Complementation method for missing value of complex equipment state monitoring data based on adaptive time sequence diagram convolutional network

Abstract

The invention discloses a complex equipment state monitoring data missing value complementing method based on an adaptive time sequence diagram convolutional network, and relates to the technical field of industrial data processing, and the method comprises the steps: obtaining incomplete state monitoring data, carrying out the preprocessing of the incomplete state monitoring data, and generating a mask matrix according to the incomplete state monitoring data, then inputting the two features into a trained mask auto-encoder time sequence interpolation model, extracting spatial-temporal features of monitoring data in an incomplete state through an adaptive graph convolutional network layer, extracting time sequence features of a mask matrix through a second bidirectional gating cycle unit, and fusing the two features to obtain a mask auto-encoder time sequence interpolation model; and decoding and outputting through a decoder to obtain complemented state monitoring data. According to the scheme, the adaptive graph convolutional network is used for operation along the time dimension, the spatio-temporal characteristics of the data can be fully mined, and the bidirectional gating cycle unit is used for mining the time sequence characteristics from the mask data so as to reconstruct the missing data and complete the missing value.

Description

Technical Field

[0001] The present application relates to the technical field of industrial data processing, and particularly to a method for filling missing values in complex equipment status monitoring data based on an adaptive temporal graph convolutional network. Background Art

[0002] With the continuous progress of intelligent manufacturing, during the operation of complex equipment, a large amount of status monitoring data is collected through various sensors. This data contains key information during equipment operation and is crucial for equipment status monitoring, fault diagnosis, and predictive maintenance. However, in actual industrial applications, due to various factors such as sensor failures, signal transmission interference, and abnormalities in data acquisition system hardware or software, missing values often appear in the status monitoring data. These missing values not only reduce the integrity of the data but may also lead to an increase in errors in subsequent analysis and processing, affecting the accurate assessment of the status of complex equipment and further affecting the formulation of key decisions. Therefore, effectively filling these missing data is of great significance for improving equipment health management capabilities, enhancing the accuracy of fault diagnosis and maintenance, and strengthening equipment safety.

[0003] In current research, various neural networks in deep learning have been widely applied to the task of filling missing values in status monitoring data. In particular, models based on graph neural networks have shown excellent performance in capturing the spatial dependence relationships of data. Although these methods have achieved certain results in filling missing values, they still have limitations. Most graph neural network-based methods rely on prior knowledge to define the graph structure through node connectivity or geographical location to determine the spatial dependence between nodes. The graph structure constructed by this method highly depends on prior knowledge and is difficult to apply to specific fields. In addition, existing graph structures mainly model feature nodes and cannot well adapt to temporal data.

[0004] In view of this, the present application proposes a new method for filling missing values in complex equipment status monitoring data, aiming to be widely applicable to different industrial fields and effectively solve the problem of filling missing values in complex equipment status monitoring data. Summary of the Invention

[0005] The purpose of the present application is to provide a method for filling missing values in complex equipment status monitoring data based on an adaptive temporal graph convolutional network, which can be widely applied to filling missing values in complex equipment status monitoring data in different industrial fields and has good innovation and practicality.

[0006] To achieve the above purpose, the present application provides the following solution:

[0007] A method for filling missing values in complex equipment status monitoring data based on an adaptive temporal graph convolutional network, comprising the following steps:

[0008] Obtain incomplete state monitoring data; the incomplete state monitoring data is the state monitoring data with incomplete values collected during the operation of complex equipment.

[0009] Preprocess the incomplete state monitoring data to obtain the preprocessed incomplete state monitoring data.

[0010] Generate a mask matrix based on the preprocessed incomplete state monitoring data; the mask matrix is used to characterize the distribution of missing values in the preprocessed incomplete state monitoring data.

[0011] Input the preprocessed incomplete state monitoring data and the mask matrix into the trained mask autoencoder time series imputation model to obtain the completed state monitoring data; the mask autoencoder time series imputation model includes an adaptive graph convolutional network layer and a first bidirectional gated recurrent unit for extracting the spatio-temporal features of the incomplete state monitoring data, a second bidirectional gated recurrent unit for extracting the time series features of the mask matrix, a data fusion layer for fusing the spatio-temporal features of the incomplete state monitoring data and the time series features of the mask matrix, and a decoder for decoding and outputting based on the fused spatio-temporal features obtained by the data fusion layer.

