A flight ground support data interpolation method and system
By using fully connected neural networks, interpolation recurrent neural networks, and deconvolutional neural networks to extract features and reduce dimensionality of flight ground support data, and combining this with loss function training of deep neural networks, the problem of lost or corrupted flight ground support data has been solved, achieving accurate data interpolation and integrity, and improving resource utilization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- THE SECOND RES INST OF CIVIL AVIATION ADMINISTRATION OF CHINA
- Filing Date
- 2022-07-01
- Publication Date
- 2026-05-08
AI Technical Summary
Existing technologies are prone to data loss or damage during the collection, transportation, and storage of flight ground support data, resulting in data gaps. Ignoring or simply estimating these missing data can lead to poor model training performance, affecting the accuracy of subsequent data analysis and resource utilization.
Fully connected neural networks, interpolation recurrent neural networks, and deconvolutional neural networks are used to extract features and reduce dimensionality of flight ground support data. A deep neural network is trained using a loss function to achieve accurate data interpolation.
It improved the interpolation accuracy of flight ground support data, ensured the integrity and accuracy of the data, reduced the deviation caused by inaccurate interpolation, and improved the utilization rate of resources.
Smart Images
Figure CN115511249B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of civil aviation data processing methods, specifically to a method and system for interpolating flight ground support data. Background Technology
[0002] The ground handling process is a crucial link connecting ground arrivals and departures with airport infrastructure. In the era of big data, with various types of sensors deployed in every corner of the airport, the dimensions and scale of airport data are increasing exponentially. This massive amount of data can provide airports with information that is beneficial for decision support.
[0003] Most current research on flight support assumes that there is no missing data. However, data is often lost or damaged during collection, transportation, and storage, which is the main reason for most missing data. Generally, research ignores this missing data or uses simple statistical methods to estimate it. This approach will lead to worse model training results and introduce greater bias into the data. Ultimately, these problems will be reflected in subsequent data analysis work such as evaluation and prediction. Therefore, how to accurately imput flight support data is a challenging problem. Summary of the Invention
[0004] To address the shortcomings of existing technologies, this invention provides a method and system for interpolating flight ground support data, which improves the accuracy of flight ground support data interpolation, ensures the integrity of flight ground support data, facilitates subsequent analysis and utilization of flight ground support data, and improves resource utilization.
[0005] In a first aspect, the data interpolation method for flight ground support provided by the present invention includes the following steps:
[0006] Analyze the initial flight ground support data, preprocess the initial flight ground support data to extract the graph structure information of the flight ground support data, and obtain the adjacency matrix of each node of the flight ground support network;
[0007] The feature tensor matrix is obtained by expanding the dimension of the flight ground support data of each node of the flight ground support network using a fully connected neural network. The first feature extraction and dimensionality reduction of the feature tensor matrix is then performed using an interpolation recurrent neural network to obtain the first feature vector.
[0008] A second feature extraction is performed on the hidden layer data of the first feature vector using a recurrent neural network to obtain the feature vector in the hidden layer;
[0009] By downsampling the feature vectors in the hidden layer using a deconvolutional neural network, the original dimensions of the flight ground support data are restored.
[0010] The parameters of a deep neural network are trained using a loss function to obtain the trained deep neural network. The trained deep neural network is then used to interpolate the data and output complete flight ground support data.
[0011] Secondly, the present invention provides an interpolation system for flight ground support data, comprising: an analysis module, a first feature extraction module, a second feature extraction module, a restoration module, and a training module.
[0012] The analysis module is used to analyze the initial flight ground support data, preprocess the initial flight ground support data to extract the graph structure information of the flight ground support data, and obtain the adjacency matrix of each node of the flight ground support network.
[0013] The first feature extraction module is used to expand the dimension of the flight ground support data of each node of the flight ground support network using a fully connected neural network to obtain a feature tensor matrix, and then use an interpolation recurrent neural network to perform the first feature extraction and dimensionality reduction on the feature tensor matrix to obtain the first feature vector.
[0014] The second feature extraction module is used to perform a second feature extraction on the hidden layer data of the first feature vector using a recurrent neural network to obtain the feature vector in the hidden layer.
[0015] The restoration module is used to downsample the feature vectors in the hidden layer through a deconvolutional neural network to restore the original dimensions of the flight ground support data.
[0016] The training module is used to train the parameters of a deep neural network using a loss function to obtain a trained deep neural network. The trained deep neural network is then used to interpolate data and output complete flight ground support data.
