Fault detection method, device and equipment of metering automation master station and storage medium

Abstract

The application discloses a kind of metrology automation main station's fault detection method, device, equipment and storage medium, by two-dimensional data sample, preset variational autoencoder is trained, until preset variational autoencoder converges, obtain target variational autoencoder;Decoder is replaced by preset convolutional neural network, obtain initial fault detection model;Based on the target two-dimensional data sample obtained after two-dimensional data sample is processed by noise based on encoder, preset convolutional neural network in initial fault detection model is trained, obtain target fault detection model;Using target fault detection model, according to the target operation data of metrology automation main station, the fault detection of metrology automation main station is carried out, and the fault result data of metrology automation main station is obtained.The application is by the noise reduction processing of data noise introduced in data acquisition stage, to repair data to a certain extent, improve the robustness of fault detection model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of metering automation master station, and particularly relates to a fault detection method, device and equipment of metering automation master station and a storage medium. BACKGROUND

[0002] The metering automation master station is a computer system connected with various metering automation terminals, and it is the information collection and control center of the entire metering automation system. In order to avoid the long-time fault of the metering automation system from affecting the normal operation, the fault detection of the metering automation master station is needed to improve the fault repair efficiency.

[0003] At present, the fault classification method of the metering automation master station mainly includes manual inspection and intelligent detection based on neural network. Among them, the manual inspection process is tedious and cannot detect global faults; the intelligent detection method based on ANN and CNN neural network can solve the problems existing in the manual inspection method, but the accuracy depends on the integrity of the data set. If the master station log is missing, damaged or has a large amount of noise, the accuracy of this method will be seriously reduced. It can be seen that the current intelligent detection method based on neural network has the problem of poor model robustness. SUMMARY

[0004] The present application provides a fault detection method, device and equipment of metering automation master station to solve the technical problem of poor model robustness of the current intelligent detection method based on neural network.

[0005] In order to solve the above technical problem, in a first aspect, the present application provides a fault detection method of metering automation master station, comprising:

[0006] Randomly sampling the historical operation data of the metering automation master station, and generating the random sampling result as a two-dimensional data sample;

[0007] Training the preset variational autoencoder based on the two-dimensional data sample until the preset variational autoencoder converges, obtaining a target variational autoencoder, the target variational autoencoder comprising an encoder and a decoder;

[0008] Replacing the decoder with a preset convolutional neural network to obtain an initial fault detection model;

[0009] Training the preset convolutional neural network in the initial fault detection model based on the target two-dimensional data sample obtained after the noise processing of the two-dimensional data sample by the encoder until the preset convolutional neural network converges, obtaining a target fault detection model;

[0010] The target fault detection model is used to detect faults of the metering automation master station according to target operation data of the metering automation master station, and fault result data of the metering automation master station is obtained.

[0011] Preferably, the preset variational autoencoder converges, comprising:

[0012] The preset evidence lower bound ELBO function is used as a target loss function to perform gradient descent on each weight variable in the encoder and the decoder of the preset variational autoencoder to update the weight variable until the target loss function reaches a preset convergence condition.

[0013] Preferably, the ELBO function comprises a reconstruction loss function and a KL divergence loss function, and the expression of the ELBO function is:

[0014]

[0015] wherein, ELBO represents the ELBO function, reconstruction loss represents the reconstruction loss function, KL divergence loss represents the KL divergence loss function, and β represents a weight value of the KL divergence loss function.

[0016] Preferably, the weight value β of the KL divergence loss function is:

[0017]

[0018] wherein c is a constant, μ is a constant to prevent β from tending to infinity in the initial stage, and i th epoch is the number of iterations, and k is a constant,

[0019] Preferably, the decoder is replaced by a preset convolutional neural network to obtain an initial fault detection model, comprising:

[0020] The weight parameters of the encoder in the target variational autoencoder are locked, and the decoder is replaced by a preset convolutional neural network to obtain an initial fault detection model.

[0021] Preferably, the target variational autoencoder comprises a convolutional layer and a fully connected layer, and the convolutional layer is used to extract associated features between two-dimensional data samples.

[0022] Preferably, the historical operation data of the metering automation master station is randomly sampled, and the random sampling result is generated as a two-dimensional data sample, comprising:

[0023] The sampling frequency when the metering automation master station collects historical operation data is used to determine a plurality of sampling time periods of the historical operation data;

[0024] The maximum value and the minimum value of the historical operation data in the sampling time period are randomly extracted by using a random sampling window;

[0025] The maximum value and the minimum value corresponding to each sampling time period are combined into a two-dimensional data sample.