[0012] Optionally, before inputting the preprocessed incomplete state monitoring data and the mask matrix into the trained mask autoencoder time series imputation model to obtain the completed state monitoring data, the method for completing missing values in complex equipment state monitoring data based on an adaptive time series graph convolutional network further includes the following steps:

[0013] Obtain a historical state monitoring data set; the historical state monitoring data set includes several complete state monitoring data, and the complete state monitoring data is the state monitoring data without incomplete values collected during the operation of complex equipment.

[0014] For any complete state monitoring data, preprocess the complete state monitoring data to obtain the preprocessed complete state monitoring data; all the preprocessed complete state monitoring data constitutes the preprocessed historical state monitoring data set.

[0015] Construct a mask autoencoder time series imputation model based on an adaptive graph convolutional network and several bidirectional gated recurrent units.

[0016] Divide the preprocessed historical state monitoring data set into a training data set, a test data set, and a validation data set; the training data set is used to train and optimize the weight parameters of the mask autoencoder time series imputation model, the validation data set is used to adjust the model hyperparameters, and the test data set is used to test and evaluate the performance of the trained mask autoencoder time series imputation model.

[0017] For any preprocessed complete condition monitoring data, use a randomly generated mask matrix to perform masking on the preprocessed complete condition monitoring data to obtain masked condition monitoring data; the masked condition monitoring data, the mask matrix, and the preprocessed complete condition monitoring data form a sample.

[0018] Use the training dataset to train the masked autoencoder time series imputation model, optimize the weights of the masked autoencoder time series imputation model, and obtain the trained masked autoencoder time series imputation model.

[0019] Optionally, use the training dataset to train the masked autoencoder time series imputation model, optimize the weights of the masked autoencoder time series imputation model, and obtain the trained masked autoencoder time series imputation model, which specifically includes the following steps:

[0020] For any sample in the training dataset, input the masked condition monitoring data and the mask matrix into the masked autoencoder time series imputation model to obtain the complemented condition monitoring data.

[0021] According to the complemented condition monitoring data and the preprocessed complete condition monitoring data, calculate the loss function value of the masked autoencoder time series imputation model.

[0022] According to the loss function of the masked autoencoder time series imputation model, optimize the weight parameters of the masked autoencoder time series imputation model.

[0023] Optionally, the loss function of the masked autoencoder time series imputation model selects the mean absolute error MAE.

[0024] Optionally, use the Adam optimization algorithm to optimize the weight parameters of the masked autoencoder time series imputation model, set the learning rate to 0.001, and set the number of training rounds to 100.

[0025] Optionally, the masked autoencoder time series imputation model includes an encoder and a decoder; the encoder includes an adaptive graph convolutional network layer, a first bidirectional gated recurrent unit, a second bidirectional gated recurrent unit, and a data fusion layer; the adaptive graph convolutional network layer and the first bidirectional gated recurrent unit are used to extract the spatio-temporal features of the incomplete condition monitoring data; the second bidirectional gated recurrent unit is used to extract the time series features of the mask matrix; the data fusion layer is used to fuse the spatio-temporal features of the incomplete condition monitoring data and the time series features of the mask matrix to obtain the fused spatio-temporal features; the decoder includes a third bidirectional gated recurrent unit, a first fully connected network layer, a second fully connected network layer, and an output layer; the third bidirectional gated recurrent unit, the first fully connected network layer, and the second fully connected network layer are used to reconstruct the complemented condition monitoring data according to the fused spatio-temporal features and output the complemented condition monitoring data through the output layer.

[0026] Optionally, an adaptive graph convolutional network layer is used to extract the spatial features of the incomplete status monitoring data, and further, a first bidirectional gated recurrent unit is used to extract the temporal features of the incomplete status monitoring data, so as to realize the extraction of the spatio-temporal features of the incomplete status monitoring data.

[0027] Optionally, the adaptive graph convolutional network layer is used to operate along the time dimension, allocate a learnable parameter space and a node embedding dictionary for each time node within the current time step, define a node graph structure according to the similarity between the previous node embedding dictionaries, and learn the parameter space and the node graph structure of each time node during the training process to capture the spatial dependence relationships among the time nodes within the time step.

