Complex equipment state monitoring data missing value completion method based on adaptive timing diagram convolution network

Abstract

The application discloses a complex equipment state monitoring data missing value completion method based on an adaptive timing diagram convolution network, and relates to the technical field of industrial data processing.The method comprises the following steps: obtaining incomplete state monitoring data and preprocessing the same, generating a mask matrix in accordance with the incomplete state monitoring data, subsequently inputting the incomplete state monitoring data and the mask matrix into a trained mask autoencoder timing interpolation model, extracting the space-time features of the incomplete state monitoring data through an adaptive graph convolution network layer, simultaneously extracting the timing features of the mask matrix through a second bidirectional gated recurrent unit, then fusing the two kinds of features, and outputting the completed state monitoring data through a decoder after decoding.The above scheme of the application utilizes the adaptive graph convolution network to operate along the time dimension, can fully mine the space-time features of the data, and utilizes the bidirectional gated recurrent unit to mine the timing features from the mask data to reconstruct the missing part of the data, thereby completing the completion of the missing values.

Description

Technical Field

[0001] This application relates to the field of industrial data processing technology, and in particular to a method for completing missing values ​​in complex equipment condition monitoring data based on adaptive temporal graph convolutional networks. Background Technology

[0002] With the continuous advancement of intelligent manufacturing, complex equipment generates a vast amount of condition monitoring data through various sensors during operation. This data contains crucial information about equipment operation and is essential for condition monitoring, fault diagnosis, and predictive maintenance. However, in practical industrial applications, due to various factors such as sensor failure, signal transmission interference, and hardware or software malfunctions in the data acquisition system, missing values ​​often appear in the condition monitoring data. These missing values ​​not only reduce data integrity but may also increase errors in subsequent analysis and processing, affecting the accurate assessment of the condition of complex equipment and consequently impacting the formulation of critical decisions. Therefore, effectively filling in 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 are widely used for missing value completion tasks in state monitoring data. In particular, graph neural network-based models excel in capturing spatial dependencies in data. While these methods have achieved some success in missing value completion, they still have limitations. Most graph neural network-based methods rely on prior knowledge, defining the graph structure through node connectivity or geographic location to determine spatial dependencies between nodes. The graph structures constructed by this approach are highly dependent on prior knowledge, making them difficult to apply to specific domains. Furthermore, existing graph structures primarily model feature nodes and are not well-suited for time-series data.

[0004] In view of this, this application proposes a new method for completing missing values ​​in the condition monitoring data of complex equipment, which is intended to be widely applicable to different industrial fields and effectively solve the problem of completing missing values ​​in the condition monitoring data of complex equipment. Summary of the Invention

[0005] The purpose of this application is to provide a method for completing missing values ​​in the condition monitoring data of complex equipment based on adaptive temporal graph convolutional networks. This method can be widely applied to completing missing values ​​in the condition monitoring data of complex equipment in different industrial fields, and has good innovation and practicality.

[0006] To achieve the above objectives, this application provides the following solution:

[0007] A method for imputing missing values ​​in complex equipment condition monitoring data based on adaptive temporal graph convolutional networks includes the following steps:

[0008] Acquire defect status monitoring data; defect status monitoring data refers to status monitoring data with defective values ​​collected during the operation of complex equipment.

[0009] The incomplete state monitoring data is preprocessed to obtain preprocessed incomplete state monitoring data.

[0010] A mask matrix is ​​generated 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] The preprocessed incomplete state monitoring data and mask matrix are input into the trained mask autoencoder temporal interpolation model to obtain the completed state monitoring data. The mask autoencoder temporal 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 and the temporal features of the mask matrix, and a decoder for decoding and outputting the fused spatiotemporal features obtained from the data fusion layer.

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

[0013] Obtain historical condition monitoring datasets; historical condition monitoring datasets include several complete condition monitoring data, which are condition monitoring data collected during the operation of complex equipment without any missing values.

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

[0015] A mask autoencoder temporal interpolation model is constructed based on an adaptive graph convolutional network and several bidirectional gated recurrent units.

