CN-115511249-B - Interpolation method and system for flight ground guarantee data

CN115511249BCN 115511249 BCN115511249 BCN 115511249BCN-115511249-B

Abstract

The invention discloses a data interpolation method and system for flight ground assurance, wherein the method comprises the steps of preprocessing initial flight ground assurance data to obtain an adjacent matrix of each node of a flight ground assurance network, performing dimension expansion on the ground flight assurance data by utilizing a fully-connected neural network to obtain a feature tensor matrix, performing first feature extraction and dimension reduction on the feature tensor matrix by utilizing an interpolation recurrent neural network to obtain a first feature vector, performing second feature extraction on the first feature vector by utilizing the recurrent neural network to obtain a feature vector in a hidden layer, performing downsampling on the feature vector in the hidden layer by utilizing a deconvolution neural network to restore original dimensions, training deep neural network parameters by adopting a loss function, performing interpolation on the trained deep neural network to obtain complete flight ground assurance data. The quality of the flight guarantee data is improved, the interpolation precision of the data is more accurate, and the integrity of the flight ground guarantee data is ensured.

Inventors

LUO QIAN
YOU YI
XIA HUAN
WEN TAO
LIU CHANG
SUN KE
ZHANG TAO
ZHANG XINGRUI
DU YUXIAN
LV MING
DENG QIANGQIANG
ZHANG ZHIWEI

Assignees

中国民用航空总局第二研究所

Dates

Publication Date: 20260508
Application Date: 20220701

Claims (7)

1. The data interpolation method for the flight ground guarantee is characterized by comprising the following steps of: Analyzing the initial flight ground guarantee data, preprocessing the initial flight ground guarantee data, extracting the graph structure information of the flight ground guarantee data, and obtaining an adjacency matrix of each node of the flight ground guarantee network; Performing dimension expansion on flight ground guarantee data of each node of a flight ground guarantee network by using a fully-connected neural network to obtain a feature tensor matrix, and performing first feature extraction and dimension reduction on the feature tensor matrix by using an interpolation recurrent neural network to obtain a first feature vector; Performing second feature extraction on the hidden layer data of the first feature vector by using a recurrent neural network to obtain a feature vector in the hidden layer; downsampling the feature vectors in the hidden layer through the deconvolution neural network to restore the original dimension of the flight ground guarantee data; training parameters of the deep neural network by adopting a loss function to obtain a trained deep neural network, and interpolating data by using the trained deep neural network to output complete flight ground guarantee data.
2. The method of claim 1, wherein the specific method for preprocessing the initial flight ground assurance data to extract the graphic structure information of the flight ground assurance data comprises the following steps: Determining an adjacency matrix G (X, A) of the ground guarantee network of the flight by adopting an Euclidean distance threshold value judging method, wherein the Euclidean distance calculating formula is as follows: (1) d 12 represents the Euclidean distance between two points of the ground guarantee network of the flight, x 1 、x 2 represents the abscissa of the two points, y 1 、y 2 represents the ordinate of the two points, and the formula for judging whether the connection exists between the two points is as follows: (2) where ρ is an a priori parameter, adjusted empirically, and if the distance between two points satisfies equation (2), then a connection is considered to exist between the two points.
3. The method as claimed in claim 2, wherein the specific method for performing dimension expansion on the flight ground guarantee data of each node of the flight ground guarantee network by using the fully-connected neural network to obtain the feature tensor matrix comprises the following steps: The flight ground guarantee network is a time sequence A network connection diagram corresponding to a flight ground guarantee under each time node is obtained, and the feature nodes of each flight ground guarantee network under each time node are expanded by using a fully-connected neural network to obtain a feature tensor matrix , wherein, And representing the feature data of the guaranteed nodes of the d nodes at time t i .
4. The method of claim 3, wherein the specific method for performing the first feature extraction and dimension reduction on the feature tensor matrix by using the interpolation recurrent neural network comprises: Calculating tensor and data missing rate by adopting a shielding matrix with the same size and scale as those of the ground guarantee data of the initial flight, wherein the shielding matrix is , The tensor formula obtained is: (3) the calculation formula of the missing rate of the data is as follows: (4) in the formula (4), M represents a covering matrix, d represents characteristic nodes, and a time tag matrix is introduced into the deep neural network The deep neural network can quantify the length in the time dimension direction, and at the beginning All being 0, different values being Expressed as the following formula: (5) The decay rate is introduced into the deep neural network for controlling the influence of the pre-observable value, and the calculation formula of the decay rate is as follows: (6) In the formula (6), the amino acid sequence of the compound, And The method is characterized in that the method is a parameter of a model, the negative power of an exponential function ensures that the decay rate is monotonically decreased from 0 to 1, the calculated decay rate is utilized to approach data to an average value, and finally, a formula for calculating a first eigenvector by the trained parameter is as follows: (7) In the formula (7), the amino acid sequence of the compound, A vector representing the last observation node, x representing the arithmetic mean of the variables.
5. The method of claim 4, wherein the specific method for performing the second feature extraction on the hidden layer data of the first feature vector by using the recurrent neural network comprises: And taking the association relation in the data time dimension into consideration, carrying out feature extraction on the first feature vector by adopting a long-short-period neural network to obtain the feature vector in the final hidden layer, wherein the feature vector of the hidden layer simultaneously takes the space-time association relation of the data into consideration.
6. The method of claim 5, wherein the loss function comprises an information loss function, an antagonism loss function, and a data interpolation loss function, wherein the information loss function has a calculation formula: (22) the calculation formula of the counterattack loss function is as follows: (23) The calculation formula of the data interpolation loss function is as follows: (24) where X is the raw data of the sample, M represents the masking matrix, Indicating the interpolated data, h adj indicating the node adjacent to h by a fixed distance, σ being the activation function, P () indicating the negative sampling distribution, Q indicating the number of negative samples, λ being the ratio between a hyper-parametric control information loss function and a counterloss function.
7. The interpolation system of the flight ground guarantee data is characterized by comprising 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 for analyzing the initial flight ground guarantee data, preprocessing the initial flight ground guarantee data, extracting the graph structure information of the flight ground guarantee data, and obtaining the adjacency matrix of each node of the flight ground guarantee network; The first-time feature extraction module is used for performing dimension expansion on flight ground guarantee data of each node of the flight ground guarantee network by using the fully-connected neural network to obtain a feature tensor matrix, and performing first-time feature extraction and dimension reduction on the feature tensor matrix by using the interpolation recurrent neural network to obtain a first feature vector; The second feature extraction module is used for carrying out second feature extraction on the hidden layer data of the first feature vector by using the recurrent neural network to obtain the feature vector in the hidden layer; the restoration module is used for downsampling the feature vectors in the hidden layer through the deconvolution neural network and restoring the original dimension of the ground guarantee data of the flight; The training module is used for training the parameters of the deep neural network by adopting the loss function to obtain the trained deep neural network, and the trained deep neural network is used for realizing the interpolation of data and outputting complete flight ground guarantee data.