[0028] Optionally, a data fusion layer is used to splice the spatio-temporal features of the incomplete status monitoring data and the temporal features of the mask matrix along the feature dimension to fuse the two types of feature information and obtain fused spatio-temporal features.

[0029] Optionally, the incomplete status monitoring data is preprocessed according to the following formula:

[0030]

[0031] wherein, X norm is the preprocessed incomplete status monitoring data, X is the incomplete status monitoring data, and X min and X max are the minimum value and the maximum value of the incomplete status monitoring data respectively.

[0032] According to the specific embodiments provided by this application, the following technical effects are disclosed by this application:

[0033] The present application provides a method for completing missing values ​​of complex equipment status monitoring data based on an adaptive temporal graph convolutional network. In the method, after preprocessing the acquired incomplete status monitoring data, a mask matrix is ​​generated accordingly, and then the preprocessed incomplete status monitoring data and the mask matrix are input into a trained masked autoencoder timing interpolation model. In the masked autoencoder timing interpolation model, the spatiotemporal features of the incomplete status monitoring data are extracted by an adaptive graph convolutional network layer and a first bidirectional gated recurrent unit, and the timing features of the mask matrix are extracted by a second bidirectional gated recurrent unit. Then, the spatiotemporal features of the incomplete status monitoring data and the timing features of the mask matrix are fused by a data fusion layer. Finally, the decoder decodes and outputs the fused spatiotemporal features obtained by the data fusion layer to obtain the completed status monitoring data. The above-mentioned scheme of the present application uses an adaptive graph convolutional network to operate along the time dimension without introducing prior knowledge, which can fully mine the spatiotemporal characteristics of the data, and use a bidirectional gated recurrent unit to mine the timing characteristics from the mask data to reconstruct the missing part of the data and complete the missing value completion. It can be widely used in the missing value completion task of complex equipment status monitoring data in different industrial fields, and has good innovation and practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1 A flowchart of a method for completing missing values ​​in complex equipment status monitoring data based on an adaptive temporal graph convolutional network is provided in accordance with an embodiment of the present application.

[0036] Figure 2 A flowchart of obtaining historical data to train and optimize a model in a method for completing missing values ​​in complex equipment status monitoring data based on an adaptive temporal graph convolutional network provided in one embodiment of the present application.

[0037] Figure 3 A flowchart of step B6 in a method for completing missing values ​​of complex equipment status monitoring data based on an adaptive temporal graph convolutional network provided in one embodiment of the present application.

[0038] Figure 4 A structural schematic diagram of a masked autoencoder time series interpolation model in a method for completing missing values ​​of complex equipment status monitoring data based on an adaptive time series graph convolutional network provided in one embodiment of the present application.

[0039] Figure 5 Schematic diagram for comparing the imputed value and the true value of oil temperature data in a method for filling missing values in complex equipment status monitoring data based on an adaptive temporal graph convolutional network provided by an embodiment of the present application.

[0040] Figure 6 Schematic diagram of the functional modules of a system for filling missing values in complex equipment status monitoring data based on an adaptive temporal graph convolutional network provided by an embodiment of the present application.

[0041] Figure 7 Schematic diagram of the structure of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0042] Next, the technical solutions in the embodiments of the present application will be clearly and completely described with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a 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 of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0043] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below with reference to the accompanying drawings and specific implementation manners.

[0044] In an exemplary embodiment, as Figure 1 shown, a method for filling missing values in complex equipment status monitoring data based on an adaptive temporal graph convolutional network is provided, including the following steps:

[0045] A1. Obtain incomplete status monitoring data; the incomplete status monitoring data is status monitoring data with incomplete values collected during the operation of complex equipment.

[0046] A2. Preprocess the incomplete status monitoring data to obtain the preprocessed incomplete status monitoring data. Specifically, in this embodiment, the incomplete status monitoring data is preprocessed according to the following formula:

[0047]

[0048] where, X norm is the preprocessed incomplete status monitoring data, X is the incomplete status monitoring data, X min and X max are the minimum value and the maximum value of the incomplete status monitoring data respectively.

[0049] A3. Generate a mask matrix according to the preprocessed incomplete status monitoring data; the mask matrix is used to represent the distribution of missing values in the preprocessed incomplete status monitoring data.