[0026] In a second aspect, the present application provides a fault detection device of a metering automation master station, comprising:

[0027] A sampling module is configured to randomly sample historical operation data of the metering automation master station and generate a random sampling result as a two-dimensional data sample;

[0028] A first training module is configured to train a preset variational autoencoder based on the two-dimensional data sample until the preset variational autoencoder converges, so as to obtain a target variational autoencoder, wherein the target variational autoencoder comprises an encoder and a decoder;

[0029] A replacement module is configured to replace the decoder with a preset convolutional neural network to obtain an initial fault detection model;

[0030] A second training module is configured to train the preset convolutional neural network in the initial fault detection model based on a target two-dimensional data sample obtained by performing noise processing on the two-dimensional data sample by using the encoder until the preset convolutional neural network converges, so as to obtain a target fault detection model;

[0031] A detection module is configured to perform fault detection on the metering automation master station based on target operation data of the metering automation master station by using the target fault detection model, so as to obtain fault result data of the metering automation master station.

[0032] In a third aspect, the present application provides a computer device comprising a processor and a memory, wherein the memory is configured to store a computer program, and the computer program is configured to implement the fault detection method of the metering automation master station according to the first aspect when executed by the processor.

[0033] In a fourth aspect, the present application provides a computer readable storage medium storing a computer program, and the computer program is configured to implement the fault detection method of the metering automation master station according to the first aspect when executed by a processor.

[0034] Compared with the prior art, the present application has the following beneficial effects:

[0035] The historical operation data of the metering automation master station is randomly sampled, and the random sampling result is generated as a two-dimensional data sample to reduce the data amount; then, the preset variational autoencoder is trained based on the two-dimensional data sample until the preset variational autoencoder converges, so as to train the target variational autoencoder capable of performing noise processing on the sample, and the decoder is replaced by the preset convolutional neural network to obtain an initial fault detection model, and the preset convolutional neural network in the initial fault detection model is trained based on the target two-dimensional data sample obtained by performing noise processing on the two-dimensional data sample by the encoder until the preset convolutional neural network converges, so as to obtain a target fault detection model, the data noise introduced in the data collection stage is denoised by the target variational autoencoder to repair the data to a certain extent, and the robustness of the fault detection model is improved; finally, the target fault detection model is used to perform fault detection on the metering automation master station according to the target operation data of the metering automation master station, so as to obtain the fault result data of the metering automation master station, so that the target fault detection model with higher robustness can be used for fault detection in actual application, the accuracy and anti-interference ability of fault detection are improved, and the method is more suitable for industrial environment. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 A flowchart of a fault detection method for a metering automation master station is shown in the embodiments of the present application;

[0037] Figure 2 A structure diagram of a preset variational autoencoder is shown in the embodiments of the present application;

[0038] Figure 3 A structure diagram of a target fault detection model is shown in the embodiments of the present application;

[0039] Figure 4 A structure diagram of a fault detection device for a metering automation master station is shown in the embodiments of the present application;

[0040] Figure 5 A structure diagram of a computer device is shown in the embodiments of the present application. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0042] Please refer to Figure 1 , Figure 1A flowchart of a fault detection method of a metering automation master station is provided for an embodiment of the present application. The fault detection method of the metering automation master station of the embodiment of the present application can be applied to a computer device, including but not limited to a smart phone, a notebook computer, a tablet computer, a desktop computer, a physical server and a cloud server, and the like. As shown in Figure 1 The fault detection method of the metering automation master station of the embodiment includes steps S101 to S105, which are described in detail as follows:

[0043] Step S101, randomly sampling historical running data of the metering automation master station, and generating a random sampling result as a two-dimensional data sample.

[0044] In this step, since network congestion, interface damage and the like are the main causes of faults in the fault detection process of the metering automation master station, the historical running data includes but is not limited to a large amount of related data such as data traffic, TCP connection number and server CPU occupancy rate. The metering automation master station is usually in the unit of a prefecture-level city, so it needs to face the problem of a large amount of data. Therefore, the embodiment reduces the amount of data by means of random sampling.

[0045] In an embodiment, the step S101 includes:

[0046] determining a plurality of sampling time periods of the historical running data according to a sampling frequency at which the metering automation master station collects the historical running data;

[0047] randomly extracting a maximum value and a minimum value of the historical running data in the sampling time periods by using a random sampling window;

[0048] combining the maximum value and the minimum value corresponding to each sampling time period into the two-dimensional data sample.

