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

Abstract

The application discloses a kind of fault prediction method, device, equipment and medium of metering automation master station, by the operation data of metering automation master station is carried out data cleaning, obtain target operation data, utilize convolutional neural network in target fault prediction model, the feature extraction of target operation data is carried out, obtain the feature sequence of target operation data, utilize long short-term memory network in target fault prediction model, according to the feature sequence, the fault prediction of metering automation master station is carried out, obtain the fault probability value of metering automation master station at future target time, the fault probability value is compared with preset fault threshold value, to determine the state information of metering automation master station at future target time, the state information is fault state or non-fault state, to improve the prediction accuracy by using the advantages of CNN and LSTM.

Description

TECHNICAL FIELD

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

[0002] The metering automation master station is a computer system connected with various metering automation terminals. It is the information collection and control center of the entire metering automation system. It collects and controls the information of the metering automation terminals through remote communication channels or subordinate system interfaces, and performs analysis, comprehensive processing and electric energy data release. In order to ensure the normal operation of the metering automation system, it is very important to perform fault early warning on the metering automation master station.

[0003] At present, the metering automation master station usually performs fault prediction based on traditional data mining technology, such as nearest neighbor algorithm, decision tree classifier or clustering method, to establish a fault prediction model of the master station system and perform fault prediction of the master station through the fault prediction model. However, due to the large scale of the metering automation master station and the large number of data sets, the traditional data mining technology has the problems of low prediction efficiency and low prediction accuracy. For example, the nearest neighbor algorithm has large calculation amount and low operation efficiency; the clustering method has slow convergence speed on large-scale data and may converge to a local minimum value; the decision tree classifier is difficult to extract the correlation between data and is prone to overfitting phenomenon. SUMMARY

[0004] The present application provides a fault prediction method, device, equipment and medium of metering automation master station to solve the technical problem of low fault prediction accuracy of the current metering automation master station.

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

[0006] performing data cleaning on the operation data of the metering automation master station to obtain target operation data;

[0007] performing feature extraction on the target operation data by using the convolutional neural network in the target fault prediction model to obtain a feature sequence of the target operation data;

[0008] performing fault prediction on the metering automation master station according to the feature sequence by using the long short-term memory network in the target fault prediction model to obtain a fault probability value of the metering automation master station at a future target time;

[0009] comparing the fault probability value with a preset fault threshold to determine the state information of the metering automation master station at the future target time, the state information being a fault state or a non-fault state.

[0010] As preferred, before the feature extraction of the target operation data by the convolutional neural network in the target fault prediction model, the method further comprises:

[0011] performing data cleaning on the historical operation data of the main station of the metering automation to obtain operation data samples;

[0012] oversampling the operation data samples to obtain target operation data samples;

[0013] sampling the target operation data samples based on a preset two-level time window to obtain a target sample set containing the two-level time window, wherein a first-level time window of the target sample set is used to record the target operation data samples, and a second-level time window is used to record a fault state label of the main station of the metering automation at a future time, and the first-level time window and the second-level time window have the same time length;

[0014] training a convolutional neural network and a long short-term memory network of a preset fault prediction model by using the target sample set until the preset fault prediction model converges, to obtain the target fault prediction model.

[0015] As preferred, the data cleaning on the historical operation data of the main station of the metering automation to obtain operation data samples comprises:

[0016] obtaining historical operation data collected by the main station of the metering automation;

[0017] performing outlier processing on the historical operation data to remove redundant data of the historical operation data to obtain the operation data samples.

[0018] As preferred, the oversampling of the operation data samples to obtain target operation data samples comprises:

[0019] classifying the operation data samples to obtain a positive sample set and a negative sample set;

[0020] generating a random number, and based on the random number, randomly selecting one positive sample from the positive sample set and one negative sample from the negative sample set;

[0021] generating a new negative sample according to the positive sample and the negative sample, and adding the new negative sample to the negative sample set until the number of samples in the positive sample set and the negative sample set is the same.