[0050] A4. Input the preprocessed incomplete condition monitoring data and the mask matrix into the trained mask autoencoder time series imputation model to obtain the completed condition monitoring data. The mask autoencoder time series imputation model includes an adaptive graph convolutional network layer and a first bidirectional gated recurrent unit for extracting the spatio-temporal features of the incomplete condition monitoring data, a second bidirectional gated recurrent unit for extracting the time series features of the mask matrix, a data fusion layer for fusing the spatio-temporal features of the incomplete condition monitoring data and the time series features of the mask matrix, and a decoder for decoding and outputting based on the fused spatio-temporal features obtained by the data fusion layer.

[0051] In another exemplary embodiment, before step A4, the method for completing missing values in complex equipment condition monitoring data based on an adaptive time series graph convolutional network further includes a process of obtaining historical data to train and optimize the model, as Figure 2 shown, and this process includes the following steps:

[0052] B1. Obtain a historical condition monitoring data set. The historical condition monitoring data set includes several complete condition monitoring data, and the complete condition monitoring data is the condition monitoring data collected during the operation of the complex equipment without incomplete values.

[0053] B2. For any complete condition monitoring data, preprocess the complete condition monitoring data to obtain the preprocessed complete condition monitoring data. All the preprocessed complete condition monitoring data constitutes the preprocessed historical condition monitoring data set.

[0054] B3. Based on an adaptive graph convolutional network and several bidirectional gated recurrent units, construct a mask autoencoder time series imputation model. In this embodiment, as Figure 3 shown, the mask autoencoder time series imputation model includes an encoder and a decoder. The encoder includes an adaptive graph convolutional network layer, a first bidirectional gated recurrent unit, a second bidirectional gated recurrent unit, and a data fusion layer. The adaptive graph convolutional network layer is used to extract the spatial features of the incomplete condition monitoring data, and further extract the time series features of the incomplete condition monitoring data through the first bidirectional gated recurrent unit to achieve the extraction of the spatio-temporal features of the incomplete condition monitoring data. The second bidirectional gated recurrent unit is used to extract the time series features of the mask matrix.

[0055] Specifically, the adaptive graph convolutional network layer is used to operate along the time dimension, allocate a learnable parameter space and a node embedding dictionary for each time node within the current time step, define the node graph structure according to the similarity between the previous node embedding dictionaries, and learn the parameter space and the node graph structure of each time node during the training process to capture the spatial dependence relationship between the time nodes within the time step.

[0056] The data fusion layer is used to fuse the spatio-temporal features of the incomplete status monitoring data and the temporal features of the mask matrix to obtain fused spatio-temporal features. Specifically, the data fusion layer is used to concatenate the spatio-temporal features of the incomplete status monitoring data and the temporal features of the mask matrix along the feature dimension to fuse the two types of feature information and obtain fused spatio-temporal features.

[0057] The decoder includes a third bidirectional gated recurrent unit, a first fully connected network layer, a second fully connected network layer, and an output layer; the third bidirectional gated recurrent unit, the first fully connected network layer, and the second fully connected network layer are used to reconstruct the complete status monitoring data based on the fused spatio-temporal features and output the complete status monitoring data through the output layer.

[0058] B4. Divide the preprocessed historical status monitoring data set into a training data set, a test data set, and a validation data set; the training data set is used to train and optimize the weight parameters of the mask autoencoder time series imputation model, the validation data set is used to adjust the model hyperparameters, and the test data set is used to test and evaluate the performance of the trained mask autoencoder time series imputation model.

[0059] B5. For any preprocessed complete status monitoring data, use a randomly generated mask matrix to perform masking on the preprocessed complete status monitoring data to obtain masked status monitoring data; the masked status monitoring data, the mask matrix, and the preprocessed complete status monitoring data form a sample.

[0060] B6. Use the training data set to train the mask autoencoder time series imputation model, optimize the weights of the mask autoencoder time series imputation model, and obtain a trained mask autoencoder time series imputation model. In this embodiment, as Figure 4 shown, step B6 specifically includes the following steps:

[0061] B61. For any sample in the training data set, input the masked status monitoring data and the mask matrix into the mask autoencoder time series imputation model to obtain the completed status monitoring data.

