Aviation monitoring data analysis method and system

Abstract

The invention relates to the technical field of data analysis, in particular to an aviation monitoring data analysis method and system. An aviation monitoring data analysis system comprises an aviation monitoring data set acquisition module and an aviation monitoring data analysis module. According to the invention, spatio-temporal characteristic analysis is carried out on the aviation monitoring data through the aviation monitoring data analysis model, so that problems can be found in time in the task execution process of the aircraft; in an aviation monitoring data analysis model, self-attention mechanism processing is carried out on an aviation monitoring data set through a spatial distribution characteristic matrix, and the time sequence change condition of sensor data which does not conform to spatial rationality can be concerned, so that the influence of redundant characteristics is inhibited, and the accuracy and universality of aviation monitoring data analysis are improved; and the time sequence change analysis of the sensor data is additionally added, so that the weakening condition of the convolution operation in the spatial feature time sequence analysis layer on the time sequence change of the sensor data is made up.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to an aviation monitoring data analysis method and system. Background Art

[0002] With the rapid development of modern aviation technology, the safety and reliability of aircraft have become the core issues of concern to the aviation industry. In order to ensure the normal operation of aircraft, aviation monitoring systems usually deploy a large number of sensors to collect real-time operating data of key aircraft components. These sensors cover multiple key parts such as engines, wings, landing gear, etc., and can record multiple physical parameters such as vibration, temperature, and pressure. Based on these sensor data, current aircraft fault diagnosis and flight problem analysis methods mainly rely on modeling and analyzing the temporal changes of sensor signals, such as identifying abnormal patterns through time series analysis, frequency domain analysis, and machine learning methods. However, such methods usually only focus on the data characteristics of a single sensor and its dynamic changes over time, while ignoring the spatial relationship between different sensors and their interactive characteristics.

[0003] As a complex whole, there are significant physical correlations between the various components of an aircraft. For example, abnormal vibration of the engine may be transmitted to the wing through the fuselage structure, resulting in asymmetric wing vibration; for another example, the imbalance of pressure distribution on the left and right wings may be closely related to the state of the landing gear. These spatial distribution characteristics are crucial for accurately identifying flight problems and locating the cause of faults. However, most existing analysis methods fail to effectively combine the spatial distribution relationship between sensors, resulting in limitations in dealing with complex flight problems. Summary of the invention

[0004] The present invention performs spatiotemporal feature analysis on aviation monitoring data through an aviation monitoring data analysis model, so as to timely discover problems in the process of aircraft performing tasks; in the aviation monitoring data analysis model, the aviation monitoring data set is processed by a self-attention mechanism through a spatial distribution feature matrix, so as to pay attention to the temporal changes of sensor data that do not conform to spatial rationality, thereby suppressing the influence of redundant features and improving the accuracy and versatility of aviation monitoring data analysis; and additionally adds temporal change analysis of the sensor data itself to compensate for the weakening of the temporal changes of the sensor data itself by the convolution operation in the spatial feature temporal analysis layer, so as to further improve the accuracy and versatility of aviation monitoring data analysis.