[0022] As preferred, the convolutional neural network in the target fault prediction model is used to perform feature extraction on the target operation data to obtain a feature sequence of the target operation data, including:

[0023] The convolutional neural network is used to convert the target operation data into a vector matrix;

[0024] Based on the multi-level convolutional layers of the convolutional neural network, a convolution operation is performed on the vector matrix to obtain a feature matrix output by each convolutional layer;

[0025] Based on the multi-level pooling layers of the convolutional neural network, feature aggregation is performed on the feature matrix output by each convolutional layer to obtain the feature sequence of the target operation data.

[0026] As preferred, the expression of the convolution operation is:

[0027] ;

[0028] wherein, is a feature matrix obtained after the convolution operation, is a weight matrix of the convolutional layer, is a vector matrix, is an offset vector, represents the convolution operation.

[0029] As preferred, the long short-term memory network in the target fault prediction model is used to perform fault prediction on the metering automation master station according to the feature sequence to obtain a fault probability value of the metering automation master station at a future target time, including:

[0030] The long short-term memory network is used to perform feature screening on the feature sequence to obtain a vector matrix;

[0031] Full connection and activation are performed on the vector matrix to obtain the fault probability value of the metering automation master station at the future target time.

[0032] In a second aspect, the present application provides a fault prediction device of a metering automation master station, including:

[0033] A cleaning module is configured to perform data cleaning on operation data of the metering automation master station to obtain target operation data;

[0034] An extraction module is configured to use a convolutional neural network in a target fault prediction model to perform feature extraction on the target operation data to obtain a feature sequence of the target operation data;

[0035] A prediction module is configured to utilize a long short-term memory network in the target fault prediction model to perform fault prediction on the metering automation master station according to the feature sequence, and obtain a fault probability value of the metering automation master station at a future target time.

[0036] A comparison module is configured to compare the fault probability value with a preset fault threshold to determine state information of the metering automation master station at the future target time, the state information being a fault state or a non-fault state.

[0037] 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 prediction method of the metering automation master station according to the first aspect when executed by the processor.

[0038] 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 prediction method of the metering automation master station according to the first aspect when executed by a processor.

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

[0040] The present application can reduce the data operation amount and improve the data processing efficiency by performing data cleaning on the operation data of the metering automation master station to obtain target operation data, can improve the feature extraction depth of the target operation data by utilizing the convolutional neural network in the target fault prediction model to extract features from the target operation data and obtain a feature sequence of the target operation data, can extract more important features from the target operation data and ensure the correlation between the time sequence features, can improve the accuracy of the fault probability value at the future time by utilizing the long short-term memory network in the target fault prediction model to perform fault prediction on the metering automation master station according to the feature sequence and obtain a fault probability value of the metering automation master station at a future target time, and can realize intelligent fault early warning based on the CNN and the LSTM, and improve the prediction accuracy by utilizing the advantages of the CNN and the LSTM. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 A flowchart of the fault prediction method of the metering automation master station according to the present application is shown in the figure;

[0042] Figure 2 A flowchart of the fault prediction method of the metering automation master station according to the present application is shown in the figure;

[0043] Figure 3 The structure of the fault prediction device of the metering automation master station is shown in the embodiment of the present application.

[0044] Figure 4 The structure of the fault prediction device of the metering automation master station is shown in the embodiment of the present application.

[0045] Figure 5 The structure of the computer device is shown in the embodiment of the present application. DETAILED DESCRIPTION

[0046] 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. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0047] Please refer to Figure 1 , Figure 1 The flowchart of the fault prediction method of the metering automation master station is provided in the embodiment of the present application. The fault prediction method of the metering automation master station in the embodiment of the present application can be applied to a computer device, which includes but is not limited to a smart phone, a notebook computer, a tablet computer, a desktop computer, a physical server, a cloud server and the like. As shown in Figure 1 The fault prediction method of the metering automation master station in the embodiment includes steps S101 to S104, which are described in detail as follows.

[0048] Step S101, data cleaning is performed on the running data of the metering automation master station to obtain target running data.

[0049] In this step, the running data includes but is not limited to data flow, server CPU occupancy, TCP connection number and the like. Since the related running data generated by the metering automation master station has a large data volume and a large amount of redundant data, the running data is cleaned. Based on the running data at the abnormal network, the data points will present outlying data. The data preprocessing method of eliminating redundant non-outlying points is adopted in the present application to reduce the data volume. The definition of the outlying point is a point whose distance from most points exceeds a certain specific value.

