CN112489497A - Airspace operation complexity evaluation method based on deep convolutional neural network - Google Patents
Airspace operation complexity evaluation method based on deep convolutional neural network Download PDFInfo
- Publication number
- CN112489497A CN112489497A CN202011294477.7A CN202011294477A CN112489497A CN 112489497 A CN112489497 A CN 112489497A CN 202011294477 A CN202011294477 A CN 202011294477A CN 112489497 A CN112489497 A CN 112489497A
- Authority
- CN
- China
- Prior art keywords
- airspace
- operation complexity
- sector
- image
- air traffic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G5/00—Traffic control systems for aircraft, e.g. air-traffic control [ATC]
- G08G5/003—Flight plan management
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Aviation & Aerospace Engineering (AREA)
- Image Analysis (AREA)
Abstract
The invention particularly relates to a spatial domain operation complexity evaluation method based on a deep convolutional neural network, which comprises the following steps of: extracting sector dynamic traffic data of a target airspace sector, and marking airspace operation complexity grades; defining a circumscribed rectangle of a target airspace sector, and carrying out gridding treatment; constructing a multi-channel air traffic situation image, and constructing an air traffic situation image library according to the operation complexity grade of an airspace; constructing a spatial domain operation complexity hierarchical network model according to the multi-channel air traffic situation image; training a space domain operation complexity hierarchical network model; and the airspace operation complexity evaluation is carried out according to the trained airspace operation complexity hierarchical network model, so that the most relevant characteristics can be automatically learned from the original data in an end-to-end mode on the premise of not depending on the complexity relevant characteristics, the establishment of the airspace operation complexity hierarchical network model is assisted, and the workload and the use threshold of the airspace complexity evaluation are greatly reduced.
Description
Technical Field
The invention belongs to the technical field of airspace situation assessment of air traffic control, and particularly relates to an airspace operation complexity assessment method based on a deep convolutional neural network.
Background
With the development of the air transportation industry, the sharply increased air traffic flight flow and limited airspace resources bring enormous workload and pressure to air traffic controllers. The airspace operation complexity is a key index for evaluating the workload of controllers, and meanwhile, the airspace operation complexity also can provide decision support for strategic and tactical air traffic management systems, so that how to determine a scientific, accurate and reliable method for evaluating the airspace operation complexity is one of the widely researched problems. In recent years, some scholars adopt a machine learning method based on manual features to solve the problem of evaluating the complexity of airspace operation, and the main research idea is as follows: the method comprises the steps of constructing a feature system influencing the operation complexity of an airspace sector, simultaneously collecting airspace operation complexity labels under different air traffic scenes, training and learning a mapping relation model between a plurality of airspace operation complexity features and complexity labels through a machine learning algorithm, and carrying out the estimation work of the airspace operation complexity by utilizing the trained machine learning model. However, the characteristics influencing the airspace operation complexity are numerous and complex, a uniformly recognized airspace operation complexity characteristic system does not exist at present, in practice, the selection of related characteristics is greatly influenced by subjectivity, and an incomplete or improper complexity characteristic system can seriously influence the performance of the airspace operation complexity evaluation model based on the machine learning algorithm.
Aiming at the problems, a unified and comprehensive airspace operation complexity feature system is lacked, so that a machine learning model is difficult to learn an evaluation model with excellent performance through a defective feature set, and a computer is expected to replace a researcher to complete automatic generation or selection of airspace operation complexity features. In deep learning, the features of a data set can be unknown, and the goal is to perform automatic feature learning through a deep convolutional neural network under the guidance of a label to mine rich feature information of original data and provide good feature vector input for a subsequent classification or regression model.
In the prior art, the spatial domain operation complexity characteristics are manually selected, although a machine learning algorithm model can learn the mapping relation between different characteristics and the spatial domain operation complexity, the mapping relation is not perfect due to possible defects of the characteristic set, and the selection and the use of the characteristic set are greatly influenced by actual scenes and professional knowledge.
Therefore, a new spatial domain operation complexity evaluation method based on the deep convolutional neural network needs to be designed based on the technical problems.
Disclosure of Invention
The invention aims to provide a spatial domain operation complexity evaluation method based on a deep convolutional neural network.
In order to solve the technical problem, the invention provides a spatial domain operation complexity evaluation method based on a deep convolutional neural network, which comprises the following steps:
extracting sector dynamic traffic data of a target airspace sector, and marking airspace operation complexity grades;
defining a circumscribed rectangle of a target airspace sector, and carrying out gridding treatment;
constructing a multi-channel air traffic situation image, and constructing an air traffic situation image library according to the operation complexity grade of an airspace;
constructing a spatial domain operation complexity hierarchical network model according to the multi-channel air traffic situation image;
training a space domain operation complexity hierarchical network model; and
and performing space domain operation complexity evaluation according to the trained space domain operation complexity hierarchical network model.