[0049] In this embodiment, the maximum value and the minimum value of each time period are extracted to reduce the amount of data, wherein the time period is determined in combination with the sampling frequency to ensure the relevance of the data as much as possible.

[0050] Step S102, training a preset variational autoencoder based on the two-dimensional data sample until the preset variational autoencoder converges, obtaining a target variational autoencoder, the target variational autoencoder including an encoder and a decoder.

[0051] In this step, as Figure 2A structure diagram of a preset variational autoencoder is shown, which includes an encoder, a hidden space and a decoder. The variational autoencoder (VAE) is a probabilistic graphical model that can generate new data. It can generate new data with variability but similar structural characteristics to the input data by learning the probability distribution of the data in a high-dimensional space. The goal of the variational autoencoder is to model the distribution of the data and represent it as a hidden space. It is usually assumed that the variables in the space are a combination of several Gaussian distributions and these Gaussian distributions are used to fit the true probability distribution.

[0052] Optionally, assuming that the sample observed by the sensor is X = [x1, x2,... xn] is an n-dimensional vector, an encoder is first constructed by a fully connected neural network method to compress the n-dimensional data into a d-dimensional vector Z = [z1, z2,... zd], where Z should satisfy the Gaussian distribution:

[0053]

[0054] where μ is the mean, ∑ is the variance, μ and ∑ are functions of x, and e is an auxiliary variable subject to a normal distribution.

[0055] Optionally, the preset variational autoencoder replaces the fully connected layer in the encoder with a combination of convolutional and fully connected layers, i.e., the target variational autoencoder includes convolutional and fully connected layers. After converting the data collected over a period of time into a two-dimensional format, the data correlation features over a period of time can be obtained by convolution. Then, a decoder is constructed using a symmetric neural network to decode the state variable Z into a vector form similar to the sample variable to generate sample-like data X'. It should be noted that the structure of the encoder and the decoder of the variational autoencoder is usually symmetrical, and multiple deconvolution layers in the decoder will make the reconstruction loss term larger, so only one deconvolution layer is set in the decoder to restore the picture dimension.

[0056] In an embodiment, the preset variational autoencoder converges, comprising:

[0057] A preset evidence lower bound ELBO function is used as a target loss function to perform gradient descent on each weight variable in the encoder and the decoder of the preset variational autoencoder to update the weight variable until the target loss function reaches a preset convergence condition.

[0058] In the embodiment, the two-dimensional data sample is input into the preset variational autoencoder for training, and the process is a self-supervised training process, which only needs to update each weight variable in the encoder and the decoder by using the principle of gradient descent according to the ELBO function, so that no sample label is needed, and in addition, there is a process of converting two-dimensional data into one-dimensional data in the process, which is directly completed by using the row-by-row and column-by-column sequential arrangement method, and similarly, the conversion of one-dimensional data into two-dimensional data is the reverse process of the process. It should be noted that, since the variational autoencoder is self-supervised learning, the main use is data generation, which solves the problem of input noise or even damage.

[0059] Optionally, in the process of model training, an ELBO function is constructed based on a conditional probability, the ELBO function includes a reconstruction loss function and a KL divergence loss function, wherein the reconstruction loss term is a function constructed by considering the similarity of input and output, and the KL divergence is used as a regularizer to force the Z variable to approach a Gaussian distribution: the expression of the ELBO function is:

[0060]

[0061] wherein, ELBO represents the ELBO function, reconstruction loss function, KL divergence loss function, and β represents a weight value for controlling the KL divergence loss function.

[0062] Optionally,

[0063] Optionally, the weight value β for controlling the KL divergence loss function is:

[0064]

[0065] wherein c is a constant, μ is a constant for preventing β from tending to infinity in the initial stage, i th epoch is the number of iterations, and k is a constant, A weight variable β is added according to the characteristics of the KL divergence to more effectively enable the model to learn the time sequence correlation between data.

[0066] In step S103, the decoder is replaced with a preset convolutional neural network to obtain an initial fault detection model.

[0067] In this step, the weight parameters of the encoder in the target variational autoencoder are locked, and the decoder is replaced with the preset convolutional neural network to obtain the initial fault detection model. Exemplarily, as shown in Figure 3The structure diagram of the target fault detection model is shown. After the training process of the target variational autoencoder is completed, the encoder parameters and the hidden space parameters in the target variational autoencoder are locked, the encoder is used to complete data repair, and then a preset convolutional neural network is used to replace the decoder part as a fault detection classifier.