[0016] The preprocessed historical state monitoring dataset is divided into a training dataset, a test dataset, and a validation dataset. The training dataset is used to train and optimize the weight parameters of the masked autoencoder temporal interpolation model, the validation dataset is used to adjust the model hyperparameters, and the test dataset is used to test and evaluate the performance of the trained masked autoencoder temporal interpolation model.

[0017] For any preprocessed complete state monitoring data, a randomly generated mask matrix is ​​used to mask the preprocessed complete state monitoring data to obtain masked state monitoring data; the masked state monitoring data, the mask matrix, and the preprocessed complete state monitoring data constitute a sample.

[0018] The mask autoencoder timing interpolation model is trained using the training dataset, and the weights of the mask autoencoder timing interpolation model are optimized to obtain the trained mask autoencoder timing interpolation model.

[0019] Optionally, the mask autoencoder temporal interpolation model is trained using the training dataset to optimize its weights, resulting in a well-trained mask autoencoder temporal interpolation model. This process includes the following steps:

[0020] For any sample in the training dataset, the mask state monitoring data and the mask matrix are input into the mask autoencoder temporal interpolation model to obtain the completed state monitoring data.

[0021] Based on the completed state monitoring data and the preprocessed complete state monitoring data, the loss function value of the mask autoencoder temporal interpolation model is calculated.

[0022] Based on the loss function of the masked autoencoder temporal interpolation model, optimize the weight parameters of the masked autoencoder temporal interpolation model.

[0023] Optionally, the loss function for the masked autoencoder temporal interpolation model is selected as the mean absolute error (MAE).

[0024] Optionally, the Adam optimization algorithm is used to optimize the weight parameters of the masked autoencoder temporal interpolation model, with the learning rate set to 0.001 and the number of training epochs set to 100.

[0025] Optionally, the mask autoencoder temporal 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 temporal features of the mask matrix; the data fusion layer is used to fuse the spatiotemporal features of the incomplete state monitoring data and the temporal 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 based on the fused spatiotemporal features, and output the completed state monitoring data through the output layer.

[0026] Optionally, 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, thereby realizing the extraction of the spatiotemporal features of the incomplete state monitoring data.

[0027] Optionally, the adaptive graph convolutional network layer is used to operate along the time dimension, assigning a learnable parameter space and node embedding dictionary to each time node within the current time step, obtaining the node graph structure based on the similarity between the previous node embedding dictionaries, and learning the parameter space and node graph structure of each time node during training, capturing the spatial dependencies of each time node within the time step.

[0028] Optionally, the data fusion layer is used to concatenate 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 and obtain fused spatiotemporal features.

[0029] Optionally, the incomplete state monitoring data can be preprocessed according to the following formula:

[0030]

[0031] Among them, X norm X represents the preprocessed incomplete state monitoring data. min and X max These are the minimum and maximum values ​​of the incomplete state monitoring data, respectively.

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

[0033] This application provides a method for completing missing values ​​in complex equipment condition monitoring data based on an adaptive temporal graph convolutional network. In this method, after preprocessing the acquired incomplete condition monitoring data, a mask matrix is ​​generated. Then, the preprocessed incomplete condition monitoring data and the mask matrix are input into a trained mask autoencoder temporal interpolation model. In this mask autoencoder temporal interpolation model, the spatiotemporal features of the incomplete condition monitoring data are extracted through an adaptive graph convolutional network layer and a first bidirectional gated recurrent unit. At the same time, the temporal features of the mask matrix are extracted through a second bidirectional gated recurrent unit. Then, a data fusion layer is used to fuse the spatiotemporal features of the incomplete condition monitoring data and the temporal features of the mask matrix. Finally, a decoder decodes and outputs the completed condition monitoring data based on the fused spatiotemporal features obtained by the data fusion layer. The above-mentioned scheme in this application operates along the time dimension using an adaptive graph convolutional network without introducing prior knowledge. It can fully mine the spatiotemporal features of the data and use bidirectional gated recurrent units to mine temporal features from the masked data to reconstruct the missing data and complete the missing value completion. It can be widely applied to the task of completing the missing value of complex equipment condition monitoring data in different industrial fields, and has good innovation and practicality. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a flowchart illustrating a method for completing missing values ​​in the status monitoring data of complex equipment based on an adaptive temporal graph convolutional network, as provided in an embodiment of this application.