Further, the method for extracting the sector dynamic traffic data of the target airspace sector and marking the airspace operation complexity level comprises the following steps:
acquiring original air traffic operation data of a target airspace sector, and extracting sector dynamic traffic data of the target airspace sector from the original operation data according to a preset time granularity period;
and dividing the dynamic traffic data of the sector according to a preset time period, and marking the complexity grade of the airspace operation on the dynamic traffic data of the sector corresponding to each time period.
Further, the method for defining the circumscribed rectangle of the target airspace sector and performing gridding processing comprises the following steps:
the method comprises the steps of obtaining sector boundary point longitude and latitude data of a target airspace sector, determining a minimum external rectangle of the target airspace sector, expanding each side of the minimum external rectangle outwards by a preset length respectively to form the target airspace sector external rectangle, and carrying out gridding processing on the target airspace sector external rectangle according to preset length intervals.
Further, the method for constructing the multi-channel air traffic situation image and constructing the air traffic situation image library according to the airspace operation complexity level label comprises the following steps:
according to the dynamic traffic data of the sectors at each time period, taking the longitude and latitude of the aircraft as coordinates, acquiring the position of the corresponding coordinates in the external rectangle of the meshed target airspace sector, filling the altitude motion parameters of the aircraft in the dynamic traffic data of the sectors at each time period into the corresponding grids as pixel values to generate an altitude historical track image channel at the corresponding time period, and filling the speed parameters of the aircraft into the external rectangle of the meshed target airspace sector to generate a speed historical track image channel;
acquiring a predicted point which is reached by the aircraft after preset time according to the longitude and latitude, the speed and the course of the last track point of each aircraft in a time period, connecting sector dynamic traffic data between the last track point and the predicted point to acquire a predicted track, mapping the predicted track to a meshed target airspace sector external rectangle through the longitude and latitude, and when the track passes through a new mesh, sequentially decreasing the filling value by a preset step length until the filling of the last point of the predicted track is completed to generate a conflict predicted track image channel;
and forming a multi-channel air traffic situation image according to the height historical track image channel, the speed historical track image channel and the conflict prediction track image channel, associating the multi-channel air traffic situation image generated at different time intervals with the airspace operation complexity level to obtain an air traffic situation image library, and dividing the air traffic situation image library into a training data set and a testing data set.
Further, the method for constructing the airspace operation complexity hierarchical network model according to the multi-channel air traffic situation image comprises the following steps:
constructing a spatial domain operation complexity hierarchical network model according to the multi-channel air traffic situation image through a deep convolutional neural network, namely
Constructing a fourteen-layer deep convolutional neural network model;
the first layer is an input layer and inputs multi-channel air traffic situation images; the second, third, fifth, sixth, eighth and ninth layers are convolution layers; the fourth, seventh and tenth layers are pooling layers; the eleventh layer is a rolled layer; the twelfth, thirteenth and fourteenth layers are full connection layers, and the output is a space domain operation complexity level vector;
the second layer and the third layer of convolution layers comprise 32 convolution kernels, the fifth layer and the sixth layer of convolution layers comprise 64 convolution kernels, the eighth layer and the ninth layer of convolution layers comprise 128 convolution kernels, and convolution calculation is carried out in an SAME filling mode according to the size of a preset convolution kernel and the movement amplitude of the preset convolution kernel;
the pooling layers of the fourth, seventh and tenth layers are subjected to pooling treatment by adopting a maximum pooling mode;
the twelfth, thirteenth and fourteenth layers are full connection layers, probability representation is carried out on the fourteenth layer output through a Softmax function, and the classification with the maximum probability value is selected as a final classification result;
the Softmax function is:
wherein i is a airspace operation complexity grade category, and i is 1, which indicates that the airspace operation complexity is 1 grade; k represents a natural number greater than zero; n is the total level number of the airspace operation complexity;
the output of the fourteenth layer is a 5-dimensional vector, and each dimension represents the probability of the airspace operation complexity belonging to the level;
and the nonlinear function used after the second, third, fifth, sixth, eighth and ninth convolution layers and the twelfth, thirteenth and fourteenth full connection layers is subjected to nonlinear transformation.
Further, the method for training the space-domain operation complexity hierarchical network model comprises the following steps:
preprocessing an image in the training data set, and performing image standardization processing on pixel values of the image:
wherein mu is the mean value of the image; x is an image matrix; σ is the standard deviation; p is the number of pixels of the image;
putting the preprocessed training data set into the airspace operation complexity hierarchical network model for training;
in the training process, the target loss function is cross entropy:
where y represents the true probability distribution of the image class;
representing the probability distribution calculated by the neural network; y isjAnd
each represents a probability value of the jth dimension in the 5-dimensional vector;
and continuously optimizing the target loss function by a random optimization method in the training process.