[0062] B62. Calculate the loss function value of the mask autoencoder time series imputation model based on the completed status monitoring data and the preprocessed complete status monitoring data. Specifically, in this embodiment, the loss function of the mask autoencoder time series imputation model selects the mean absolute error MAE.

[0063] B63. Optimize the weight parameters of the mask autoencoder time series imputation model according to the loss function of the mask autoencoder time series imputation model. In this embodiment, the Adam optimization algorithm is used to optimize the weight parameters of the mask autoencoder time series imputation model, the learning rate is set to 0.001, and the number of training epochs is set to 100.

[0064] To better illustrate the application effect of the above method provided by this application, next, the ETTh1 power transformer oil temperature dataset is used to experimentally verify a method for filling missing values in complex equipment status monitoring data based on an adaptive temporal graph convolutional network proposed by this application.

[0065] The ETT dataset consists of power transformer monitoring data from two different regions in the same province. The dataset provides two years of data, with each data point recorded every minute. Due to the extremely large amount of data, the dataset provides dataset variants with data granularities of 15 minutes and 1 hour for use, denoted as ETTm1, ETTm2, ETTh1, and ETTh2. Each data point consists of 1 oil temperature data and 6 power load characteristics. The ETTh1 dataset used in this example is the power transformer status monitoring dataset with a data granularity of 1 hour in the first region.

[0066] After obtaining the above dataset, first, preprocess these monitoring data. Specifically, use the following formula to normalize the status monitoring data:

[0067]

[0068] where, X norm is the normalized status monitoring data, X is the original status monitoring data, X max and X min are the maximum and minimum values of each status monitoring data along the time axis, respectively. Through normalization, the status monitoring data is transformed into data within the range of -1 to 1 for subsequent processing. Data normalization is a commonly used data preprocessing method. Data normalization can not only accelerate the convergence speed of the model but also effectively improve the accuracy of the model.

[0069] Next, perform the division of the dataset. This step mainly divides the dataset into a training dataset, a validation dataset, and a test dataset according to a certain ratio. The training dataset is used to calculate the gradient to update the weights, that is, to train the model; the validation dataset is used for model selection, that is, to adjust the hyperparameters of the model; the test dataset is used to test the final performance of the model and evaluate indicators such as the accuracy and error of the model. In this example, the dataset is divided according to the ratio of 8:1:1.

[0070] In order to use the structure of the autoencoder to recover the real data from the missing data, it is necessary to perform masking processing on the original status monitoring data. The specific method of masking processing is as follows: Define a masking matrix M, whose dimension is the same as the original data, used to indicate whether the data is masked. The values in the masking matrix are specifically defined as follows:

[0071]

[0072] Within each time step, a portion of the elements in the masking matrix for that time step are randomly set to 0 according to the masking ratio, and the other elements are set to 1. After the masking matrix is set, multiplying the masking matrix by the original condition monitoring data gives the masked condition monitoring data. Since power data is periodic, the actual power usage should be on a daily cycle. In this embodiment, the time step is set to 48 with a cycle of two days, and the masking ratio is taken as 20%.

[0073] Subsequently, based on the adaptive graph convolutional network and the bidirectional gated recurrent unit, a masked autoencoder time series interpolation model as shown in Figure 4 is constructed. The encoder therein includes: an adaptive graph convolutional network, a first bidirectional gated recurrent unit, a second bidirectional gated recurrent unit, and a data fusion layer; the decoder includes: a third bidirectional gated recurrent unit, a first fully connected network layer, a second fully connected network layer, and an output layer.

[0074] In the encoder, the masked condition monitoring data passes through the adaptive graph convolutional network and the first bidirectional gated recurrent unit to extract its spatio-temporal features; the masking matrix is separately input into the second bidirectional gated recurrent unit to extract its masking information; the above two results are input into the data fusion layer to obtain the result of data fusion, which is then input into the decoder.

[0075] In this example, the number of neurons in the adaptive graph convolutional network is set to 64, and the feature dimension of the learnable parameters is all set to 100. In a specific implementation, the adaptive graph convolutional network layer operates along the time dimension, assigns a learnable parameter space and a node embedding dictionary for each time node within the current time step, and learns the parameter space and node graph structure of the nodes during the training process to capture the spatial dependence relationships among the time nodes within the time step.