[0005] The present invention provides an aviation monitoring data analysis method, comprising: Acquire an aviation monitoring data set at the current monitoring time point. The aviation monitoring data set includes a number of sensor data of different parts of the aircraft, and the aviation monitoring data set is stored in the form of a matrix, where the horizontal dimension represents the type of sensor data and the vertical dimension represents the part of the aircraft; The aviation monitoring data set obtained at the current monitoring time point and the aviation monitoring data set obtained at the previous N-1 detection time points are combined into an aviation monitoring time series analysis data set, where N is generally 10. The aviation monitoring time series analysis data set is then sent to the aviation monitoring data analysis model for processing, and the aviation monitoring data analysis label is output; The aviation monitoring data analysis model includes a spatial distribution feature extraction layer, a spatial feature time series analysis layer, a time series feature analysis layer, a feature splicing layer and an aviation monitoring data analysis label output layer, wherein the spatial distribution feature extraction layer is used to process each aviation monitoring data set in the aviation monitoring time series analysis data set to output a spatial distribution feature matrix; the spatial feature time series analysis layer is used to perform a self-attention mechanism on the corresponding aviation monitoring data set based on the spatial distribution weight matrix, and perform time series analysis on the aviation monitoring data set processed by the self-attention mechanism to construct a spatial feature time series vector; the time series feature analysis layer is established based on the LSTM model, and is used to perform time series analysis on the aviation monitoring time series analysis data set to construct a time series feature vector; the feature splicing layer is used to splice the spatial feature time series vector and the time series feature vector to construct a time series comprehensive feature vector; the aviation monitoring data analysis label output layer is used to perform a full connection operation on the time series comprehensive feature vector to output the aviation monitoring data analysis label.

[0006] As a preferred aspect of the present invention, each aviation monitoring data set in the aviation monitoring time series analysis data set is processed by a spatial distribution feature extraction layer to output a spatial distribution feature matrix, which specifically includes the following steps: Each row in the aviation monitoring data set is regarded as a feature node, and a feature edge is constructed between any two feature nodes. The similarity between the two feature nodes is recorded as the feature edge weight. The aviation monitoring graph structure data is constructed based on all feature nodes and feature edges; the aviation monitoring adjacency matrix is ​​constructed, and the feature edge weights are stored in the aviation monitoring adjacency matrix; the aviation monitoring degree matrix is ​​constructed, and the aviation monitoring degree matrix is ​​a diagonal matrix; The graph convolution operation is performed on the aviation monitoring dataset using the following formula: G(t+1)=σ[HG(t)W(t)], where G(t) is the aviation monitoring dataset after the tth iteration, t=1, 2, …, U, U is the maximum number of iterations, σ[] is the activation function, H is the normalized aviation monitoring adjacency matrix, and the aviation monitoring adjacency matrix H=E -0.5 RE -0.5, R is the aviation monitoring adjacency matrix, E is the aviation monitoring degree matrix, W(t) is the graph convolution weight matrix of the tth iteration; the aviation monitoring dataset that has undergone U iterations is recorded as the spatial distribution feature matrix.

[0007] As a preferred aspect of the present invention, in the spatial feature time series analysis layer, the corresponding aviation monitoring data set is processed by the self-attention mechanism based on the spatial distribution weight matrix, and the aviation monitoring data set processed by the self-attention mechanism is subjected to time series analysis to construct a spatial feature time series vector, which specifically includes the following steps: For the corresponding spatial distribution weight matrix and aviation monitoring data set, the aviation monitoring data set is multiplied with the value weight matrix and the key weight matrix respectively to construct the aviation monitoring value vector V and the aviation monitoring key vector K. The spatial distribution weight matrix is ​​multiplied with the query weight matrix to construct the aviation monitoring query vector Q. The self-attention weight matrix ATT=softmax(QK T / (D) 0.5 ), where D is the horizontal dimension size of the aviation monitoring dataset, T is the matrix transpose operation, and then the self-attention weight matrix ATT is multiplied by the aviation monitoring value vector V to obtain the enhanced aviation monitoring dataset; All enhanced aviation monitoring data sets are subjected to time series analysis to output time series analysis graphs, which are then sent to a multi-layer perceptron for processing to construct an inter-feature time series vector.

[0008] As a preferred aspect of the present invention, before performing time series analysis on the aviation monitoring time series analysis data set through the time series feature analysis layer, it also includes a preprocessing operation on the aviation monitoring time series analysis data set. The specific steps are as follows: for each aviation monitoring data set in the aviation monitoring time series analysis data set, the aviation monitoring data set is flattened into one-dimensional data by row.