[0050] Optionally, normal points are first aggregated and sorted according to data density. Similar points in the smaller sets formed by the dataset are removed, and a point is selected, with its distances to other points calculated. If this distance is less than the radius of the set's center, the point belongs to that set; otherwise, it becomes the new set center. Sets with fewer than k points are retained, and all other sets are deleted. Based on these operations, some redundant non-outlier points can be eliminated, reducing the amount of data the fault prediction model needs to process and improving prediction efficiency.

[0051] Step S102: Using the convolutional neural network in the target fault prediction model, feature extraction is performed on the target running data to obtain the feature sequence of the target running data.

[0052] In this step, such as Figure 2 The diagram illustrates the target fault prediction model. In this embodiment, the Convolutional Neural Network (CNN) includes convolutional layers and pooling layers for feature extraction from the running data. The pooling layers can be K-max pooling layers. Optionally, the CNN's convolutional and pooling layers extract features from the running data through convolution and pooling operations, respectively, to obtain the most important feature information from the running data.

[0053] In one embodiment, step S102 includes:

[0054] The target running data is converted into a vector matrix using the convolutional neural network.

[0055] Based on the multi-level convolutional layers of the convolutional neural network, the vector matrix is ​​convolved to obtain the feature matrix output by each level of the convolutional layer;

[0056] Based on the multi-level pooling layers of the convolutional neural network, feature aggregation is performed on the feature matrix output by each convolutional layer to obtain the feature sequence of the target running data.

[0057] In this embodiment, the CNN employs local connections and weight sharing, which reduces the complexity of the neural network model and the number of weights. Alternating processing of the original data through convolutional and pooling layers automatically extracts local features and establishes corresponding feature vectors. The feature extraction process consists of the following three steps:

[0058] (1) Embedding Layer (Input Layer). The processing data needs to be input from the embedding layer and vectorized into a matrix before being fed into the network. The length of the processing data is limited to m, and each data point is transformed into an n-dimensional data vector. The output of the embedding layer is a... Two-dimensional matrix ,in, It is data The data vector.

[0059] (2) Convolutional Layers. Convolutional layers are used to extract features, with each convolutional kernel corresponding to a specific part of the extracted features. The convolutional kernels output the matrix of the embedding layer. Perform the following convolution operation:

[0060] ;

[0061] in, This is the feature matrix obtained after the convolution operation. Here is the weight matrix of the convolutional layer. It is a vector matrix. For offset vectors, This indicates a convolution operation. and The parameters are learned through network learning. In the convolution operation, the filter weights remain unchanged, reducing the number of model parameters that need to be learned, thus simplifying the neural network structure. After the convolutional layer, the activation function is added to the output of the convolutional layer, playing a non-linear role in the output of the neural network's convolutional layer. Optionally, the activation function used in this embodiment is:

[0062] ;

[0063] Among them, the ReLU function has the characteristics of fast convergence speed and simple gradient calculation.

[0064] (3) Pooling layer. After the convolution operation, the extracted feature matrix is ​​fed into the K-max pooling layer. The pooling layer further aggregates the features and selects the K most significant features to simplify the feature representation.

[0065] Step S103: Using the long short-term memory network in the target fault prediction model, based on the feature sequence, perform fault prediction on the metering automation master station to obtain the fault probability value of the metering automation master station at a future target time.

[0066] In this step, such as Figure 2 As shown, the feature sequences of a CNN are used as input to an LSTM to predict faults. Long Short-Term Memory (LSTM) networks are an improved method based on recurrent neural networks, offering advantages in learning long-term dependencies. This embodiment uses LSTM to predict the probability of a fault occurring in the future based on data accumulated over a certain time period. LSTM can address the problems of gradient vanishing and long-term memory loss that exist in long-sequence data processing.

[0067] In one embodiment, step S103 includes:

[0068] The long short-term memory network is used to perform feature screening on the feature sequence to obtain a vector matrix.

[0069] The vector matrix is fully connected and activated to obtain a fault probability value of the metering automation master station at a future target time.