Further, the method for performing the spatial domain operation complexity evaluation according to the trained spatial domain operation complexity hierarchical network model comprises the following steps:
and preprocessing the images in the test data set, inputting the images after the preprocessing in the test data set into the trained airspace operation complexity grading network model to obtain an airspace operation complexity grading result, and finishing the airspace operation complexity evaluation.
The method has the advantages that the sector dynamic traffic data of the target airspace sector are extracted, and airspace operation complexity level marking is carried out; defining a circumscribed rectangle of a target airspace sector, and carrying out gridding treatment; constructing a multi-channel air traffic situation image, and constructing an air traffic situation image library according to the operation complexity grade of an airspace; constructing a spatial domain operation complexity hierarchical network model according to the multi-channel air traffic situation image; training a space domain operation complexity hierarchical network model; and the airspace operation complexity evaluation is carried out according to the trained airspace operation complexity hierarchical network model, so that the selection of the most relevant characteristics can be directly learned from the original data in an end-to-end mode without depending on the complexity relevant characteristics, the establishment of the airspace operation complexity hierarchical network model is assisted, and the workload and the use threshold of the airspace complexity evaluation are greatly reduced.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and drawings.
In order to make the aforementioned and other objects, features and advantages of the present invention comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a flow chart of a spatial domain operation complexity evaluation method based on a deep convolutional neural network according to the present invention;
FIG. 2 is a specific flowchart of the spatial domain operation complexity evaluation method based on the deep convolutional neural network according to the present invention;
FIG. 3(a) is a schematic diagram of a high historical track path in accordance with the present invention;
FIG. 3(b) is a schematic diagram of a speed history track path in accordance with the present invention;
FIG. 4(a) is a schematic diagram of a collision-free scenario in a collision-predicted track path according to the present invention;
FIG. 4(b) is a schematic diagram of a collision scenario in a collision prediction track channel according to the present invention;
FIG. 5 is a schematic diagram of a deep convolutional neural network architecture in accordance with the present invention;
FIG. 6 is a graph of evaluation accuracy performance index and loss function convergence according to the present invention.
Detailed Description
To make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
FIG. 1 is a flow chart of a spatial domain operation complexity evaluation method based on a deep convolutional neural network according to the present invention.
As shown in fig. 1, the present embodiment provides a spatial domain operation complexity evaluation method based on a deep convolutional neural network, including: extracting sector dynamic traffic data of a target airspace sector, and marking airspace operation complexity grades; defining a circumscribed rectangle of a target airspace sector, and carrying out gridding treatment; constructing a multi-channel air traffic situation image, and constructing an air traffic situation image library according to the operation complexity grade of an airspace; constructing a spatial domain operation complexity hierarchical network model according to the multi-channel air traffic situation image; training a space domain operation complexity hierarchical network model; the airspace operation complexity evaluation is carried out according to the trained airspace operation complexity hierarchical network model, the selection of relevant features independent of complexity is realized, and the most relevant features can be directly learned from the original data in an end-to-end mode, so that the establishment of the airspace operation complexity hierarchical network model is assisted, and the workload and the use threshold of the airspace complexity evaluation are greatly reduced; the method can be flexibly applied to sectors with different airway structure configurations as a part of an air traffic control decision-making system.
Fig. 2 is a specific flowchart of the spatial domain operation complexity evaluation method based on the deep convolutional neural network according to the present invention.
As shown in fig. 2, in this embodiment, the method for extracting sector dynamic traffic data of a target airspace sector and performing airspace operation complexity level labeling includes: acquiring original air traffic operation data of a target airspace sector, and extracting sector dynamic traffic data of the target airspace sector from the original operation data in time intervals according to preset time granularity (1min time granularity), wherein the sector dynamic traffic data mainly comprises aircraft position information (longitude and latitude) and aircraft motion parameters (flight height, course and speed) in the sector; the sector dynamic traffic data are divided according to preset time periods, and the sector dynamic traffic data corresponding to each time period are subjected to airspace operation complexity level (airspace operation complexity level) marking, namely the extracted sector dynamic traffic data in each 1 minute time period are one sample, corresponding to a corresponding air traffic scene, each sample is subjected to airspace operation complexity level marking (label for short) by an experienced air traffic controller, the size of the corresponding airspace operation complexity is represented, 5 labels (1-5 levels) in different categories are totally provided, and the larger the number is, the higher the airspace operation complexity is represented.