[0009] As a preferred aspect of the present invention, training is performed on the aviation monitoring data analysis model, specifically comprising the following steps: Acquire several aviation monitoring data analysis training samples, which include N aviation monitoring data sets sorted by time. Label the aviation monitoring data analysis training samples through aviation monitoring data analysis labels, and combine all labeled aviation monitoring data analysis training samples into an aviation monitoring data analysis training set. Send the aviation monitoring data analysis training set to the aviation monitoring data analysis model with initialized parameters for training. Calculate the loss value through the cross entropy loss function, and determine whether the loss value is within a preset range. If the loss value is within the preset range, output the trained aviation monitoring data analysis model; otherwise, continue to train the aviation monitoring data analysis model through the aviation monitoring data analysis training set.

[0010] As a preferred aspect of the present invention, the similarity between two feature nodes is calculated using a cosine similarity algorithm.

[0011] The present invention also provides an aviation monitoring data analysis system, comprising: The aviation monitoring data set acquisition module is used to acquire the aviation monitoring data set at the current monitoring time point. The aviation monitoring data set includes a number of sensor data of different parts of the aircraft, and the aviation monitoring data set is stored in the form of a matrix, where the horizontal dimension represents the type of sensor data and the vertical dimension represents the part of the aircraft; The aviation monitoring data analysis module is used to combine the aviation monitoring data set obtained at the current monitoring time point and the aviation monitoring data set obtained at the previous N-1 detection time points into an aviation monitoring time series analysis data set, where N is generally 10, and then send the aviation monitoring time series analysis data set to the aviation monitoring data analysis model for processing, and output aviation monitoring data analysis labels; The aviation monitoring data analysis model includes a spatial distribution feature extraction layer, a spatial feature time series analysis layer, a time series feature analysis layer, a feature splicing layer and an aviation monitoring data analysis label output layer, wherein the spatial distribution feature extraction layer is used to process each aviation monitoring data set in the aviation monitoring time series analysis data set to output a spatial distribution feature matrix; the spatial feature time series analysis layer is used to perform a self-attention mechanism on the corresponding aviation monitoring data set based on the spatial distribution weight matrix, and perform time series analysis on the aviation monitoring data set processed by the self-attention mechanism to construct a spatial feature time series vector; the time series feature analysis layer is established based on the LSTM model, and is used to perform time series analysis on the aviation monitoring time series analysis data set to construct a time series feature vector; the feature splicing layer is used to splice the spatial feature time series vector and the time series feature vector to construct a time series comprehensive feature vector; the aviation monitoring data analysis label output layer is used to perform a full connection operation on the time series comprehensive feature vector to output the aviation monitoring data analysis label.

[0012] The present invention has the following advantages: The present invention performs spatiotemporal feature analysis on aviation monitoring data through an aviation monitoring data analysis model, so as to timely discover problems in the process of aircraft performing tasks; in the aviation monitoring data analysis model, the aviation monitoring data set is processed by a self-attention mechanism through a spatial distribution feature matrix, so as to pay attention to the temporal changes of sensor data that do not conform to spatial rationality, thereby suppressing the influence of redundant features and improving the accuracy and versatility of aviation monitoring data analysis; and additionally adds temporal change analysis of the sensor data itself to compensate for the weakening of the temporal changes of the sensor data itself by the convolution operation in the spatial feature temporal analysis layer, so as to further improve the accuracy and versatility of aviation monitoring data analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A schematic diagram of the structure of the aviation monitoring data analysis model used in an embodiment of the present invention.