[0017] The beneficial effects of this invention are:
[0018] The present invention provides a method and system for interpolating flight ground support data, which improves the quality of flight support data, makes the interpolation accuracy more precise, thereby ensuring the integrity of flight ground support data, which is beneficial to the subsequent analysis and utilization of flight support data, reduces the deviation caused by inaccurate interpolation, and improves the utilization rate of resources. Attached Figure Description
[0019] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.
[0020] Figure 1 This is a flowchart of a method for interpolating flight ground support data provided in the first embodiment of the present invention;
[0021] Figure 2 This is a flowchart illustrating the specific flight ground support service process in the first embodiment of the present invention;
[0022] Figure 3 This is a schematic diagram of the network structure for the feature extraction part of the ground support map structure data of a certain flight in the first embodiment of the present invention;
[0023] Figure 4 This is a schematic diagram of the flight ground support data interpolation model provided in the first embodiment of the present invention;
[0024] Figure 5 This is a schematic diagram of flight ground support data upsampling provided in the first embodiment of the present invention;
[0025] Figure 6 This is a structural block diagram of the flight ground support data interpolation system provided in the second embodiment of the present invention. Detailed Implementation
[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0027] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.
[0028] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.
[0029] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes such combinations.
[0030] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [described condition or event] is detected" may be interpreted, depending on the context, as "once determined," "in response to determination," "once [described condition or event] is detected," or "in response to detection of [described condition or event]."
[0031] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application should have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.
[0032] like Figure 1 The diagram shows a flowchart of a data interpolation method for flight ground support provided in the first embodiment of the present invention. The method includes the following steps:
[0033] S1: Analyze the initial flight ground support data, preprocess the initial flight ground support data to extract the graph structure information of the flight ground support data, and obtain the adjacency matrix of each node of the flight ground support network;
[0034] S2: The flight ground support data of each node of the flight ground support network is expanded in dimension using a fully connected neural network to obtain a feature tensor matrix. The feature tensor matrix is then subjected to the first feature extraction and dimensionality reduction using an interpolation recurrent neural network to obtain the first feature vector.
[0035] S3: Use a recurrent neural network to perform a second feature extraction on the hidden layer data of the first feature vector to obtain the feature vector in the hidden layer;
[0036] S4: The feature vectors in the hidden layer are downsampled by a deconvolutional neural network to restore the original dimensions of the flight ground support data;
[0037] S5: The parameters of the deep neural network are trained using a loss function to obtain the trained deep neural network. The trained deep neural network is then used to interpolate the data and output complete flight ground support data.
[0038] In this embodiment, the initial flight ground support process is first analyzed, treating the initial flight ground support data as graph-structured data. Graph structure recognition is performed on the flight ground support data to facilitate subsequent analysis. A spatial domain-based convolutional method is used to initially extract and complete the data to obtain a first feature vector. A recurrent neural network is applied to further analyze and extract the time dimension features of the first feature vector to obtain the hidden layer feature vector. Based on this, a deconvolutional neural network structure is used to downsample the intermediate hidden layer feature vectors to restore them to their original dimensions. A loss function that simultaneously considers the actual differences and distribution differences in the data is designed to train the model's neural network. This method improves the accuracy of flight ground support data interpolation, thereby ensuring data integrity, which is beneficial for subsequent analysis and utilization of flight ground support data, reduces the bias caused by inaccurate interpolation, and improves resource utilization.
[0039] like Figure 2 As shown, flight ground support service is a dynamic and complex process dominated by parallel workflows, mainly referring to a series of support operations during the time from wheel chock loading to wheel chock removal. Based on the actual flight ground support service process, it can be roughly divided into four main parallel processes: fuel replenishment, cabin service, cargo handling, and aircraft maintenance inspection. The cabin service process is the most complex, encompassing catering service, cabin cleaning, waste disposal, and decontamination operations. These processes not only involve sequential relationships but also a certain logical order.
[0040] In flight ground support services, timestamp data of each node is recorded. This node data forms a data matrix. However, the underlying relationships between the nodes need to be further processed to obtain the adjacency matrix of its graph structure. The Euclidean distance calculation formula is as follows:
[0041] (1)
[0042] d 12 —The European distance between two points in the flight support process
[0043] x and y represent the x and y coordinates of the point, respectively. The formula for determining whether there is a connection between two points is:
[0044] (2)
[0045] Where ρ is a priori parameter that is adjusted based on experience. If the distance between two points satisfies this formula, then the two points are considered to be connected, that is, the corresponding coordinate of the adjacency matrix is 1.