[0036] Figure 2 This is a flowchart illustrating the process 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, as provided in an embodiment of this application.

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

[0038] Figure 4 This is a schematic diagram of the structure of a mask autoencoder temporal interpolation model in a method for filling missing values ​​in complex equipment status monitoring data based on an adaptive temporal graph convolutional network, provided in an embodiment of this application.

[0039] Figure 5 This diagram illustrates the comparison between the interpolated and true values ​​of oil temperature data in a method for completing missing values ​​in complex equipment condition monitoring data based on an adaptive temporal graph convolutional network, as provided in an embodiment of this application.

[0040] Figure 6 This is a schematic diagram of the functional modules of a system for completing missing values ​​in complex equipment status monitoring data based on an adaptive temporal graph convolutional network, provided in an embodiment of this application.

[0041] Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0042] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0043] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

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

[0045] A1. Obtain defect status monitoring data; defect status monitoring data refers to status monitoring data with defective values ​​collected during the operation of complex equipment.

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

[0047]

[0048] Among them, X norm X represents the preprocessed incomplete state monitoring data. min and X max These are the minimum and maximum values ​​of the incomplete state monitoring data, respectively.

[0049] A3. 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.

[0050] A4. Input the preprocessed incomplete state monitoring data and mask matrix into the trained mask autoencoder temporal interpolation model to obtain the completed state monitoring data. The mask autoencoder temporal 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 and the temporal features of the mask matrix, and a decoder for decoding and outputting the fused spatiotemporal features obtained from the data fusion layer.

[0051] In another exemplary embodiment, prior to step A4, the method for imputing missing values ​​in complex equipment status monitoring data based on adaptive temporal graph convolutional networks further includes a process of acquiring historical data to train and optimize the model, such as... Figure 2 As shown, the process includes the following steps:

[0052] B1. Obtain historical condition monitoring dataset; The historical condition monitoring dataset includes several complete condition monitoring data, which are condition monitoring data collected during the operation of complex equipment without any missing values.

[0053] B2. For any complete state monitoring data, preprocess the complete state monitoring data to obtain preprocessed complete state monitoring data; all preprocessed complete state monitoring data constitute the preprocessed historical state monitoring dataset.

[0054] B3. Based on an adaptive graph convolutional network and several bidirectional gated recurrent units, a masked autoencoder temporal interpolation model is constructed. In this embodiment, as shown... Figure 3 As shown, the mask autoencoder temporal 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 is used to extract the spatial features of the incomplete state monitoring data, and further, the first bidirectional gated recurrent unit extracts the temporal features of the incomplete state monitoring data, thus realizing the extraction of the spatiotemporal features of the incomplete state monitoring data. The second bidirectional gated recurrent unit is used to extract the temporal features of the mask matrix.

[0055] Specifically, the adaptive graph convolutional network layer is used to operate along the time dimension, assigning a learnable parameter space and node embedding dictionary to each time node within the current time step, obtaining the node graph structure based on the similarity between the previous node embedding dictionaries, and learning the parameter space and node graph structure of each time node during training, capturing the spatial dependencies between time nodes within the time step.

[0056] The data fusion layer is used to fuse the spatiotemporal features of the incomplete state monitoring data and the temporal features of the mask matrix to obtain fused spatiotemporal features. Specifically, the data fusion layer is used to concatenate 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 and obtain fused spatiotemporal features.

[0057] The decoder includes a third bidirectional gated loop unit, a first fully connected network layer, a second fully connected network layer, and an output layer. The third bidirectional gated loop unit, the first fully connected network layer, and the second fully connected network layer are used to reconstruct the completed state monitoring data based on the fused spatiotemporal features, and output the completed state monitoring data through the output layer.

[0058] B4. Divide the preprocessed historical state monitoring dataset into a training dataset, a test dataset, and a validation dataset. The training dataset is used to train and optimize the weight parameters of the mask autoencoder temporal interpolation model. The validation dataset is used to adjust the model hyperparameters. The test dataset is used to test and evaluate the performance of the trained mask autoencoder temporal interpolation model.