In this embodiment, the method for defining a circumscribed rectangle of a target airspace sector and performing meshing processing includes: acquiring sector boundary point longitude and latitude data of a target airspace sector, determining a minimum external rectangle of the target airspace sector, wherein the speed of an aircraft is assumed to be 900km/h (15km/min), the time granularity of an air traffic situation scene is 1min, in order to ensure that the sector boundary position of the target airspace sector can reflect the information of the predicted flight path of the aircraft, each side of the minimum external rectangle is respectively extended outwards by a preset length (45km and 3min flight time distance) to form the external rectangle of the target airspace sector, gridding the external rectangle of the target airspace sector according to a preset length interval, and gridding the external rectangle of the target airspace sector by using 2km as an interval to obtain a plurality of square grids with the side length of 2km, and the square grids are used as a basis for subsequently generating an air traffic situation image.
FIG. 3(a) is a schematic diagram of a high historical track path in accordance with the present invention;
FIG. 3(b) is a schematic diagram of a speed history track path in accordance with the present invention;
FIG. 4(a) is a schematic diagram of a collision-free scenario in a collision-predicted track path according to the present invention;
fig. 4(b) is a schematic diagram of a collision scenario in a collision prediction track channel according to the present invention.
In this embodiment, the method for constructing the multi-channel air traffic situation image and constructing the air traffic situation image library according to the airspace operation complexity level label includes: according to the dynamic traffic data of the sectors at each time interval, taking the longitude and latitude of the aircraft as coordinates, acquiring the specific position of the corresponding coordinates in the external rectangle of the meshed target airspace sector, filling the altitude motion parameters of the aircraft in the dynamic traffic data of the sectors at each time interval into the corresponding grid as pixel values to generate an altitude historical track image channel (shown in figure 3 (a)) at the corresponding time interval, and filling the aircraft speed parameters into a new external rectangle of the meshed target airspace sector to generate a speed historical track image channel (shown in figure 3 (b)) in the same way; acquiring a predicted point (the predicted arrival position of the aircraft) of the aircraft after the preset time (3min) according to the longitude and latitude, the speed and the course of the last track point of each aircraft in a time period, connecting the sector dynamic traffic data between the last track point and the predicted point to acquire a predicted track, mapping the predicted track onto a new external rectangle of a gridded target airspace sector through the longitude and latitude, setting the filling value of the starting point of the predicted track of the aircraft in the external rectangle of the gridded target airspace sector to 10000, and sequentially decreasing the filling value by a preset step length (100) when the track passes through a new grid, the uncertainty size is used for reflecting the future aircraft position and the collision influence, and a collision prediction track image channel is generated until the filling of the last point of the prediction track is completed (as shown in fig. 4(a) and fig. 4 (b)); the method comprises the steps of forming a multi-channel air traffic situation image according to a height historical track image channel, a speed historical track image channel and a conflict prediction track image channel, wherein the size of the image is 173 × 3, associating the multi-channel air traffic situation image generated in different time periods with an airspace operation complexity level to obtain an air traffic situation image library, dividing the air traffic situation image library into a training data set and a test data set according to the proportion of 8:2, and using the training data set and the test data set for the training of an airspace operation complexity hierarchical network model and the test of the airspace operation complexity evaluation performance.
Fig. 5 is a schematic diagram of a deep convolutional neural network structure according to the present invention.