[0070] In the embodiment, as Figure 3 The network structure diagram of the LSTM is shown. Optionally, to prevent overfitting of data, the vector matrix output by the LSTM model is input to a dropout layer for feature screening. The feature screening process is as follows: the basic unit of the LSTM network includes a forgetting gate, an input gate, and an output gate. As shown in FIG. The forgetting gate is The input gate is The output gate is The sigmoid activation function is used. The input of the forgetting gate is , the intermediate input is and the state storage unit together determine which information needs to be forgotten by the state storage unit. On the other hand, the of the input gate is and the tanh function, and the output thereof is used to determine which information should be retained in the state storage unit. Based on the structure of the input gate and the forgetting gate, the LSTM network can more effectively determine which information should be retained and which information should be forgotten. Subsequently, the and the output are updated, and the intermediate output is obtained based on the two to complete a complete updating process. Exemplarily, the calculation formula is shown in the formula.

[0071] ;

[0072] ;

[0073] ;

[0074] ;

[0075] ;

[0076] ;

[0077] ;

[0078] ;

[0079] wherein , , respectively are weight values corresponding to the forget gate, the input gate, the output gate and the state storage unit respectively, respectively are offset values corresponding to the state gate. is an activation function.

[0080] Optionally, the vector matrix is input to a fully connected layer for dimension reduction processing to reduce the vector dimension, and then an activation function is used to process the output of the fully connected layer to obtain the probability of the prediction result

[0081] In step S104, the fault probability value is compared with a preset fault threshold to determine the state information of the metering automation master station at a future target time, the state information being a fault state or a non-fault state.

[0082] In this step, when the prediction probability is greater than the threshold, it is considered that the metering automation master station will fail; otherwise, it is considered that the metering automation master station will not fail.

[0083] In an embodiment, based on the embodiment shown in Figure 1 The method further includes:

[0084] Data cleaning is performed on historical operation data of the metering automation master station to obtain operation data samples;

[0085] The operation data samples are oversampled to obtain target operation data samples;

[0086] Based on a preset two-level time window, the target operation data samples are sampled to obtain a target sample set containing the two-level time window, wherein the first-level time window of the target sample set is used to record the target operation data samples, and the second-level time window is used to record the fault state label of the metering automation master station at a future time, and the time length of the first-level time window is the same as that of the second-level time window;

[0087] The convolutional neural network and the long short-term memory network of a preset fault prediction model are trained using the target sample set until the preset fault prediction model converges, and the target fault prediction model is obtained.

[0088] ​​​​​​​In the present embodiment, in the fault early warning process of the metering automation master station, network congestion, interface damage and other faults are relatively typical fault types. Therefore, first, relevant data such as data flow, server CPU occupancy, and TCP connection number need to be collected. Since the relevant network data generated by the metering automation master station system has the characteristics of large data volume, many redundant data, and few abnormal data, and the data is incomplete, it is necessary to pre-process the data collected by the metering automation master station based on the elimination of redundant non-abnormal data points and oversampling scheme. Subsequently, two-level time windows are used to extract sample information. The first level window is used to record the current running data of the metering automation master station system and serve as the input sequence of the prediction model. The second level window is used to record the future communication network security state to be predicted and serve as the label. The lengths of the two-level windows should be consistent. The data after the data preprocessing process is input into the preset fault prediction model based on CNN-LSTM for training until the loss function of the preset fault prediction model is less than the preset threshold, or the iteration number of the training process reaches the preset upper limit number of iterations, then the preset fault prediction model converges, and a target fault prediction model is obtained.

[0089] Optionally, the data cleaning on the historical running data of the metering automation master station to obtain the running data sample comprises:

[0090] Obtaining historical running data collected by the metering automation master station;

[0091] Performing outlier processing on the historical running data to remove redundant data of the historical running data to obtain the running data sample.

[0092] In the present embodiment, the data cleaning is the same as the data cleaning process of step S101, and the relevant description of step S101 can be referred to.