[0059] B5. For any preprocessed complete state monitoring data, use a randomly generated mask matrix to mask the preprocessed complete state monitoring data to obtain masked state monitoring data; the masked state monitoring data, the mask matrix, and the preprocessed complete state monitoring data constitute a sample.

[0060] B6. Train the masked autoencoder timing interpolation model using the training dataset, optimize the weights of the masked autoencoder timing interpolation model, and obtain the trained masked autoencoder timing interpolation model. In this embodiment, as... Figure 4 As shown, step B6 specifically includes the following steps:

[0061] B61. For any sample in the training dataset, input the mask state monitoring data and the mask matrix into the mask autoencoder temporal interpolation model to obtain the completed state monitoring data.

[0062] B62. Based on the completed state monitoring data and the preprocessed complete state monitoring data, the loss function value of the masked autoencoder timing interpolation model is calculated. Specifically, in this embodiment, the loss function of the masked autoencoder timing interpolation model is selected as the mean absolute error (MAE).

[0063] B63. Optimize the weight parameters of the masked autoencoder temporal interpolation model based on its loss function. In this embodiment, the Adam optimization algorithm is used to optimize the weight parameters of the masked autoencoder temporal interpolation model, with a learning rate of 0.001 and a training epoch count of 100.

[0064] To better illustrate the application effect of the method provided in this application, the following experiment uses the ETTh1 power transformer oil temperature dataset to verify the method for completing missing values ​​in complex equipment condition monitoring data based on adaptive temporal graph convolutional networks proposed in this application.

[0065] The ETT dataset consists of power transformer monitoring data from two different regions within the same province. The dataset provides two years of data, with each data point recorded once per minute. Due to the massive amount of data, variant datasets with granularities of 15 minutes and 1 hour are provided for use, denoted as ETTm1, ETTm2, ETTh1, and ETTh2. Each data point in these variants consists of one oil temperature data point and six electrical load characteristics. This example uses the ETTh1 dataset, which is the power transformer condition monitoring dataset from the first region with a granularity of 1 hour.

[0066] After obtaining the above dataset, the monitoring data is first preprocessed, specifically by normalizing the status monitoring data using the following formula:

[0067]

[0068] Among them, X norm This is the normalized state monitoring data, where X is the original state monitoring data. max and X min These represent the maximum and minimum values ​​of each state monitoring data along the time axis. Through normalization, the state 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; it not only accelerates model convergence but also effectively improves model accuracy.

[0069] The next step is to partition the dataset. This step involves dividing the dataset into training, validation, and test datasets according to a certain ratio. The training dataset is used to calculate gradients and update weights, i.e., to train the model; the validation dataset is used for model selection, i.e., to adjust the model's hyperparameters; and the test dataset is used to test the model's final performance and evaluate metrics such as accuracy and error. In this example, the dataset is partitioned in an 8:1:1 ratio.

[0070] To recover true data from missing data using the structure of an autoencoder, the original state monitoring data needs to be masked. The specific method for masking is as follows: Define a mask matrix M with the same dimensions as the original data, used to indicate whether the data is masked. The values ​​in the mask matrix are defined as follows:

[0071]

[0072] Within each time step, some elements of the mask matrix within that time step are randomly set to 0, and other elements are set to 1, according to the mask ratio. After setting the mask matrix, multiplying the mask matrix with the original state monitoring data yields the masked state monitoring data. Since power data is periodic, and actual power usage should be on a daily cycle, this embodiment sets the time step to 48 based on a two-day cycle, and the mask ratio to 20%.

[0073] Subsequently, based on an adaptive graph convolutional network and a bidirectional gated recurrent unit, a system was built as follows: Figure 4 The mask autoencoder temporal interpolation model shown includes an encoder comprising an adaptive graph convolutional network, a first bidirectional gated recurrent unit, a second bidirectional gated recurrent unit, and a data fusion layer; and a decoder comprising 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 state monitoring data is processed by an adaptive graph convolutional network and a first bidirectional gated recurrent unit to extract its spatiotemporal features; the mask matrix is ​​input separately into a second bidirectional gated recurrent unit to extract its mask information; the above two results are input into the data fusion layer to obtain the data fusion result, and 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 all learnable parameters is set to 100. In one specific implementation, the adaptive graph convolutional network layer operates along the time dimension, assigning a learnable parameter space and node embedding dictionary to each time node within the current time step, and learning the parameter space and node graph structure of the nodes during training to capture the spatial dependencies between time nodes within the time step.