In this embodiment, the method for constructing the spatial domain operation complexity hierarchical network model according to the multi-channel air traffic situation image includes: constructing a spatial domain operation complexity hierarchical network model according to the multi-channel air traffic situation image through a deep convolutional neural network, namely
Constructing a fourteen-layer deep convolutional neural network model through a TensorFlow framework;
the first layer is an input layer and inputs multi-channel air traffic situation images; the second, third, fifth, sixth, eighth and ninth layers are convolution layers, and each convolution layer is subjected to nonlinear transformation operation by using a ReLU function; the fourth, seventh and tenth layers are pooling layers; the eleventh layer is a rolling layer and converts all the two-dimensional characteristic graphs into one-dimensional vectors; the twelfth, thirteenth and fourteenth layers are full-connection layers, and the output is a space-domain operation complexity level vector (5 levels);
the second and third convolutional layers comprise 32 convolution kernels, the fifth and sixth convolutional layers comprise 64 convolution kernels, the eighth and ninth convolutional layers comprise 128 convolution kernels, and convolution calculation is performed in an SAME filling mode according to the size of a preset convolution kernel (for example, the size of each convolution kernel is 3 x 3) and the movement amplitude of the preset convolution kernel (for example, the movement amplitude of the convolution kernel is 1);
performing pooling treatment on the pooling layers of the fourth, seventh and tenth layers in a maximum pooling mode, wherein the size of a pooling core is (2 x 2), the moving amplitude of the pooling core is 2, and padding is not filled;
the twelfth, thirteenth and fourteenth layers are all connected layers, the dimensions are (1 x 320), (1 x 160) and (1 x 5), wherein 5 represents the number of the spatial domain operation complexity levels, the output of the fourteenth layer carries out probability representation on the output of the fourteenth layer through a Softmax function, and the classification with the maximum probability is selected as the final classification result;
the Softmax function is:
wherein i is a airspace operation complexity grade category, and i is 1, which indicates that the airspace operation complexity is 1 grade; k represents a natural number greater than zero for counting; n is the total level number of the airspace operation complexity;
the output of the fourteenth layer is a 5-dimensional vector, and each dimension represents the probability of the airspace operation complexity belonging to the level;
nonlinear transformation is carried out on a nonlinear function ReLU function max (0, x) used after the second, third, fifth, sixth, eighth and ninth convolutional layers and the twelfth, thirteenth and fourteenth fully-connected layers; feature learning is automatically carried out from the provided virtual air multi-channel air traffic situation image data set by utilizing a deep convolutional neural network, the problem that manual features constructed by experts are relied on in the existing airspace operation complexity evaluation is solved, the evaluation of an end-to-end mode is realized, and the cost of constructing the manual features is eliminated; compared with the manually constructed shallow features, the deep features extracted by the deep neural network can be closer to the connotation understanding of the real airspace operation complexity of a controller, cross the barrier of the semantic gap, better represent the internal information of air traffic, and exceed the existing machine learning method based on the manual features in the airspace operation complexity evaluation performance;
in the present embodiment, specifically, as shown in fig. 5, [ 1 ] is an input layer, the input data is a generated multi-channel air traffic situation image, the size is (173 × 3), the width and height of the image are 173, and 3 indicates the number of channels of the image;
【2】 Representing a convolution layer, wherein the neuron on each receptive field in the layer is locally connected with the neuron in [ 1 ], and shares a weight; in the present embodiment, 32 convolution kernels are used for the [ 1 ] layer convolution, each convolution kernel has a size of (3 × 3), the convolution kernel shift amplitude is 2, and in the convolution, 0 is added to the [ 1 ] layer image, so that the image size after convolution is expressed by the formula:
Convl(x,y)＝Filterl*PreLayer(x,y)[filterarea],l＝1,2,...,32；
wherein, Convl(x, y) represents the convolved value of the neuron on the (x, y) th slice in the convolution layer, FilterlRepresenting convolution kernels, wherein neurons on the same slice l share one convolution kernel and are sequentially convolved with the neurons in the previous layer;
【3】 The layers are convolution layers same as the (2) layers, the convolution layer principle of the (5) layers and the (6) layers is the same as the (2) layers, but the layers use 64 convolution kernels, and the convolution layer principle of the (8) layers and the (9) layers is the same as the convolution layer principle
【2】 The layers are identical, but this layer uses 128 convolution kernels;
the convolution layer is connected with nonlinear operation to carry out nonlinear transformation on the numerical value of the convolution layer, thereby enhancing the universality of the neural network,
Outputl(x,y)＝nonlinear(Convl(x,y)+biasl) Wherein nonlinear represents a non-linear function, Convl(x, y) represents the value of the neuron at the previous layer (x, y), biaslRepresenting a bias value, which is shared by all neurons in the l-th receptive field of the upper layer; common activation functions include a sigmod activation function, a tanh activation function, an ELU activation function, a ReLU activation function, and the like; the activation function in this embodiment is a ReLU activation function, where the expression of the ReLU activation function is: (x) max (0, x), where x represents the output of each convolution layer; the ReLU activation function is used to enable the output of a part of neurons to be 0, so that the network becomes sparse, the interdependence relation of parameters is reduced, the occurrence of the overfitting problem is relieved, and meanwhile, the calculation amount is greatly reduced;
【4】 The layers, [ 7 ] and [ 10 ] are down-sampling layers, are used for reducing the dimension of the output vector of the previous layer, are used for preventing overfitting and simplifying calculation, the down-sampling operation here has modes of maximum pooling, average pooling and the like, in the embodiment, maximum pooling is adopted, the pooled kernel size is (2 × 2), and the pooled kernel moving amplitude is 2, so the image dimensionality after pooling becomes half of the original image dimensionality;
【12】 The layers [ 13 ] and [ 14 ] are fully connected layers with the dimensions of (1 × 320) and (14),
(1 x 160) and (1 x 5), wherein 5 represents the number of the spatial domain operation complexity grades, and the higher the grade is, the higher the operation complexity of the corresponding air traffic scene is; finally, probability representation is carried out on the [ 14 ] layer output by using a softmax function,
and finally outputting a 5-dimensional vector, wherein each dimension represents the probability that the spatial domain operation complexity belongs to a certain level.