[0068] In step S104, the preset convolutional neural network in the initial fault detection model is trained based on the target two-dimensional data sample obtained after the encoder performs noise processing on the two-dimensional data sample, until the preset convolutional neural network converges, and the target fault detection model is obtained.

[0069] In this step, the convolutional neural network (CNN) belongs to the feedforward neural network, and the hierarchical structure is in series. The whole is composed of one or more convolutional layers and a fully connected layer at the back end. The original input is convolved by the convolution kernel of the convolutional layer to obtain a multi-channel feature slice, and then is aggregated by the pooling layer. The neurons in the whole network model are activated by a suitable activation function to work, and finally a confidence vector of fault labels is output in the fully connected layer.

[0070] In the convolutional neural network, the whole convolution process is performed by a two-dimensional convolution kernel (also known as a discrete two-dimensional filter). After one convolution process, a feature map is generated from the original input matrix. The whole process is often performed by multiple two-dimensional convolution kernels. Therefore, the whole convolution process will generate a multi-channel feature map to complete the convolution feature extraction of the original input matrix. One convolution process is as follows: the convolution kernel is sequentially moved to all positions on the two-dimensional matrix with a designed step as the sliding step, and an inner product operation is performed on the convolution kernel and the pixels at each position. Since the convolution process shares the weight of the convolution kernel, it can greatly reduce the number of model parameters and the amount of model calculation while ensuring the accuracy of the final result.

[0071] It can be understood that after the training of the target variational autoencoder is completed, all weight variable values are fixed, and the decoder part is removed. Then, a new convolutional neural network is added after the hidden space as a fault classifier, and the training is performed again until the parameters converge. This process can be regarded as the training process of the ordinary convolutional neural network.

[0072] In step S105, the target fault detection model is used to perform fault detection on the metering automation master station according to the target operation data of the metering automation master station, and fault result data of the metering automation master station is obtained.

[0073] In this step, the target running data collected by the metering automation master station is converted into two-dimensional data, and then the two-dimensional data is input into the target fault detection model of the target variational autoencoder to convert the two-dimensional data into one-dimensional data, and then the target two-dimensional data is regenerated according to the one-dimensional data, so as to achieve the purpose of noise reduction, and finally the target two-dimensional data is input into the convolutional neural network of the target fault detection model. The fault result data is obtained.

[0074] It should be noted that, compared with the prior art, the present application has the following beneficial effects:

[0075] (1) The fault detection method of the metering automation master station mixed with the convolutional neural network and the variational autoencoder is safer, faster and can reduce the cost compared with manual fault inspection.

[0076] (2) The noise introduced by the fault in the data acquisition stage is processed by the variational autoencoder, the lost data is repaired to a certain extent, and the robustness of the fault detection module is improved.

[0077] (3) Compared with the VAE, CNN, ANN and other fault detection algorithms, the accuracy of the present application is higher, the anti-interference performance is stronger, and the present application is more suitable for industrial environment.

[0078] (4) For the training process of the variational autoencoder, a variable β is added according to the characteristics of the KL divergence, so that the model can learn the time sequence correlation between the data more effectively.

[0079] (5) For the network system with large amount of data, the maximum and minimum value method in a certain time period is adopted to reduce the amount of data, effectively solving the problem of large amount of calculation.

[0080] In order to perform the fault detection method of the metering automation master station corresponding to the above-mentioned method embodiment, so as to realize the corresponding functions and technical effects. Referring to Figure 4 , Figure 4 The structure block diagram of the fault detection device of the metering automation master station provided by the embodiment of the present application is shown. For the convenience of description, only the part related to the present embodiment is shown. The fault detection device of the metering automation master station provided by the embodiment of the present application comprises:

[0081] The sampling module 401 is used for randomly sampling the historical running data of the metering automation master station, and generating the random sampling result as a two-dimensional data sample;

[0082] The first training module 402 is used for training the preset variational autoencoder based on the two-dimensional data sample until the preset variational autoencoder converges, so as to obtain a target variational autoencoder, and the target variational autoencoder comprises an encoder and a decoder.

[0083] The replacement module 403 is configured to replace the decoder with a preset convolutional neural network to obtain an initial fault detection model.