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

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

[0078] In the decoder, the data fusion result is passed through a third bidirectional gated recurrent unit, two fully connected layers, and an output layer to finally obtain the reconstructed original data. Specifically, in the decoder, the third bidirectional gated recurrent unit has 128 neurons, and the activation function is the "tanh" function. The two fully connected layers have 128 and 50 neurons respectively, and their activation functions are both rectified linear units (ReLU). The output layer is a linear layer with 7 neurons, and the output is the reconstructed value of 7 feature data.

[0079] The samples from the training dataset obtained by partitioning the dataset above are input into the masked autoencoder temporal interpolation 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 training the masked autoencoder temporal interpolation model described above, samples from the validation dataset are input into the model to obtain the completed state monitoring data, which is then compared with the original state monitoring data. Three commonly used evaluation functions in regression problems are selected as evaluation parameters in this example: root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). Smaller RMSE and MAE values ​​indicate more accurate interpolation results, while a R² closer to 1 indicates a better fit between the interpolation result and the true value, resulting in higher interpolation accuracy.

[0081] To avoid the influence of random factors, each experiment was repeated ten times, and the RMSE, MAE, and R² values ​​of the ten experiments were recorded. The mean and variance were then used as the final evaluation parameters. In this example, the mean and standard deviation of the three evaluation parameters are shown in the table below:

[0082] Table 1. Model Validation and 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 mean absolute error of the experimental results are small, and the coefficient of determination is close to 1, indicating that the interpolation method proposed in this application has good interpolation 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 interpolated values ​​obtained by completing the oil temperature data in this example and the original true values ​​is shown. These results demonstrate that the proposed solution can effectively interpolate missing data in the condition monitoring data of complex equipment.

[0085] The proposed solution, without introducing prior knowledge, utilizes an adaptive temporal graph convolutional network along the time dimension to fully extract the spatiotemporal features of the data. It then reconstructs the original data from the masked data using a masked autoencoder structure. This method can be widely applied to missing value completion tasks in complex equipment condition monitoring data across various industrial sectors. In the field of missing value completion for complex equipment condition monitoring data, this application is the first to propose using an adaptive temporal graph convolutional network to extract the spatiotemporal features of the data and employing a masked autoencoder structure to reconstruct the original data from the masked data, ultimately achieving missing value completion for complex equipment condition monitoring data. This method demonstrates significant innovation and practicality.

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

[0087] The condition monitoring data acquisition module is used to acquire incomplete condition monitoring data; incomplete condition monitoring data refers to condition monitoring data with incomplete values ​​collected during the operation of complex equipment.

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

[0089]

[0090] Among them, X norm X represents the preprocessed incomplete state monitoring data. min and X max These are the minimum and maximum values ​​of the incomplete state monitoring data, respectively.

[0091] The mask matrix generation module is used to 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.

[0092] The state monitoring data completion module is used to input the preprocessed incomplete state monitoring data and the mask matrix into the trained mask autoencoder temporal interpolation model to obtain the completed state monitoring data. The mask autoencoder temporal 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 and the temporal features of the mask matrix, and a decoder for decoding and outputting the fused spatiotemporal features obtained from the data fusion layer.

[0093] certainly, Figure 6 The architecture shown is merely exemplary; it can be omitted as needed when implementing different functionalities. Figure 6 One or at least two components of the system shown.

[0094] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 7 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it can implement the method for completing missing values ​​in complex equipment status monitoring data based on adaptive temporal graph convolutional networks provided in the above embodiment.

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

[0096] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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, they should be considered to be within the scope of this specification.