[0046] Since flight ground support data only has time as a feature, the data is augmented before the first feature extraction. A fully connected neural network is used to augment the dimensionality of the flight ground support data, resulting in a feature tensor matrix: ,in, This indicates that d nodes are at time t. i The guaranteed node feature data is used here, and we expand the dimension of each node d.
[0047] During flight ground support services, special vehicles should arrive in a timely manner and carry out service operations in the prescribed order. If vehicles do not arrive in a timely manner, it will cause a delay in a certain service, which will then affect subsequent service operations and create a ripple effect.
[0048] Spatial domain-based convolutional strategies determine the update relationships of nodes at different layers based on the information transmission relationships between nodes in the network. Based on the actual operational situation of flight support, a feature update strategy was determined, such as... Figure 3 As shown, the information transmission paths were determined based on the flight support process. Each path was analyzed and updated individually. The intersection of these paths was averaged. A recurrent neural network was used to perform preliminary feature extraction on the flight ground support network data. Since the data was incomplete at this stage, a masking matrix was used. Its size and scale are equivalent to the feature data, and the calculated tensor formula is: (3) The formula for calculating the missing data rate is: (4),
[0049] In equation (4), M represents the masking matrix, d represents the feature node, and a time label matrix is introduced into the deep neural network. The deep neural network can quantize the length in the time dimension. Initially, All are 0, different values are in It can be expressed as the following formula:
[0050] (5)
[0051] In deep neural networks, a decay rate is introduced to control the influence of the previously observed values. The formula for calculating the decay rate is as follows:
[0052] (6)
[0053] in, and These are all parameters of the model. The negative power of the exponential function ensures that the decay rate monotonically decreases from 0 to 1. The calculated decay rate is used to approximate the data to the average value. The final trained parameters are used to calculate the first feature vector using the following formula:
[0054] (7)
[0055] in, Let represent the vector of the previous observation node, and x represent the arithmetic mean of the variables. To apply the effect of this decay to the hidden layers, the network update function can be represented by the following process:
[0056] (8)
[0057] (9)
[0058] (10)
[0059] (11)
[0060] (12)
[0061] (13)
[0062] in, and It is a trainable parameter function that takes the hidden state from the previous time step. Passed to an estimable vector Z represents the update door, and R represents the reset door; these are alternative hidden states. It is an activation function, and another matrix , , , , , , , , sum vector , , These are all parameters to be trained.
[0063] The specific method for performing a second feature extraction on the hidden layer data of the first feature vector using a recurrent neural network includes: considering the correlation between data in the time dimension, using a long short-term neural network to extract features from the first feature vector to obtain the final feature vector in the hidden layer. The feature vector in the hidden layer also considers the spatiotemporal correlation between the data.
[0064] like Figure 4As shown, based on the above design, a method for mimicking the information transmission process and feature node update in flight support procedures can initially obtain the extracted feature nodes. To further understand the node update mechanism in the time dimension, a recurrent neural network is used to update the nodes, and its update formula is as follows:
[0065] (14)
[0066] (15)
[0067] (16)
[0068] (17)
[0069] (18)
[0070] (19)
[0071] in, It is the output of the previous hidden layer, and the dimension of the input features. This refers to the hidden state. Before training the entire network, we need to define the size of the matrices within it. Size and The size of the main layer and the size of the hidden layer are the same.
[0072] Based on the spatial convolution strategy using graph structures and the control of the temporal dimension by recurrent neural networks, feature vectors in the hidden layers were obtained. To restore the flight ground support data to its original dimension, a deconvolutional neural network structure was designed to downsample the feature vectors in the hidden layers, such as... Figure 5 As shown, the process is as follows:
[0073] (20)
[0074] (twenty one)
[0075] Where X represents the sampled raw data, and A is the adjacency matrix of each node in the flight ground support network. A loss function is designed as needed to adjust the network parameters, ultimately outputting more accurate and complete flight ground support data. Specifically, traditional networks only consider the distance between real and generated data during training. However, when the data scale is large, this loss function design often has shortcomings. Therefore, an adversarial loss is introduced to compare the data distribution. Simultaneously, to synthesize the two loss functions, a hyperparameter λ is introduced to balance their relationship. In this embodiment, the loss functions include an information loss function, an adversarial loss function, and a data imputation loss function.