FIG. 6 is a graph of evaluation accuracy performance index and loss function convergence according to the present invention.
In this embodiment, the method for training the complexity-hierarchical network model in the air domain includes: preprocessing an image in a training data set, and carrying out image standardization on pixel values of the image so that all pixel values of the image are between 0 and 1:
wherein mu is the mean value of the image; x is an image matrix; σ is the standard deviation; p is the number of pixels of the image;
putting the preprocessed training data set into the airspace operation complexity hierarchical network model for training; when the model is trained, a class equilibrium sampling method is adopted for model training, so that the trained samples are balanced among different classes every time, and the problem of unbalance of air traffic sample data is solved; meanwhile, in the process of sampling each time, data enhancement processing is carried out on the image in a rotating mode to generate a large number of new images, and the number of sufficient training samples is ensured; respectively collecting training samples from different types of data sets according to fixed data, and performing random data enhancement on training images after collection, wherein an image turning mode is adopted in the embodiment in consideration of actual problem conditions;
in the training process, the target loss function is cross entropy:
where y represents the reality of the image classProbability distribution;
representing the probability distribution calculated by the neural network; y isjAnd
each represents a probability value of the jth dimension in the 5-dimensional vector;
continuously optimizing a target loss function by a random optimization method in the training process; in this embodiment, an Adam optimizer is used, which is a random optimization method, and only requires a first-order gradient, and has a small memory footprint, and has a good effect on solving a model with a large number of parameters, and the Adam optimization method includes the following key steps:
t←t+ 1
mt←β1·mt-1+(l-β1)·gt
where α represents the algorithm learning rate, β1，β2M, v are parameters in the algorithm, f (p) represents a target loss function, and t represents the iteration number of the algorithm; training process and testedThe model evaluation performance iteration curve of the routine is shown in FIG. 6; the method for enhancing the image data by adopting category balanced sampling and random rotation in the model training stage avoids the tendency that the model is prone to selecting the category containing most samples, relieves the influence of the imbalance problem to a certain extent, increases the richness of a training data set, and further improves the accuracy of the network to airspace operation complexity evaluation.
In this embodiment, the method for performing spatial domain operation complexity evaluation according to the trained spatial domain operation complexity hierarchical network model includes: and preprocessing the images in the test data set, inputting the images after the preprocessing in the test data set into the trained airspace operation complexity grading network model to obtain an airspace operation complexity grading result, and finishing the airspace operation complexity evaluation.
In conclusion, the sector dynamic traffic data of the target airspace sector are extracted, and the airspace operation complexity level is labeled; defining a circumscribed rectangle of a target airspace sector, and carrying out gridding treatment; constructing a multi-channel air traffic situation image, and constructing an air traffic situation image library according to the operation complexity grade of an airspace; constructing a spatial domain operation complexity hierarchical network model according to the multi-channel air traffic situation image; training a space domain operation complexity hierarchical network model; and the airspace operation complexity evaluation is carried out according to the trained airspace operation complexity hierarchical network model, so that the selection of the most relevant characteristics can be directly learned from the original data in an end-to-end mode without depending on the complexity relevant characteristics, the establishment of the airspace operation complexity evaluation model is assisted, and the workload and the use threshold of the airspace complexity evaluation are greatly reduced.
In the embodiments provided herein, the flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
In addition, the functional modules in the embodiments of the present invention may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.
The functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
In light of the foregoing description of the preferred embodiment of the present invention, many modifications and variations will be apparent to those skilled in the art without departing from the spirit and scope of the invention. The technical scope of the present invention is not limited to the content of the specification, and must be determined according to the scope of the claims.
Claims (7)
1. A spatial domain operation complexity evaluation method based on a deep convolutional neural network is characterized by comprising the following steps:
extracting sector dynamic traffic data of a target airspace sector, and marking airspace operation complexity grades;
defining a circumscribed rectangle of a target airspace sector, and carrying out gridding treatment;
constructing a multi-channel air traffic situation image, and constructing an air traffic situation image library according to the operation complexity grade of an airspace;
constructing a spatial domain operation complexity hierarchical network model according to the multi-channel air traffic situation image;
training a space domain operation complexity hierarchical network model; and
and performing space domain operation complexity evaluation according to the trained space domain operation complexity hierarchical network model.