[0084] The second training module 404 is configured to train the preset convolutional neural network in the initial fault detection model based on a target two-dimensional data sample obtained by performing noise processing on the two-dimensional data sample by using the encoder, until the preset convolutional neural network converges, to obtain the target fault detection model.

[0085] The detection module 405 is configured to perform fault detection on the metering automation master station based on target operation data of the metering automation master station by using the target fault detection model, to obtain fault result data of the metering automation master station.

[0086] In an embodiment, the preset variational autoencoder converges, and the method comprises:

[0087] A preset evidence lower bound ELBO function is used as a target loss function to perform gradient descent on each weight variable in the encoder and the decoder of the preset variational autoencoder, to update the weight variable, until the target loss function reaches a preset convergence condition.

[0088] In an embodiment, the ELBO function comprises a reconstruction loss function and a KL divergence loss function, and an expression of the ELBO function is:

[0089]

[0090] wherein, ELBO represents the ELBO function, reconstruction loss represents the reconstruction loss function, KL divergence loss represents the KL divergence loss function, and β represents a weight value of the KL divergence loss function.

[0091] In an embodiment, the weight value β of the KL divergence loss function is:

[0092]

[0093] wherein c is a constant, μ is a constant for preventing β from tending to infinity in an initial stage, i th epoch is the number of iterations, and k is a constant,

[0094] In an embodiment, the replacement module 403 is specifically configured to:

[0095] Lock the weight parameter of the encoder in the target variational autoencoder, and replace the decoder with the preset convolutional neural network to obtain the initial fault detection model.

[0096] In an embodiment, the target variational autoencoder comprises a convolutional layer and a fully connected layer, and the convolutional layer is configured to extract associated features between the two-dimensional data samples.

[0097] In an embodiment, the sampling module 401 is configured to:

[0098] determine a plurality of sampling time periods of the historical operation data according to a sampling frequency at which the metering automation master station collects the historical operation data;

[0099] randomly extract maximum values and minimum values of the historical operation data in the sampling time periods by using a random sampling window;

[0100] combine the maximum values and the minimum values corresponding to each sampling time period into the two-dimensional data samples.

[0101] The fault detection device of the metering automation master station described above can implement the fault detection method of the metering automation master station of the method embodiments described above. The options in the method embodiments described above are also applicable to this embodiment, and will not be described in detail here. The remaining contents of the embodiments of the present application can refer to the contents of the method embodiments described above, and will not be described in detail in this embodiment.

[0102] Figure 5 A structural schematic diagram of a computer device according to an embodiment of the present application is shown in FIG. 5. As shown in FIG. 5, the computer device 5 of this embodiment includes at least one processor 50 (only one processor is shown in the figure), a memory 51, and a computer program 52 stored in the memory 51 and executable on the at least one processor 50, wherein the processor 50 implements the steps in any of the method embodiments described above when executing the computer program 52. Figure 5 Figure 5 The computer device 5 can be a smart phone, a tablet computer, a desktop computer, a cloud server, and the like. The computer device can include but is not limited to the processor 50 and the memory 51. Those skilled in the art can understand that the computer device 5 shown in the figure is only an example and does not constitute a limitation on the computer device 5, which can include more or fewer components than those shown in the figure, or combine certain components, or different components, for example, can also include an input / output device, a network access device, and the like.

[0103] The computer device 5 can be a smart phone, a tablet computer, a desktop computer, a cloud server, and the like. The computer device can include but is not limited to the processor 50 and the memory 51. Those skilled in the art can understand that the computer device 5 shown in the figure is only an example and does not constitute a limitation on the computer device 5, which can include more or fewer components than those shown in the figure, or combine certain components, or different components, for example, can also include an input / output device, a network access device, and the like. Figure 5

[0104] ​​The processor 50 can be a central processing unit (CPU), and can also be other general-purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0105] The memory 51 can be an internal storage unit of the computer device 5 in some embodiments, for example, a hard disk or a memory of the computer device 5. The memory 51 can also be an external storage device of the computer device 5 in other embodiments, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Further, the memory 51 can include both an internal storage unit and an external storage device of the computer device 5. The memory 51 is used to store an operating system, an application program, a boot loader, data, and other programs, for example, program codes of the computer program, etc. The memory 51 can also be used to temporarily store data that has been output or will be output.

[0106] In addition, the embodiments of the present application further provide a computer readable storage medium, the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps in any method embodiment described above.

[0107] The embodiments of the present application provide a computer program product, when the computer program product runs on a computer device, the computer device executes the steps in each method embodiment described above.