[0076] The formula for the information loss function is as follows:
[0077] (twenty two)
[0078] The formula for calculating the adversarial loss function is:
[0079] (twenty three)
[0080] The formula for calculating the data interpolation loss function is:
[0081] (twenty four)
[0082] Where X is the original sampled data, and M represents the masking matrix. h represents the interpolated data. adj Let represent nodes that are adjacent to h at a fixed distance, σ be the activation function, P() represent the negative sampling distribution, Q represent the number of negative samples, and λ be a hyperparameter controlling the ratio between the information loss function and the adversarial loss function. By adjusting the parameters of the deep neural network through the loss function, the adjusted neural network is used to interpolate the data, resulting in a more accurate and complete output of flight ground support data.
[0083] The first embodiment of the present invention provides a flight ground support data interpolation method, which improves the quality of flight support data, makes the data interpolation accuracy more precise, thereby ensuring the integrity of flight ground support data, which is beneficial to the subsequent analysis and utilization of flight support data, reduces the deviation caused by inaccurate interpolation, and improves the utilization rate of resources.
[0084] In the first embodiment described above, a method for interpolating flight ground support data is provided. Correspondingly, this application also provides a system for interpolating flight ground support data. Please refer to... Figure 6 This is a structural block diagram of a flight ground support data interpolation method provided in the second embodiment of the present invention. Since the device embodiment is basically similar to the method embodiment, it is described simply; relevant details can be found in the description of the method embodiment. The device embodiment described below is merely illustrative.
[0085] like Figure 6The diagram illustrates a structural block diagram of a flight ground support data interpolation system according to a second embodiment of the present invention. The system includes an analysis module, a first feature extraction module, a second feature extraction module, a restoration module, and a training module. The analysis module analyzes initial flight ground support data, preprocesses the initial flight ground support data to extract graph structure information, and obtains the adjacency matrix of each node in the flight ground support network. The first feature extraction module uses a fully connected neural network to expand the dimension of the flight ground support data of each node in the flight ground support network to obtain a feature tensor matrix. An interpolation recurrent neural network is then used to perform first feature extraction and dimensionality reduction on the feature tensor matrix to obtain a first feature vector. The second feature extraction module uses a recurrent neural network to perform second feature extraction on the hidden layer data of the first feature vector to obtain feature vectors in the hidden layer. The restoration module uses a deconvolutional neural network to downsample the feature vectors in the hidden layer to restore the original dimension of the flight ground support data. The training module uses a loss function to train the parameters of a deep neural network to obtain a trained deep neural network. The trained deep neural network is then used to perform data interpolation and output complete flight ground support data.
[0086] In this embodiment, the analysis module analyzes the flight ground support process, treating the flight ground support data as data with a graph structure, and performs graph structure recognition on the flight support data to facilitate subsequent analysis. The first feature extraction module uses a spatial domain convolution method to initially extract and complete the data to obtain the first feature vector. The second feature extraction module applies a recurrent neural network to further analyze and extract the time dimension features of the first feature vector to obtain the hidden layer feature vector. Based on this, the restoration module uses a deconvolutional neural network structure to downsample the intermediate hidden layer feature vectors to restore them to their original dimensions. The training module designs a loss function that simultaneously considers the actual differences and distribution differences in the data to train the neural network of the model.
[0087] The flight ground support data interpolation system provided in this invention improves the quality of flight support data, makes the data interpolation accuracy more precise, thereby ensuring the integrity of flight ground support data, which is beneficial for subsequent analysis and utilization of flight support data, reduces the deviation caused by inaccurate interpolation, and improves resource utilization.
[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.
Claims
1. A data interpolation method for flight ground support, characterized in that, Includes the following steps: Analyze the initial flight ground support data, preprocess the initial flight ground support data to extract the graph structure information of the flight ground support data, and obtain the adjacency matrix of each node of the flight ground support network; The feature tensor matrix is obtained by expanding the dimension of the flight ground support data of each node of the flight ground support network using a fully connected neural network. The first feature extraction and dimensionality reduction of the feature tensor matrix is then performed using an interpolation recurrent neural network to obtain the first feature vector. A second feature extraction is performed on the hidden layer data of the first feature vector using a recurrent neural network to obtain the feature vector in the hidden layer; By downsampling the feature vectors in the hidden layer using a deconvolutional neural network, the original dimensions of the flight ground support data are restored. The parameters of a deep neural network are trained using a loss function to obtain the trained deep neural network. The trained deep neural network is then used to interpolate the data and output complete flight ground support data.