[0093] Optionally, the oversampling on the running data sample to obtain a target running data sample comprises:

[0094] Classifying the running data sample to obtain a positive sample set and a negative sample set;

[0095] Generating a random number, and based on the random number, randomly extracting a positive sample from the positive sample set and a negative sample from the negative sample set;

[0096] Generating a new negative sample according to the positive sample and the negative sample, and adding the new negative sample to the negative sample set until the sample numbers of the positive sample set and the negative sample set are the same.

[0097] In the embodiment, since the metering automation master station is in normal operation state most of the time, the abnormal data information collected is very limited, which is not conducive to the training of the fault prediction model. Based on this, the number of normal samples and abnormal samples is balanced by using oversampling technology. Oversampling can effectively solve the problem that the computer cannot learn the characteristics of abnormal information due to data imbalance.

[0098] Optionally, the collected data is divided into a positive (normal) sample set and a negative (abnormal) sample set. At this time, the data amount in the negative sample set is much smaller than that in the positive sample set. Then, a random number is generated, a sample in a negative sample is randomly extracted and recorded as x, a positive sample is randomly extracted and recorded as y, and a new negative sample is generated. Repeat this step until the sum of the element numbers of the negative sample set and the newly generated negative sample set is equal to the element number of the positive sample set, and the oversampling ends.

[0099] It should be noted that, compared with the prior art, the present application has the following advantages:

[0100] (1) The metering automation master station fault prediction method based on the convolutional neural network and the long short-term memory network is safer, faster and more predictive than the daily routine inspection work relying only on manual work.

[0101] (2) By removing part of the redundant non-abnormal information and oversampling, the problem of large amount of data collected by the metering automation master station, multiple redundant data, and few abnormal data, and multiple damaged data can be solved.

[0102] (3) Compared with using CNN alone to train the model or using random forest to predict the model, the fault prediction model based on CNN-LSTM has higher prediction accuracy.

[0103] In order to perform the metering automation master station fault prediction method corresponding to the above method embodiment to realize the corresponding functions and technical effects. Referring to Figure 4 , Figure 4 A structure block diagram of a metering automation master station fault prediction device provided by an embodiment of the present application is shown. For ease of illustration, only the part related to the present embodiment is shown. The metering automation master station fault prediction device provided by the embodiment of the present application comprises:

[0104] The cleaning module 401 is configured to clean the operation data of the metering automation master station to obtain target operation data.

[0105] The extraction module 402 is configured to use the convolutional neural network in the target fault prediction model to extract features of the target operation data to obtain a feature sequence of the target operation data.

[0106] The prediction module 403 is configured to perform fault prediction on the metering automation master station according to the feature sequence by using a long short-term memory network in the target fault prediction model, to obtain a fault probability value of the metering automation master station at a target future time.

[0107] The comparison module 404 is configured to compare the fault probability value with a preset fault threshold, to determine state information of the metering automation master station at the target future time, the state information being a fault state or a non-fault state.

[0108] In an embodiment, the fault prediction apparatus further comprises:

[0109] The cleaning module is configured to clean historical operation data of the metering automation master station, to obtain an operation data sample.

[0110] The oversampling module is configured to oversample the operation data sample, to obtain a target operation data sample.

[0111] The sampling module is configured to sample the target operation data sample based on a preset two-level time window, to obtain a target sample set containing the two-level time window, wherein a first-level time window of the target sample set is configured to record the target operation data sample, and a second-level time window is configured to record a fault state label of the metering automation master station at a future time, and the first-level time window and the second-level time window have the same time length.

[0112] The training module is configured to train a convolutional neural network and a long short-term memory network of a preset fault prediction model by using the target sample set, until the preset fault prediction model converges, to obtain the target fault prediction model.

[0113] In an embodiment, the cleaning module is specifically configured to:

[0114] Obtain historical operation data collected by the metering automation master station.

[0115] Perform outlier processing on the historical operation data, to remove redundant data of the historical operation data, to obtain the operation data sample.

[0116] In an embodiment, the oversampling module is specifically configured to:

[0117] Classify the operation data sample, to obtain a positive sample set and a negative sample set.

[0118] Generate a random number, and randomly select one positive sample from the positive sample set and one negative sample from the negative sample set based on the random number.