[0076] The two bidirectional gated recurrent units are respectively used to process the results of graph convolution and the information of the masking matrix. The number of neurons in both layers of the network is 64, and the activation function uses the "tanh" function.

[0077] In the data fusion layer, the above two results passing through the bidirectional gated recurrent units are concatenated along the feature dimension to fuse the two types of information and facilitate the processing of subsequent networks.

[0078] In the decoder, the result of data fusion passes through the third bidirectional gated recurrent unit, two layers of fully connected networks, and the output layer, and finally the reconstructed original data is obtained. Specifically, in the decoder, the number of neurons in the third bidirectional gated recurrent unit is 128, and the activation function uses the "tanh" function. The number of neurons in the two layers of fully connected networks is 128 and 50 respectively, and their activation functions both use the rectified linear unit (ReLU). The output layer uses a linear layer, and the number of its neurons is 7, and the outputs are the reconstructed values of 7 feature data respectively.

[0079] The samples in the training dataset obtained by dividing the above dataset are input into the masked autoencoder time series imputation model to train the model. The loss function for model training is the mean absolute error (MAE), the learner uses the Adam optimization algorithm, the learning rate is set to 0.001, and the number of training epochs is set to 100.

[0080] After completing the training of the above masked autoencoder time series imputation model, the samples in the validation dataset are input into the model to obtain the completed condition monitoring data, and it is compared with the original condition monitoring data. Three evaluation functions commonly used in regression problems are selected as the evaluation parameters in this example, namely the root mean square error (RMSE), the mean absolute error (MAE), and the coefficient of determination (R2). The smaller the RMSE and MAE, the more accurate the imputation result, and the closer the R2 is to 1, the better the imputation result fits the true value and the higher the imputation accuracy.

[0081] To avoid the influence of accidental factors, each group of experiments is repeated ten times, and the RMSE, MAE, and R2 values of the ten experiments are recorded, and their average values and variances are taken as the final evaluation parameters. In this example, the average values and standard deviations of the three evaluation parameters are shown in the following table:

[0082] Table 1 Model Validation Evaluation Results

[0083] RMSE MAE R2 Average value 0.3961 0.1209 0.9705 Standard deviation 0.0388 0.0112 0.0025

[0084] As can be seen from Table 1 above, after multiple experiments, the average values of the root mean square error and the mean absolute error of the experimental results are small, and the coefficient of determination is close to 1, indicating that the above imputation method proposed in this application has good imputation accuracy. At the same time, the standard deviations of the three evaluation parameters are small, and the experimental results are relatively stable, indicating that it has good stability. Figure 5 The comparison between the imputation values obtained by completing the oil temperature data in this example and the original true values is shown. The above results indicate that the above solution of this application can effectively impute the missing data in the complex equipment condition monitoring data.

[0085] Based on the above solution of the present application, without introducing prior knowledge, operations are carried out along the time dimension using an adaptive temporal graph convolutional network, fully extracting the spatio-temporal features of the data, and reconstructing the original data from the masked data through the structure of a masked autoencoder, which can be widely applied to the missing value completion task of complex equipment status monitoring data in different industrial fields. In the field of missing value completion of complex equipment status monitoring data, the present application first proposes to use an adaptive temporal graph convolutional network to extract the spatio-temporal features of the data, and use the structure of a masked autoencoder to reconstruct the original data from the masked data, ultimately realizing the missing value completion of complex equipment status monitoring data. This method has good innovation and practicability.

[0086] Based on the same inventive concept, the present application also provides a system for implementing the method for completing missing values in complex equipment status monitoring data based on an adaptive temporal graph convolutional network involved in the above embodiments. The implementation solution provided by this system for solving problems is similar to the implementation solution described in the above method. In an exemplary embodiment, as Figure 6 shown, a system for completing missing values in complex equipment status monitoring data based on an adaptive temporal graph convolutional network is provided, including the following modules:

[0087] A status monitoring data acquisition module, configured to acquire incomplete status monitoring data; the incomplete status monitoring data is status monitoring data with incomplete values collected during the operation of complex equipment.