Description

Interpolation method and system for flight ground guarantee data Technical Field The invention relates to the technical field of civil aviation data processing methods, in particular to an interpolation method and system of flight ground guarantee data. Background The flight ground guarantee process is a key link of associating a ground departure and arrival with an airport infrastructure, and in a big data age, as various types of sensors are arranged at any corner of an airport, the data dimension and the scale of the airport are multiplied, and a large amount of data can provide data information beneficial to decision support for the airport. At present, under the condition that most of flight guarantee researches assume that data is not missing, however, the data is usually lost and damaged in the process of collecting, transporting and storing, which is also a main reason for most of missing data, general researches ignore the missing data or estimate the missing data by a simple statistical method, the missing data can cause a worse model training effect, and larger deviation is introduced into the data, and finally, the problems can reflect the follow-up evaluation, prediction and other data analysis work, so that how to accurately interpolate the flight guarantee data is a challenging problem. Disclosure of Invention Aiming at the defects in the prior art, the invention provides the interpolation method and the system for the flight ground guarantee data, which improve the accuracy of the flight ground guarantee data interpolation, ensure the integrity of the flight ground guarantee data, facilitate the subsequent analysis and utilization of the flight ground guarantee data and improve the utilization rate of resources. In a first aspect, the data interpolation method for flight ground assurance provided by the invention comprises the following steps: Analyzing the initial flight ground guarantee data, preprocessing the initial flight ground guarantee data, extracting the graph structure information of the flight ground guarantee data, and obtaining an adjacency matrix of each node of the flight ground guarantee network; Performing dimension expansion on flight ground guarantee data of each node of a flight ground guarantee network by using a fully-connected neural network to obtain a feature tensor matrix, and performing first feature extraction and dimension reduction on the feature tensor matrix by using an interpolation recurrent neural network to obtain a first feature vector; Performing second feature extraction on the hidden layer data of the first feature vector by using a recurrent neural network to obtain a feature vector in the hidden layer; downsampling the feature vectors in the hidden layer through the deconvolution neural network to restore the original dimension of the flight ground guarantee data; training parameters of the deep neural network by adopting a loss function to obtain a trained deep neural network, and interpolating data by using the trained deep neural network to output complete flight ground guarantee data. In a second aspect, the invention provides an interpolation system of flight ground guarantee data, which comprises 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 for analyzing the initial flight ground guarantee data, preprocessing the initial flight ground guarantee data, extracting the graph structure information of the flight ground guarantee data, and obtaining the adjacency matrix of each node of the flight ground guarantee network; The first-time feature extraction module is used for performing dimension expansion on flight ground guarantee data of each node of the flight ground guarantee network by using the fully-connected neural network to obtain a feature tensor matrix, and performing first-time feature extraction and dimension reduction on the feature tensor matrix by using the interpolation recurrent neural network to obtain a first feature vector; The second feature extraction module is used for carrying out second feature extraction on the hidden layer data of the first feature vector by using the recurrent neural network to obtain the feature vector in the hidden layer; the restoration module is used for downsampling the feature vectors in the hidden layer through the deconvolution neural network and restoring the original dimension of the ground guarantee data of the flight; The training module is used for training the parameters of the deep neural network by adopting the loss function to obtain the trained deep neural network, and the trained deep neural network is used for realizing the interpolation of data and outputting complete flight ground guarantee data. The invention has the beneficial effects that: the interpolation method and the system for the flight ground guarantee data improve the quality of the flight guarantee data,