[0119] A new negative sample is generated according to the positive sample and the negative sample, and the new negative sample is added to the negative sample set until the number of samples in the positive sample set and the negative sample set is the same.

[0120] In an embodiment, the extraction module 402 is specifically configured to:

[0121] The target operation data is converted into a vector matrix by using the convolutional neural network.

[0122] The vector matrix is subjected to a convolution operation based on the multi-level convolutional layer of the convolutional neural network, to obtain a feature matrix output by each level of convolutional layer.

[0123] The feature matrix output by each level of convolutional layer is subjected to feature aggregation based on the multi-level pooling layer of the convolutional neural network, to obtain a feature sequence of the target operation data.

[0124] In an embodiment, the expression of the convolution operation is as follows:

[0125] ;

[0126] wherein, is a feature matrix obtained after the convolution operation, is a weight matrix of the convolutional layer, is a vector matrix, is an offset vector, represents the convolution operation.

[0127] In an embodiment, the prediction module 403 is specifically configured to:

[0128] The feature sequence is subjected to feature screening by using the long short-term memory network, to obtain a vector matrix.

[0129] The vector matrix is subjected to full connection and activation, to obtain a fault probability value of the metering automation master station at a future target time.

[0130] The fault prediction device of the metering automation master station described above can implement the fault prediction method of the metering automation master station of the method embodiment described above. The options in the method embodiment described above are also applicable to the present embodiment, and will not be described in detail herein. The remaining content of the embodiment of the present application can be referred to the content of the method embodiment described above, and will not be described in detail herein.

[0131] Figure 5 A structural schematic diagram of a computer device provided by an embodiment of the present application is shown in FIG. 5. As shown in FIG. 5, the computer device 5 of the embodiment includes at least one processor 50 (e.g., a central processing unit (CPU), a microprocessor, a microcomputer, a microcontroller, or the like). Figure 5 Figure 5 ​(Only one is shown in the diagram) a processor, 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 executes the computer program 52 to implement the steps in any of the above method embodiments.

[0132] The computer device 5 may be a smartphone, tablet, desktop computer, or cloud server, among other computing devices. This computer device may include, but is not limited to, a processor 50 and a memory 51. Those skilled in the art will understand that... Figure 5 The computer device 5 is merely an example and does not constitute a limitation on the computer device 5. It may include more or fewer components than shown, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0133] The processor 50 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0134] In some embodiments, the memory 51 may be an internal storage unit of the computer device 5, such as a hard disk or memory of the computer device 5. In other embodiments, the memory 51 may be an external storage device of the computer device 5, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 5. Furthermore, the memory 51 may include both internal and external storage units of the computer device 5. The memory 51 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory 51 can also be used to temporarily store data that has been output or will be output.

[0135] In addition, the embodiment of the present application further provides a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the steps in any method embodiment described above.

[0136] The embodiment of the present application provides a computer program product. When the computer program product is run on a computer device, the computer device is caused to execute the steps in each method embodiment described above.

[0137] In several embodiments provided by 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 implementation, the functions noted in the blocks can occur in a different order than that noted in the figure. For example, two blocks which are consecutive in the figure can actually be executed substantially in parallel, or they can be executed in reverse order, depending on the functions involved.

[0138] If the functions are implemented in the form of software function modules and sold or used as independent products, the software function modules can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or partially contribute to the prior art, or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device to execute all or part of the steps of the methods described in the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media which can store program codes.

[0139] 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 should be included in the protection scope of the present application.