2. The spatial domain operational complexity evaluation method of claim 1,
the method for extracting the sector dynamic traffic data of the target airspace sector and marking the airspace operation complexity level comprises the following steps:
acquiring original air traffic operation data of a target airspace sector, and extracting sector dynamic traffic data of the target airspace sector from the original operation data according to a preset time granularity period;
and dividing the dynamic traffic data of the sector according to a preset time period, and marking the complexity grade of the airspace operation on the dynamic traffic data of the sector corresponding to each time period.
3. The spatial domain operational complexity evaluation method of claim 2,
the method for demarcating the external rectangle of the target airspace sector and carrying out gridding processing comprises the following steps:
the method comprises the steps of obtaining sector boundary point longitude and latitude data of a target airspace sector, determining a minimum external rectangle of the target airspace sector, expanding each side of the minimum external rectangle outwards by a preset length respectively to form the target airspace sector external rectangle, and carrying out gridding processing on the target airspace sector external rectangle according to preset length intervals.
4. The spatial domain operational complexity evaluation method of claim 3,
the method for constructing the multi-channel air traffic situation image and constructing the air traffic situation image library according to the airspace operation complexity level marking comprises the following steps:
according to the dynamic traffic data of the sectors at each time period, taking the longitude and latitude of the aircraft as coordinates, acquiring the position of the corresponding coordinates in the external rectangle of the meshed target airspace sector, filling the altitude motion parameters of the aircraft in the dynamic traffic data of the sectors at each time period into the corresponding grids as pixel values to generate an altitude historical track image channel at the corresponding time period, and filling the speed parameters of the aircraft into the external rectangle of the meshed target airspace sector to generate a speed historical track image channel;
acquiring a predicted point which is reached by the aircraft after preset time according to the longitude and latitude, the speed and the course of the last track point of each aircraft in a time period, connecting sector dynamic traffic data between the last track point and the predicted point to acquire a predicted track, mapping the predicted track to a meshed target airspace sector external rectangle through the longitude and latitude, and when the track passes through a new mesh, sequentially decreasing the filling value by a preset step length until the filling of the last point of the predicted track is completed to generate a conflict predicted track image channel;
and forming a multi-channel air traffic situation image according to the height historical track image channel, the speed historical track image channel and the conflict prediction track image channel, associating the multi-channel air traffic situation image generated at different time intervals with the airspace operation complexity level to obtain an air traffic situation image library, and dividing the air traffic situation image library into a training data set and a testing data set.
5. The spatial domain operational complexity evaluation method of claim 4,
the method for constructing the airspace operation complexity hierarchical network model according to the multi-channel air traffic situation image comprises the following steps:
constructing a spatial domain operation complexity hierarchical network model according to the multi-channel air traffic situation image through a deep convolutional neural network, namely
Constructing a fourteen-layer deep convolutional neural network model;
the first layer is an input layer and inputs multi-channel air traffic situation images; the second, third, fifth, sixth, eighth and ninth layers are convolution layers; the fourth, seventh and tenth layers are pooling layers; the eleventh layer is a rolled layer; the twelfth, thirteenth and fourteenth layers are full connection layers, and the output is a space domain operation complexity level vector;
the second layer and the third layer of convolution layers comprise 32 convolution kernels, the fifth layer and the sixth layer of convolution layers comprise 64 convolution kernels, the eighth layer and the ninth layer of convolution layers comprise 128 convolution kernels, and convolution calculation is carried out in an SAME filling mode according to the size of a preset convolution kernel and the movement amplitude of the preset convolution kernel;
the pooling layers of the fourth, seventh and tenth layers are subjected to pooling treatment by adopting a maximum pooling mode;
the twelfth, thirteenth and fourteenth layers are full connection layers, probability representation is carried out on the fourteenth layer output through a Softmax function, and the classification with the maximum probability value is selected as a final classification result;
the Softmax function is:
wherein i is a airspace operation complexity grade category, and i is 1, which means that the airspace operation complexity is 1 grade; k represents a natural number greater than zero; n is the total level number of the airspace operation complexity;
the output of the fourteenth layer is a 5-dimensional vector, and each dimension represents the probability of the airspace operation complexity belonging to the level;
and the nonlinear function used after the second, third, fifth, sixth, eighth and ninth convolution layers and the twelfth, thirteenth and fourteenth full connection layers is subjected to nonlinear transformation.
6. The spatial domain operational complexity evaluation method of claim 5,
the method for training the space-domain operation complexity hierarchical network model comprises the following steps:
preprocessing an image in the training data set, and performing image standardization processing on pixel values of the image:
wherein mu is the mean value of the image; x is an image matrix; σ is the standard deviation; p is the number of pixels of the image;
putting the preprocessed training data set into the airspace operation complexity hierarchical network model for training;
in the training process, the target loss function is cross entropy:
where y represents the true probability distribution of the image class;
representing the probability distribution calculated by the neural network; y isjAnd
each represents a probability value of the jth dimension in the 5-dimensional vector;
and continuously optimizing the target loss function by a random optimization method in the training process.