[0088] A data preprocessing module, configured to preprocess the incomplete status monitoring data to obtain the preprocessed incomplete status monitoring data. Specifically, in this embodiment, the incomplete status monitoring data is preprocessed according to the following formula:

[0089]

[0090] where, X norm is the preprocessed incomplete status monitoring data, X is the incomplete status monitoring data, X min and X max are the minimum value and the maximum value of the incomplete status monitoring data respectively.

[0091] A masked matrix generation module, configured to generate a masked matrix according to the preprocessed incomplete status monitoring data; the masked matrix is used to characterize the distribution of missing values in the preprocessed incomplete status monitoring data.

[0092] The status monitoring data completion module is used to input the preprocessed incomplete status monitoring data and the mask matrix into the trained mask autoencoder time series interpolation model to obtain the completed status monitoring data. The mask autoencoder time series interpolation model includes an adaptive graph convolutional network layer and a first bidirectional gated recurrent unit for extracting the spatio-temporal features of the incomplete status monitoring data, a second bidirectional gated recurrent unit for extracting the time series features of the mask matrix, a data fusion layer for fusing the spatio-temporal features of the incomplete status monitoring data and the time series features of the mask matrix, and a decoder for decoding and outputting according to the fused spatio-temporal features obtained by the data fusion layer.

[0093] Of course, Figure 6 the architecture shown is only exemplary. When implementing different functions, one or at least two components in the Figure 6 shown system can be omitted according to actual needs.

[0094] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as Figure 7 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, the method for completing missing values of complex equipment status monitoring data based on the adaptive time series graph convolutional network provided in the above embodiment can be implemented.

[0095] Those skilled in the art can understand that Figure 7 the structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0096] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0097] In this text, specific examples are used to elaborate on the principles and implementation manners of this application. The descriptions of the above embodiments are only used to help understand the method and its core idea of this application. At the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A method for completing missing values ​​of complex equipment status monitoring data based on an adaptive temporal graph convolutional network, characterized in that: include: Obtaining defect status monitoring data; The incomplete state monitoring data is state monitoring data with incomplete values ​​collected during the operation of complex equipment; Preprocessing the defective state monitoring data to obtain preprocessed defective state monitoring data; Generate a mask matrix according to the preprocessed incomplete state monitoring data; The mask matrix is ​​used to characterize the distribution of missing values ​​in the preprocessed incomplete state monitoring data; The preprocessed incomplete state monitoring data and the mask matrix are input into the trained masked autoencoder timing interpolation model to obtain the completed state monitoring data; the masked autoencoder timing interpolation model includes an adaptive graph convolutional network layer and a first bidirectional gated recurrent unit for extracting the spatiotemporal features of the incomplete state monitoring data, a second bidirectional gated recurrent unit for extracting the temporal features of the mask matrix, a data fusion layer for fusing the spatiotemporal features of the incomplete state monitoring data with the temporal features of the mask matrix, and a decoder for decoding and outputting the fused spatiotemporal features obtained by the data fusion layer.

2. According to claim 1, the method for completing missing values ​​of complex equipment status monitoring data based on an adaptive temporal graph convolutional network is characterized in that: Before inputting the preprocessed incomplete state monitoring data and the mask matrix into the trained masked autoencoder time series interpolation model to obtain the completed state monitoring data, the method for completing missing values ​​of complex equipment state monitoring data based on an adaptive time series graph convolutional network also includes: Acquire a historical state monitoring data set; the historical state monitoring data set includes a plurality of complete state monitoring data, wherein the complete state monitoring data is state monitoring data collected during the operation of complex equipment and does not contain incomplete values; For any complete state monitoring data, preprocessing the complete state monitoring data to obtain preprocessed complete state monitoring data; all preprocessed complete state monitoring data constitute a preprocessed historical state monitoring data set; Based on an adaptive graph convolutional network and several bidirectional gated recurrent units, a masked autoencoder temporal interpolation model is constructed; The preprocessed historical state monitoring data set is divided into a training data set, a test data set and a validation data set; the training data set is used to train and optimize the weight parameters of the masked autoencoder time series interpolation model, the validation data set is used to adjust the model hyperparameters, and the test data set is used to test and evaluate the performance of the trained masked autoencoder time series interpolation model; For any preprocessed complete state monitoring data, mask the preprocessed complete state monitoring data using a randomly generated mask matrix to obtain masked state monitoring data; the masked state monitoring data, the mask matrix and the preprocessed complete state monitoring data constitute a sample; The masked autoencoder temporal interpolation model is trained using the training data set, the weight of the masked autoencoder temporal interpolation model is optimized, and a trained masked autoencoder temporal interpolation model is obtained.