2. The method as described in claim 1, characterized in that, The specific method for preprocessing initial flight ground support data to extract graph structure information of the flight ground support data includes: The adjacency matrix G(X,A) of the flight ground support network is determined using the Euclidean distance threshold method. The Euclidean distance calculation formula is as follows: （1） d 12 This represents the Euclidean distance between two points in the flight ground support network. x 1. x 2 represents the x-coordinate of the two points. y 1. y 2 The formula for determining whether there is a connection between two points is: (This represents the ordinates of the two points.) （2） Where ρ is a priori parameter, which is adjusted based on experience. If the distance between two points satisfies equation (2), then it is considered that there is a connection between the two points.
3. The method as described in claim 2, characterized in that, The specific method for using a fully connected neural network to augment the dimensionality of flight ground support data at each node of the flight ground support network to obtain the feature tensor matrix includes: The flight ground support network is time-series. Each time point corresponds to a network connection graph for flight ground support. A fully connected neural network is used to expand the feature nodes of each flight support network at each time point to obtain a feature tensor matrix. ,in, This indicates that d nodes are at time t. i The following are the characteristics of the protected nodes.
4. The method as described in claim 3, characterized in that, The specific method for performing the first feature extraction and dimensionality reduction of the feature tensor matrix using an interpolation recurrent neural network includes: The tensor and data missing rate are calculated using an occlusion matrix of the same size and scale as the initial flight ground support data. The occlusion matrix is: , The resulting tensor formula is: （3） The formula for calculating the missing data rate is: （4） In equation (4), M represents the masking matrix, d represents the feature node, and a time label matrix is introduced into the deep neural network. The deep neural network can quantize the length in the time dimension, starting at... All are 0, different values are in It can be expressed as the following formula: （5） In deep neural networks, a decay rate is introduced to control the influence of the previously observed values. The formula for calculating the decay rate is as follows: （6） In equation (6), and These are all parameters of the model. The negative power of the exponential function ensures that the decay rate monotonically decreases from 0 to 1. The calculated decay rate is used to approximate the data to the average value. The final trained parameters are used to calculate the first feature vector using the following formula: （7） In equation (7), Let x represent the vector of the previous observation node, and let x represent the arithmetic mean of the variables.
5. The method as described in claim 4, characterized in that, The specific method for performing a second feature extraction on the hidden layer data of the first feature vector using a recurrent neural network includes: Considering the correlation of data over time, a long short-term neural network is used to extract features from the first feature vector to obtain the feature vector in the final hidden layer. The feature vector in the hidden layer also takes into account the spatiotemporal correlation of the data.
6. The method as described in claim 5, characterized in that, The loss function includes an information loss function, an adversarial loss function, and a data imputation loss function. The formula for calculating the information loss function is as follows: （22） The formula for calculating the adversarial loss function is: （23） The formula for calculating the data interpolation loss function is: （24） Where X is the original sampled data, and M represents the masking matrix. h represents the interpolated data. adj Let h represent nodes that are adjacent to h at a fixed distance, σ be the activation function, P() represent the negative sampling distribution, Q represent the number of negative samples, and λ be a hyperparameter that controls the ratio between the information loss function and the adversarial loss function.
7. An interpolation system for flight ground support data, characterized in that, include: The module consists of an analysis module, a first feature extraction module, a second feature extraction module, a restoration module, and a training module. The analysis module is used to analyze the initial flight ground support data, preprocess the initial flight ground support data to extract the graph structure information of the flight ground support data, and obtain the adjacency matrix of each node of the flight ground support network. The first feature extraction module is used to expand the dimension of the flight ground support data of each node of the flight ground support network using a fully connected neural network to obtain a feature tensor matrix, and then use an interpolation recurrent neural network to perform the first feature extraction and dimensionality reduction on the feature tensor matrix to obtain the first feature vector. The second feature extraction module is used to perform a second feature extraction on the hidden layer data of the first feature vector using a recurrent neural network to obtain the feature vector in the hidden layer. The restoration module is used to downsample the feature vectors in the hidden layer through a deconvolutional neural network to restore the original dimensions of the flight ground support data. The training module is used to train the parameters of a deep neural network using a loss function to obtain a trained deep neural network. The trained deep neural network is then used to interpolate data and output complete flight ground support data.