[0097] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for completing missing values ​​in complex equipment condition monitoring data based on adaptive temporal graph convolutional networks, characterized in that, include: Acquire incomplete condition monitoring data; The defect status monitoring data refers to the status monitoring data with defective values ​​collected during the operation of complex equipment. The incomplete state monitoring data is preprocessed to obtain preprocessed incomplete state monitoring data; A mask matrix is ​​generated 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; The preprocessed incomplete state monitoring data and the mask matrix are input into the trained mask autoencoder temporal interpolation model to obtain the completed state monitoring data. The mask autoencoder temporal 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 temporal features of the mask matrix. The data fusion layer is used to fuse the spatiotemporal features of the incomplete state monitoring data and the temporal 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 based on the fused spatiotemporal features, and output the completed state monitoring data through the output layer.

2. The method for completing missing values ​​in complex equipment status monitoring data based on adaptive temporal graph convolutional networks according to claim 1, characterized in that, Before inputting the preprocessed incomplete state monitoring data and the mask matrix into the trained mask autoencoder temporal interpolation model to obtain the completed state monitoring data, the method for completing missing values ​​in complex equipment state monitoring data based on an adaptive temporal graph convolutional network further includes: Obtain a historical condition monitoring dataset; the historical condition monitoring dataset includes several complete condition monitoring data, which are condition monitoring data collected during the operation of complex equipment without any missing values; For any complete state monitoring data, the complete state monitoring data is preprocessed to obtain preprocessed complete state monitoring data; all preprocessed complete state monitoring data constitute a preprocessed historical state monitoring dataset. A mask autoencoder temporal interpolation model is constructed based on an adaptive graph convolutional network and several bidirectional gated recurrent units. The preprocessed historical state monitoring dataset is divided into a training dataset, a test dataset, and a validation dataset. The training dataset is used to train and optimize the weight parameters of the mask autoencoder temporal interpolation model. The validation dataset is used to adjust the model hyperparameters. The test dataset is used to test and evaluate the performance of the trained mask autoencoder temporal interpolation model. For any preprocessed complete state monitoring data, a randomly generated mask matrix is ​​used to mask the preprocessed complete state monitoring data 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 mask autoencoder timing interpolation model is trained using the training dataset, and the weights of the mask autoencoder timing interpolation model are optimized to obtain a trained mask autoencoder timing interpolation model.

3. The method for completing missing values ​​in complex equipment status monitoring data based on adaptive temporal graph convolutional networks according to claim 2, characterized in that, The mask autoencoder temporal interpolation model is trained using the training dataset, and the weights of the mask autoencoder temporal interpolation model are optimized to obtain a trained mask autoencoder temporal interpolation model, specifically including: For any sample in the training dataset, the mask state monitoring data and the mask matrix are input into the mask autoencoder temporal interpolation model to obtain the completed state monitoring data; Based on the completed state monitoring data and the preprocessed complete state monitoring data, the loss function value of the mask autoencoder temporal interpolation model is calculated. Based on the loss function of the mask autoencoder temporal interpolation model, optimize the weight parameters of the mask autoencoder temporal interpolation model.

4. The method for completing missing values ​​in complex equipment status monitoring data based on adaptive temporal graph convolutional networks according to claim 3, characterized in that, The loss function of the masked autoencoder temporal interpolation model is selected as the mean absolute error (MAE).

5. The method for completing missing values ​​in complex equipment status monitoring data based on adaptive temporal graph convolutional networks according to claim 2, characterized in that, The Adam optimization algorithm was used to optimize the weight parameters of the mask autoencoder temporal interpolation model, with the learning rate set to 0.001 and the number of training epochs set to 100.

6. The method for completing missing values ​​in complex equipment status monitoring data based on adaptive temporal graph convolutional networks according to claim 1, 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, thereby realizing the extraction of the spatiotemporal features of the incomplete state monitoring data.

7. The method for completing missing values ​​in complex equipment status monitoring data based on adaptive temporal graph convolutional networks according to claim 1, characterized in that, The adaptive graph convolutional network layer is used to operate along the time dimension, assigning a learnable parameter space and node embedding dictionary to each time node within the current time step, obtaining the node graph structure based on the similarity definition between the previous node embedding dictionaries, and learning the parameter space and node graph structure of each time node during training, capturing the spatial dependencies of each time node within the time step.

8. The method for completing missing values ​​in complex equipment status monitoring data based on adaptive temporal graph convolutional networks according to claim 1, 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 and obtain fused spatiotemporal features.

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

Images