[0014] Figure 2 A schematic diagram of the structure of an aviation monitoring data analysis system used in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0016] Embodiment 1, an aviation monitoring data analysis method, comprising: The aviation monitoring data set is obtained at the current monitoring time point. The aviation monitoring data set includes several sensor data of different parts of the aircraft, such as the temperature, vibration frequency and pressure data of the engine, left wing or right wing. These data are obtained through the corresponding temperature sensors, accelerometers and pressure sensors. The aviation monitoring data set is stored in the form of a matrix. The horizontal dimension represents the type of sensor data and the vertical dimension represents the part of the aircraft. Assuming that the matrix is ​​A, the data a stored in it ij is the jth sensor data of the i-th part of the aircraft; it should be noted that the time interval between two adjacent monitoring time points is determined by the developer, which is generally 100s; The aviation monitoring data set obtained at the current monitoring time point and the aviation monitoring data set obtained at the previous N-1 detection time points are combined into an aviation monitoring time series analysis data set, where N is generally 10. The aviation monitoring time series analysis data set is then sent to the aviation monitoring data analysis model for processing, and an aviation monitoring data analysis label is output. The aviation monitoring data analysis label generally includes normal and abnormal. By analyzing the aviation monitoring data through the aviation monitoring data analysis model, problems can be discovered in a timely manner during the aircraft's mission execution to avoid the occurrence of flight accidents. like Figure 1As shown in the figure, the aviation monitoring data analysis model includes a spatial distribution feature extraction layer, a spatial feature time series analysis layer, a time series feature analysis layer, a feature splicing layer and an aviation monitoring data analysis label output layer, wherein the spatial distribution feature extraction layer is used to process each aviation monitoring data set in the aviation monitoring time series analysis data set to output a spatial distribution feature matrix, and the spatial distribution weight matrix can characterize the spatial distribution of sensor data in different parts, such as the conduction of temperature between different parts of the aircraft, and whether the vibration data matches between the symmetrical parts of the aircraft. The spatial distribution weight matrix can analyze the rationality of the aviation monitoring data in space, and then realize the subsequent aviation monitoring data analysis; the spatial feature time series analysis layer is established based on the ConvLSTM model, and the ConvLSTM model can realize the extraction of spatiotemporal features. The spatial feature time series analysis layer is used to perform self-attention mechanism processing on the corresponding aviation monitoring data set based on the spatial distribution weight matrix, and perform time series analysis on the aviation monitoring data set processed by the self-attention mechanism to construct a spatial The spatial feature time series vector is used to process each aviation monitoring data set in the aviation monitoring time series analysis data set through a self-attention mechanism. In this way, more attention can be paid to the time series changes of sensor data that do not conform to spatial rationality during the time series analysis process, thereby suppressing the influence of redundant features and improving the accuracy and versatility of aviation monitoring data analysis. The time series feature analysis layer is established based on the LSTM model and is used to perform time series analysis on the aviation monitoring time series analysis data set to construct a time series feature vector. When the spatial feature time series analysis layer is used for processing, the time series changes of the sensor data itself will be weakened due to the convolution operation. Therefore, the time series feature analysis layer established based on the LSTM model is used to extract the time series change characteristics of each sensor data itself, thereby further improving the accuracy of aviation monitoring data analysis. The feature concatenation layer is used to concatenate the spatial feature time series vector and the time series feature vector to construct a time series comprehensive feature vector. The aviation monitoring data analysis label output layer is used to perform a full connection operation on the time series comprehensive feature vector to output the aviation monitoring data analysis label. Before performing time series analysis on the aviation monitoring time series analysis data set through the time series feature analysis layer, a preprocessing operation on the aviation monitoring time series analysis data set is also included. The specific steps are as follows: for each aviation monitoring data set in the aviation monitoring time series analysis data set, the aviation monitoring data set is flattened into one-dimensional data by row, specifically, each row of data in the aviation monitoring data set is flattened by connecting the end to the end.

[0017] This application uses an aviation monitoring data analysis model to perform spatiotemporal feature analysis on aviation monitoring data, which can realize timely discovery of problems during the aircraft's mission execution; in the aviation monitoring data analysis model, the aviation monitoring data set is processed by a self-attention mechanism through a spatial distribution feature matrix, which can focus on the temporal changes of sensor data that do not conform to spatial rationality, thereby suppressing the influence of redundant features and improving the accuracy and versatility of aviation monitoring data analysis; and additionally adds temporal change analysis of the sensor data itself to compensate for the weakening of the temporal changes of the sensor data itself by the convolution operation in the spatial feature temporal analysis layer, thereby further improving the accuracy and versatility of aviation monitoring data analysis.