Claims

1. A fault prediction method for an automated metering master station, characterized in that, include: The operational data of the metering automation master station is cleaned to obtain target operational data; wherein, the target operational data includes: data traffic, server CPU utilization, and TCP connection data; Using a convolutional neural network in the target fault prediction model, the target running data is converted into a vector matrix; based on the multi-level convolutional layers of the convolutional neural network, the vector matrix is ​​convolved to obtain the feature matrix output by each level of the convolutional layer; based on the multi-level pooling layers of the convolutional neural network, the feature matrix output by each level of the convolutional layer is aggregated to obtain the feature sequence of the target running data. Using the long short-term memory network in the target fault prediction model, feature filtering is performed on the feature sequence to obtain a vector matrix; the vector matrix is ​​fully connected and activated to obtain the fault probability value of the metering automation master station at the future target time. The fault probability value is compared with a preset fault threshold to determine the status information of the metering automation master station at a future target time, wherein the status information is a fault state or a non-fault state; Before extracting features from the target operating data using the convolutional neural network in the target fault prediction model to obtain the feature data of the target operating data, the method further includes: Data cleaning was performed on the historical operational data of the metering automation master station to obtain operational data samples; The running data sample is oversampled to obtain the target running data sample; Based on a preset two-level time window, the target operation data sample is sampled to obtain a target sample set containing two-level time windows. The first-level time window of the target sample set is used to record the target operation data sample, and the second-level time window is used to record the fault status label of the metering automation master station at a future time. The first-level time window and the second-level time window have the same time length. Using the target sample set, the convolutional neural network and long short-term memory network of the preset fault prediction model are trained until the preset fault prediction model converges, thus obtaining the target fault prediction model; The process of cleaning historical operational data from the automated metering station to obtain operational data samples includes: Acquire historical operational data collected by the automated metering master station; Outlier processing is performed on the historical operation data to remove redundant data and obtain the operation data sample.

2. The fault prediction method for the metering automation master station as described in claim 1, characterized in that, The process of oversampling the running data sample to obtain the target running data sample includes: The operational data samples are classified to obtain a positive sample set and a negative sample set; Generate a random number, and based on the random number, randomly select a positive sample from the positive sample set and a negative sample from the negative sample set; New negative samples are generated based on the positive and negative samples, and these new negative samples are added to the negative sample set until the number of samples in the positive and negative sample sets are the same.

3. The fault prediction method for the metering automation master station as described in claim 2, characterized in that, The expression for the convolution operation is: ； in, This is the feature matrix obtained after the convolution operation. Here is the weight matrix of the convolutional layer. It is a vector matrix. For offset vectors, This indicates a convolution operation.

4. A fault prediction device for an automated metering master station, characterized in that, include: The cleaning module is used to clean the operating data of the metering automation master station to obtain target operating data; wherein, the target operating data includes: data traffic, server CPU utilization, and TCP connection data; The extraction module is used to convert the target running data into a vector matrix using the convolutional neural network in the target fault prediction model; perform convolution operations on the vector matrix based on the multi-level convolutional layers of the convolutional neural network to obtain the feature matrix output by each level of the convolutional layer; and perform feature aggregation on the feature matrix output by each level of the convolutional layer based on the multi-level pooling layers of the convolutional neural network to obtain the feature sequence of the target running data. The prediction module is used to use the long short-term memory network in the target fault prediction model to perform feature filtering on the feature sequence to obtain a vector matrix; and to perform full connection and activation on the vector matrix to obtain the fault probability value of the metering automation master station at a future target time. The comparison module is used to compare the fault probability value with a preset fault threshold to determine the status information of the metering automation master station at a future target time, wherein the status information is a fault state or a non-fault state; Before extracting features from the target operating data using the convolutional neural network in the target fault prediction model to obtain the feature data of the target operating data, the method further includes: Data cleaning was performed on the historical operational data of the metering automation master station to obtain operational data samples; The running data sample is oversampled to obtain the target running data sample; Based on a preset two-level time window, the target operation data sample is sampled to obtain a target sample set containing two-level time windows. The first-level time window of the target sample set is used to record the target operation data sample, and the second-level time window is used to record the fault status label of the metering automation master station at a future time. The first-level time window and the second-level time window have the same time length. Using the target sample set, the convolutional neural network and long short-term memory network of the preset fault prediction model are trained until the preset fault prediction model converges, thus obtaining the target fault prediction model; The process of cleaning historical operational data from the automated metering station to obtain operational data samples includes: Acquire historical operational data collected by the automated metering master station; Outlier processing is performed on the historical operation data to remove redundant data and obtain the operation data sample.

5. A computer device, characterized in that, It includes a processor and a memory, the memory being used to store a computer program, which, when executed by the processor, implements the fault prediction method for the metering automation master station as described in any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the fault prediction method for the metering automation master station as described in any one of claims 1 to 3.

Images