[0108] In several embodiments provided in the present application, it can be understood that each block in the flowchart or block diagram can represent a module, a segment or a portion of code which includes one or more executable instructions for implementing the specified logic function. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figure. For example, two blocks noted in succession can in fact be executed substantially concurrently or in the reverse order, depending on the functionality involved.

[0109] The functions described above if implemented in software can be stored in a computer readable storage medium. Based on the understanding thus, the technical solutions of the present application essentially or partly or the part of the technical solutions that make contributions to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions that cause a computer device to execute all or part of the steps of the methods according to the embodiments of the present application. The foregoing storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, and various media that can store program codes.

[0110] The above specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only for specific embodiments of the present application and is not intended to limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A fault detection method of a metering automation master station, characterized by, The method comprises the following steps: determining a plurality of sampling time periods of the historical operation data according to a sampling frequency at which the metering automation master station collects the historical operation data; randomly extracting maximum values and minimum values of the historical operation data in the sampling time periods by using a random sampling window; combining the maximum values and the minimum values corresponding to each sampling time period into two-dimensional data samples; training a preset variational autoencoder based on the two-dimensional data samples until the preset variational autoencoder converges, so as to obtain a target variational autoencoder, wherein the target variational autoencoder comprises an encoder and a decoder; locking weight parameters of the encoder in the target variational autoencoder and replacing the decoder with a preset convolutional neural network to obtain an initial fault detection model; training the preset convolutional neural network in the initial fault detection model based on target two-dimensional data samples obtained by performing noise processing on the two-dimensional data samples by using the encoder until the preset convolutional neural network converges, so as to obtain a target fault detection model; performing fault detection on the metering automation master station according to target operation data of the metering automation master station by using the target fault detection model, so as to obtain fault result data of the metering automation master station.

2. The method for failure detection of a metrology automation master station of claim 1, wherein, The preset variational autoencoder converges, comprising: performing gradient descent on each weight variable in the encoder and the decoder of the preset variational autoencoder by using a preset evidence lower bound ELBO function as a target loss function, so as to update the weight variable until the target loss function reaches a preset convergence condition.

3. The fault detection method of a metrology automation master station of claim 2, wherein, The ELBO function comprises a reconstruction loss function and a KL divergence loss function, and an expression of the ELBO function is: wherein, represents the ELBO function, represents the reconstruction loss function, represents the KL divergence loss function, represents the weight value controlling the KL divergence loss function.

4. The fault detection method of a metrology automation master station according to claim 3, wherein, a weight value of the control kl divergence loss function is: wherein is a constant, to prevent a constant that tends to infinity in the initial phase, is the iteration number, is a constant, = .

5. The method for failure detection of a metrology automation host station of claim 1, wherein, The target variational autoencoder comprises a convolutional layer and a fully connected layer, and the convolutional layer is used to extract associated features between the two-dimensional data samples.

6. A fault detection apparatus of a metering automation master station, characterized by The method comprises the following steps: a sampling module is configured to determine a plurality of sampling time periods of historical operation data according to a sampling frequency at which a metering automation master station collects the historical operation data; a random sampling window is used to randomly extract maximum values and minimum values of the historical operation data in the sampling time periods; and the maximum values and the minimum values corresponding to each sampling time period are combined into two-dimensional data samples; a first training module is configured to train a preset variational autoencoder based on the two-dimensional data samples until the preset variational autoencoder converges, so as to obtain a target variational autoencoder, wherein the target variational autoencoder comprises an encoder and a decoder; a replacement module is configured to lock weight parameters of the encoder in the target variational autoencoder and replace the decoder with a preset convolutional neural network to obtain an initial fault detection model; a second training module is configured to train the preset convolutional neural network in the initial fault detection model based on target two-dimensional data samples obtained by performing noise processing on the two-dimensional data samples by using the encoder until the preset convolutional neural network converges, so as to obtain a target fault detection model; and A detection module is configured to utilize the target fault detection model to perform fault detection on the metering automation master station according to the target operation data of the metering automation master station, and obtain fault result data of the metering automation master station.

7. A computer device, comprising: The metering automation master station comprises a processor and a memory, the memory is configured to store a computer program, and the computer program is executed by the processor to implement the fault detection method of the metering automation master station according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer program is stored in the memory and is executed by the processor to implement the fault detection method of the metering automation master station according to any one of claims 1 to 5.

Images