[0018] Each aviation monitoring data set in the aviation monitoring time series analysis data set is processed through the spatial distribution feature extraction layer to output a spatial distribution feature matrix, which specifically includes the following steps: Each row in the aviation monitoring data set is regarded as a feature node, and a feature edge is constructed between any two feature nodes. The similarity between the two feature nodes is recorded as the feature edge weight. The similarity calculation generally adopts the cosine similarity algorithm. The aviation monitoring graph structure data is constructed based on all feature nodes and feature edges. The aviation monitoring data set is constructed as graph structure data, which can analyze the interaction relationship between sensor data of different parts of the aircraft, and then analyze the spatial distribution of sensor data of different parts; construct an aviation monitoring adjacency matrix, which stores the feature edge weights. For example, if there are M feature nodes, the size of the aviation monitoring adjacency matrix is ​​M×M, and the data in the i-th row and j-th column of the aviation monitoring adjacency matrix is ​​the feature edge weight between the i-th feature node and the j-th feature node; construct an aviation monitoring degree matrix, which is a diagonal matrix, and the data on the diagonal is the number of connected feature edges corresponding to each feature node. Since any two feature nodes in this application are connected, the data on the diagonal are generally M; The graph convolution operation is performed on the aviation monitoring dataset using the following formula: G(t+1)=σ[HG(t)W(t)], where G(t) is the aviation monitoring dataset after the tth iteration, t=1, 2, …, U, U is the maximum number of iterations, U is generally 3, σ[] is the activation function, and the activation function generally uses ReLU, H is the normalized aviation monitoring adjacency matrix, and the aviation monitoring adjacency matrix H=E -0.5 RE -0.5 , R is the aviation monitoring adjacency matrix, E is the aviation monitoring degree matrix, W(t) is the graph convolution weight matrix of the tth iteration, and the parameters are set following the model training; the aviation monitoring data set after U iterations is recorded as the spatial distribution feature matrix; In the spatial feature time series analysis layer, the corresponding aviation monitoring data set is processed by the self-attention mechanism based on the spatial distribution weight matrix, and the aviation monitoring data set processed by the self-attention mechanism is subjected to time series analysis to construct the spatial feature time series vector, which specifically includes the following steps: For the corresponding spatial distribution weight matrix and aviation monitoring data set, the aviation monitoring data set is multiplied with the value weight matrix and the key weight matrix respectively to construct the aviation monitoring value vector V and the aviation monitoring key vector K. The spatial distribution weight matrix is ​​multiplied with the query weight matrix to construct the aviation monitoring query vector Q. The self-attention weight matrix ATT=softmax(QK T / (D) 0.5 ), where D is the horizontal dimension size of the aviation monitoring data set, T is the matrix transpose operation, and then the self-attention weight matrix ATT is multiplied by the aviation monitoring value vector V to obtain the enhanced aviation monitoring data set; the value weight matrix, key weight matrix and query weight matrix are set based on the self-attention mechanism in the Transformer model, and the value weight matrix, key weight matrix and query weight matrix are parameterized following the training of the aviation monitoring data analysis model; All enhanced aviation monitoring data sets are subjected to time series analysis. Specifically, the enhanced aviation monitoring data sets are processed through the ConvLSTM model to output a time series analysis graph, and then the time series analysis graph is sent to a multi-layer perceptron for processing. The specific operation is to send each row in the time series analysis graph to the neural nodes of the input layer of the multi-layer perceptron to construct an inter-feature time series vector.