3. According to claim 2, the method for completing missing values ​​of complex equipment status monitoring data based on an adaptive temporal graph convolutional network is characterized in that: The masked autoencoder temporal interpolation model is trained by using the training data set, and the weight of the masked autoencoder temporal interpolation model is optimized to obtain a trained masked autoencoder temporal interpolation model, specifically including: For any sample in the training data set, inputting the masked state monitoring data and the mask matrix into the masked autoencoder time series interpolation model to obtain the completed state monitoring data; According to the completed state monitoring data and the preprocessed complete state monitoring data, a loss function value of the masked autoencoder time series interpolation model is calculated; According to the loss function of the masked autoencoder temporal interpolation model, the weight parameters of the masked autoencoder temporal interpolation model are optimized.

4. According to claim 3, the method for completing missing values ​​of complex equipment status monitoring data based on an adaptive temporal graph convolutional network is characterized in that: The loss function of the masked autoencoder time series interpolation model selects the mean absolute error MAE.

5. According to claim 2, the method for completing missing values ​​of complex equipment status monitoring data based on an adaptive temporal graph convolutional network is characterized in that: The Adam optimization algorithm is used to optimize the weight parameters of the masked autoencoder time series interpolation model, the learning rate is set to 0.001, and the number of training rounds is set to 100.

6. The method for completing missing values ​​of complex equipment status monitoring data based on an adaptive temporal graph convolutional network according to claim 1 is characterized in that: The masked autoencoder timing interpolation model includes an encoder and a decoder; the encoder includes an adaptive graph convolutional network layer, a first bidirectional gated recurrent unit, a second bidirectional gated recurrent unit and a data fusion layer; the adaptive graph convolutional network layer and the first bidirectional gated recurrent unit are used to extract the spatiotemporal features of the incomplete state monitoring data; the second bidirectional gated recurrent unit is used to extract the timing features of the mask matrix; the data fusion layer is used to fuse the spatiotemporal features of the incomplete state monitoring data with the timing features of the mask matrix to obtain fused spatiotemporal features; the decoder includes a third bidirectional gated recurrent unit, a first fully connected network layer, a second fully connected network layer and an output layer; the third bidirectional gated recurrent unit, the first fully connected network layer and the second fully connected network layer are used to reconstruct the completed state monitoring data according to the fused spatiotemporal features, and output the completed state monitoring data through the output layer.

7. The method for completing missing values ​​of complex equipment status monitoring data based on an adaptive temporal graph convolutional network according to claim 6 is characterized in that: The adaptive graph convolutional network layer is used to extract the spatial features of the incomplete state monitoring data, and further extracts the temporal features of the incomplete state monitoring data through the first bidirectional gated recurrent unit to achieve the extraction of the spatiotemporal features of the incomplete state monitoring data.

8. The method for completing missing values ​​of complex equipment status monitoring data based on an adaptive temporal graph convolutional network according to claim 6 is characterized in that: The adaptive graph convolutional network layer is used to operate along the time dimension, allocate a learnable parameter space and node embedding dictionary to each time node in the current time step, obtain the node graph structure according to the similarity definition between the previous node embedding dictionaries, and learn the parameter space and node graph structure of each time node during the training process to capture the spatial dependencies of each time node in the time step.

9. The method for completing missing values ​​of complex equipment status monitoring data based on an adaptive temporal graph convolutional network according to claim 1 is characterized in that: The data fusion layer is used to splice the spatiotemporal features of the incomplete state monitoring data and the temporal features of the mask matrix along the feature dimension to fuse the two types of feature information to obtain fused spatiotemporal features.

10. The method for completing missing values ​​of complex equipment status monitoring data based on an adaptive temporal graph convolutional network according to claim 1 is characterized in that: The incomplete state monitoring data is preprocessed according to the following formula: Among them, X norm is the incomplete state monitoring data after preprocessing, X is the incomplete state monitoring data, X min and X max are respectively the minimum and maximum values ​​of the incomplete state monitoring data.

Images