; Selecting reference parameters whose correlation with any preset fault type is greater than the correlation threshold.

5. The fault diagnosis method of the multi-connected air conditioner system according to claim 1, characterized in that Obtaining the parameter values of the target parameters in the multi-connected air-conditioning system includes: Obtaining the data uploaded by the configured data acquisition module in the multi-connected air-conditioning system according to the data upload period; the data acquisition module includes a signal sampling module, a filter circuit, a processor, and a communication module; the sensor data collected by the signal sampling module sequentially passes through the filter circuit and the noise reduction algorithm in the processor and is uploaded through the communication module; Selecting the parameter values of the reference parameters from the data; Calculating the sensor aging time based on the time tag of the data uploaded by the data acquisition module or the time tag of receiving the data; Taking the parameter values of the reference parameters and the sensor aging time as the parameter values of the target parameters.

6. The fault diagnosis method of the multi-connected air conditioner system according to claim 1, characterized in that Configuring the position coding identifier of the parameter value of the target parameter includes: Performing a position coding operation on the parameter value of the target parameter to obtain the position coding identifier of the parameter value of the target parameter.

7. The fault diagnosis method of the multi-connected air conditioner system according to claim 1, characterized in that, The training process of the preset fault diagnosis model includes: Obtaining data samples; the data samples include the sample values of the target parameters and the fault identifiers of the target parameters; Train the fault diagnosis model using the data samples until the model training stop condition is met and then stop; Evaluate the performance of the trained fault diagnosis model using a confusion matrix. If the performance meets the performance requirements, use the trained fault diagnosis model as the preset fault diagnosis model; the confusion matrix includes accuracy, recall, and F1 score.

8. A fault diagnosis device for a multi-connected air conditioner system, characterized in that Comprising: A data acquisition module for acquiring the parameter values of the target parameters in the multi-connected air-conditioning system; The target parameters include: the sensor aging time, and reference parameters whose correlation satisfies the correlation condition selected from multiple parameters; A position encoding module for configuring the position encoding identifiers of the parameter values of the target parameters; A fault diagnosis module for using the parameter values of the target parameters configured with position encoding identifiers as input data and inputting them into the preset fault diagnosis model to obtain a fault diagnosis result; the preset fault diagnosis model includes an adaptive multi-head attention layer, a convolutional layer, a fully connected layer, a normalization layer, and an output layer. In the preset fault diagnosis model, the input data sequentially passes through the adaptive multi-head attention layer, the convolutional layer, and the fully connected layer and then performs a normalization operation with the input data in the normalization layer. The data obtained from the normalization operation is used as the input of the output layer, and the output of the output layer is the fault diagnosis result; in the adaptive multi-head attention layer, the weights of different attention heads are obtained through dynamic learning.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein the program, when running, executes the fault diagnosis method of the multi-connected air-conditioning system according to any one of claims 1 to 7.

10. An electronic device, comprising a memory and a processor, characterized in that, A computer program is stored in the memory, and the processor is configured to execute the fault diagnosis method of the multi-connected air-conditioning system according to any one of claims 1 to 7 through the computer program.

Images