[0019] Training the aviation monitoring data analysis model includes the following steps: A number of aviation monitoring data analysis training samples are obtained. The aviation monitoring data analysis training samples include N aviation monitoring data sets sorted by time. These aviation monitoring data analysis training samples can be obtained through actual aircraft flight tests or through flight simulation software. The aviation monitoring data analysis training samples are labeled with aviation monitoring data analysis labels. The labeling method can be 1 for normal and 0 for abnormal. All labeled aviation monitoring data analysis training samples form an aviation monitoring data analysis training set. The aviation monitoring data analysis training set is sent to the aviation monitoring data analysis model with parameter initialization for training. The loss value is calculated through the cross entropy loss function, and it is determined whether the loss value is within a preset range. The preset range is set by the developer. If the loss value is within the preset range, the trained aviation monitoring data analysis model is output; otherwise, the aviation monitoring data analysis model is continuously trained through the aviation monitoring data analysis training set.

[0020] Embodiment 2, an aviation monitoring data analysis system, such as Figure 2 As shown, including: The aviation monitoring data set acquisition module is used to acquire the aviation monitoring data set at the current monitoring time point. The aviation monitoring data set includes several sensor data of different parts of the aircraft, such as the temperature, vibration frequency and pressure data of the engine, left wing or right wing. These data are acquired through corresponding temperature sensors, accelerometers and pressure sensors. The aviation monitoring data set is stored in the form of a matrix. The horizontal dimension represents the type of sensor data and the vertical dimension represents the part of the aircraft. Assuming the matrix is ​​A, the data stored in it is ij is the jth sensor data of the i-th part of the aircraft; it should be noted that the time interval between two adjacent monitoring time points is determined by the developer, which is generally 100s; The aviation monitoring data analysis module is used to combine the aviation monitoring data set obtained at the current monitoring time point and the aviation monitoring data set obtained at the previous N-1 detection time points into an aviation monitoring time series analysis data set, where N is generally 10, and then send the aviation monitoring time series analysis data set to the aviation monitoring data analysis model for processing, and output aviation monitoring data analysis labels. Aviation monitoring data analysis labels generally include normal and abnormal. By analyzing the aviation monitoring data through the aviation monitoring data analysis model, problems can be discovered in a timely manner during the aircraft's mission execution to avoid the occurrence of flight accidents. The aviation monitoring data analysis model includes a spatial distribution feature extraction layer, a spatial feature time series analysis layer, a time series feature analysis layer, a feature splicing layer and an aviation monitoring data analysis label output layer. The spatial distribution feature extraction layer is used to process each aviation monitoring data set in the aviation monitoring time series analysis data set to output a spatial distribution feature matrix. The spatial distribution weight matrix can characterize the spatial distribution of sensor data in different parts, such as the conduction of temperature between different parts of the aircraft, and whether the vibration data matches between the symmetrical parts of the aircraft. The spatial distribution weight matrix can analyze the rationality of the aviation monitoring data in space, and then realize the subsequent aviation monitoring data analysis; the spatial feature time series analysis layer is established based on the ConvLSTM model, which can realize the extraction of spatiotemporal features. The spatial feature time series analysis layer is used to perform self-attention mechanism processing on the corresponding aviation monitoring data set based on the spatial distribution weight matrix, and perform time series analysis on the aviation monitoring data set processed by the self-attention mechanism to construct a spatial The feature time series vector is processed by the self-attention mechanism for each aviation monitoring data set in the aviation monitoring time series analysis data set. In the process of time series analysis, more attention can be paid to the time series changes of sensor data that do not conform to spatial rationality, thereby suppressing the influence of redundant features and improving the accuracy and versatility of aviation monitoring data analysis. The time series feature analysis layer is established based on the LSTM model and is used to perform time series analysis on the aviation monitoring time series analysis data set to construct a time series feature vector. When processed by the spatial feature time series analysis layer, the time series changes of the sensor data itself will be weakened due to the convolution operation. Therefore, the time series feature analysis layer established based on the LSTM model is used to extract the time series change characteristics of each sensor data itself, further improving the accuracy of aviation monitoring data analysis. The feature splicing layer is used to splice the spatial feature time series vector and the time series feature vector to construct a time series comprehensive feature vector. The aviation monitoring data analysis label output layer is used to perform a full connection operation on the time series comprehensive feature vector to output the aviation monitoring data analysis label.

[0021] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A method for analyzing aviation monitoring data, characterized in that: include: Acquire an aviation monitoring data set at the current monitoring time point. The aviation monitoring data set includes a number of sensor data of different parts of the aircraft, and the aviation monitoring data set is stored in the form of a matrix, where the horizontal dimension represents the type of sensor data and the vertical dimension represents the part of the aircraft; The aviation monitoring data set obtained at the current monitoring time point and the aviation monitoring data set obtained at the previous N-1 detection time points are combined into an aviation monitoring time series analysis data set, and then the aviation monitoring time series analysis data set is sent to the aviation monitoring data analysis model for processing, and the aviation monitoring data analysis label is output; The aviation monitoring data analysis model includes a spatial distribution feature extraction layer, a spatial feature time series analysis layer, a time series feature analysis layer, a feature splicing layer, and an aviation monitoring data analysis label output layer, wherein the spatial distribution feature extraction layer is used to process each aviation monitoring data set in the aviation monitoring time series analysis data set to output a spatial distribution feature matrix; space The feature time series analysis layer is used to process the corresponding aviation monitoring data set with a self-attention mechanism based on the spatial distribution weight matrix, and to perform time series analysis on the aviation monitoring data set processed by the self-attention mechanism to construct a spatial feature time series vector; The time series feature analysis layer is built based on the LSTM model and is used to perform time series analysis on the aviation monitoring time series analysis dataset to construct a time series feature vector; The feature concatenation layer is used to concatenate the spatial feature time series vector and the time series feature vector to construct a time series comprehensive feature vector; the aviation monitoring data analysis label output layer is used to perform a full connection operation on the time series comprehensive feature vector to output the aviation monitoring data analysis label.

2. The aviation monitoring data analysis method according to claim 1, characterized in that: Each aviation monitoring data set in the aviation monitoring time series analysis data set is processed through the spatial distribution feature extraction layer to output a spatial distribution feature matrix, which specifically includes the following steps: Each row in the aviation monitoring data set is regarded as a feature node, and a feature edge is constructed between any two feature nodes. The similarity between the two feature nodes is recorded as the feature edge weight. The aviation monitoring graph structure data is constructed based on all feature nodes and feature edges; the aviation monitoring adjacency matrix is ​​constructed, and the feature edge weights are stored in the aviation monitoring adjacency matrix; Construct an aviation monitoring degree matrix, which is a diagonal matrix; The graph convolution operation is performed on the aviation monitoring dataset using the following formula: G(t+1)=σ[HG(t)W(t)], where G(t) is the aviation monitoring dataset after the tth iteration, t=1, 2, …, U, U is the maximum number of iterations, σ[] is the activation function, H is the normalized aviation monitoring adjacency matrix, and the aviation monitoring adjacency matrix H=E -0.5 RE -0.5 , R is the aviation monitoring adjacency matrix, E is the aviation monitoring degree matrix, W(t) is the graph convolution weight matrix of the tth iteration; the aviation monitoring data set that has undergone U iterations is recorded as the spatial distribution feature matrix.

3. The aviation monitoring data analysis method according to claim 2, characterized in that: In the spatial feature time series analysis layer, the corresponding aviation monitoring data set is processed by the self-attention mechanism based on the spatial distribution weight matrix, and the aviation monitoring data set processed by the self-attention mechanism is subjected to time series analysis to construct the spatial feature time series vector, which specifically includes the following steps: For the corresponding spatial distribution weight matrix and aviation monitoring data set, the aviation monitoring data set is multiplied with the value weight matrix and the key weight matrix respectively to construct the aviation monitoring value vector V and the aviation monitoring key vector K. The spatial distribution weight matrix is ​​multiplied with the query weight matrix to construct the aviation monitoring query vector Q. The self-attention weight matrix ATT=softmax(QK T / (D) 0.5 ), where D is the horizontal dimension size of the aviation monitoring dataset, T is the matrix transpose operation, and then the self-attention weight matrix ATT is multiplied by the aviation monitoring value vector V to obtain the enhanced aviation monitoring dataset; All enhanced aviation monitoring data sets are subjected to time series analysis to output time series analysis graphs, which are then sent to a multi-layer perceptron for processing to construct an inter-feature time series vector.

4. The aviation monitoring data analysis method according to claim 3, characterized in that: Before performing time series analysis on the aviation monitoring time series analysis data set through the time series feature analysis layer, a preprocessing operation on the aviation monitoring time series analysis data set is also included. The specific steps are as follows: for each aviation monitoring data set in the aviation monitoring time series analysis data set, the aviation monitoring data set is flattened into one-dimensional data by row.

5. The aviation monitoring data analysis method according to claim 4, characterized in that: Training the aviation monitoring data analysis model includes the following steps: Acquire several aviation monitoring data analysis training samples, which include N aviation monitoring data sets sorted by time. Label the aviation monitoring data analysis training samples through aviation monitoring data analysis labels, and combine all labeled aviation monitoring data analysis training samples into an aviation monitoring data analysis training set. Send the aviation monitoring data analysis training set to the aviation monitoring data analysis model with initialized parameters for training. Calculate the loss value through the cross entropy loss function, and determine whether the loss value is within a preset range. If the loss value is within the preset range, output the trained aviation monitoring data analysis model; otherwise, continue to train the aviation monitoring data analysis model through the aviation monitoring data analysis training set.

6. The aviation monitoring data analysis method according to claim 5, characterized in that: The similarity between two feature nodes is calculated using the cosine similarity algorithm.

7. An aviation monitoring data analysis system, characterized in that: The system applies an aviation monitoring data analysis method according to any one of claims 1 to 6, including: The aviation monitoring data set acquisition module is used to acquire the aviation monitoring data set at the current monitoring time point. The aviation monitoring data set includes a number of sensor data of different parts of the aircraft, and the aviation monitoring data set is stored in the form of a matrix, where the horizontal dimension represents the type of sensor data and the vertical dimension represents the part of the aircraft; The aviation monitoring data analysis module is used to combine the aviation monitoring data set obtained at the current monitoring time point and the aviation monitoring data set obtained at the previous N-1 detection time points into an aviation monitoring time series analysis data set, and then send the aviation monitoring time series analysis data set to the aviation monitoring data analysis model for processing, and output aviation monitoring data analysis labels; The aviation monitoring data analysis model includes a spatial distribution feature extraction layer, a spatial feature time series analysis layer, a time series feature analysis layer, a feature splicing layer and an aviation monitoring data analysis label output layer, wherein the spatial distribution feature extraction layer is used to process each aviation monitoring data set in the aviation monitoring time series analysis data set to output a spatial distribution feature matrix; the spatial feature time series analysis layer is used to perform a self-attention mechanism on the corresponding aviation monitoring data set based on the spatial distribution weight matrix, and perform time series analysis on the aviation monitoring data set processed by the self-attention mechanism to construct a spatial feature time series vector; the time series feature analysis layer is established based on the LSTM model, and is used to perform time series analysis on the aviation monitoring time series analysis data set to construct a time series feature vector; the feature splicing layer is used to splice the spatial feature time series vector and the time series feature vector to construct a time series comprehensive feature vector; the aviation monitoring data analysis label output layer is used to perform a full connection operation on the time series comprehensive feature vector to output the aviation monitoring data analysis label.

Images