7. The spatial domain operational complexity evaluation method of claim 6,
the method for evaluating the complexity of the airspace operation according to the trained airspace operation complexity hierarchical network model comprises the following steps:
and preprocessing the images in the test data set, inputting the images after the preprocessing in the test data set into the trained airspace operation complexity grading network model to obtain an airspace operation complexity grading result, and finishing the airspace operation complexity evaluation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011294477.7A CN112489497B (en) | 2020-11-18 | 2020-11-18 | Airspace operation complexity evaluation method based on deep convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011294477.7A CN112489497B (en) | 2020-11-18 | 2020-11-18 | Airspace operation complexity evaluation method based on deep convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112489497A true CN112489497A (en) | 2021-03-12 |
| CN112489497B CN112489497B (en) | 2022-03-11 |
Family
ID=74931680
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011294477.7A Active CN112489497B (en) | 2020-11-18 | 2020-11-18 | Airspace operation complexity evaluation method based on deep convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112489497B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112989287A (en) * | 2021-04-27 | 2021-06-18 | 北京航空航天大学 | Traffic situation real-time calculation method based on streaming big data |
| CN113139576A (en) * | 2021-03-22 | 2021-07-20 | 广东省科学院智能制造研究所 | Deep learning image classification method and system combining image complexity |
| CN115223402A (en) * | 2022-06-29 | 2022-10-21 | 北京航空航天大学 | Space domain sector complexity prediction method based on space-time graph convolutional network |
| CN115527397A (en) * | 2022-09-30 | 2022-12-27 | 中国民用航空飞行学院 | Air traffic control situation feature extraction method and device based on multimode neural network |
| CN116312072A (en) * | 2023-03-21 | 2023-06-23 | 中国人民解放军93209部队 | Flight path operation conflict decoupling control method based on airspace grids |
| CN116935700A (en) * | 2023-09-18 | 2023-10-24 | 四川大学 | Sector traffic situation prediction method based on multi-source features |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103106813A (en) * | 2013-01-18 | 2013-05-15 | 南京航空航天大学 | Traffic situation complexity assessment method based on index system |
| US20140188378A1 (en) * | 2011-01-25 | 2014-07-03 | Bruce K. Sawhill | Method and apparatus for dynamic aircraft trajectory management |
| CN106357461A (en) * | 2016-11-04 | 2017-01-25 | 中国民航大学 | Measuring method for air traffic display complexity |
| US20170301247A1 (en) * | 2016-04-19 | 2017-10-19 | George Mason University | Method And Apparatus For Probabilistic Alerting Of Aircraft Unstabilized Approaches Using Big Data |
| CN108090613A (en) * | 2017-12-18 | 2018-05-29 | 南京航空航天大学 | A kind of approach control sector ATC controller workload Forecasting Methodology |
| CN109886352A (en) * | 2019-03-04 | 2019-06-14 | 北京航空航天大学 | A kind of unsupervised appraisal procedure of airspace complexity |
| CN109993225A (en) * | 2019-03-29 | 2019-07-09 | 北京航空航天大学 | A kind of airspace complexity classification method and device based on unsupervised learning |
| CN111009155A (en) * | 2019-12-06 | 2020-04-14 | 南京莱斯信息技术股份有限公司 | Air traffic flow complexity quantitative analysis method based on airspace structure and flight flow |
| CN111047182A (en) * | 2019-12-10 | 2020-04-21 | 北京航空航天大学 | Airspace complexity evaluation method based on deep unsupervised learning |
| CN111461482A (en) * | 2020-02-25 | 2020-07-28 | 北京航空航天大学 | Airspace dynamic management method and device |
-
2020
- 2020-11-18 CN CN202011294477.7A patent/CN112489497B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140188378A1 (en) * | 2011-01-25 | 2014-07-03 | Bruce K. Sawhill | Method and apparatus for dynamic aircraft trajectory management |
| CN103106813A (en) * | 2013-01-18 | 2013-05-15 | 南京航空航天大学 | Traffic situation complexity assessment method based on index system |
| US20170301247A1 (en) * | 2016-04-19 | 2017-10-19 | George Mason University | Method And Apparatus For Probabilistic Alerting Of Aircraft Unstabilized Approaches Using Big Data |
| CN106357461A (en) * | 2016-11-04 | 2017-01-25 | 中国民航大学 | Measuring method for air traffic display complexity |
| CN108090613A (en) * | 2017-12-18 | 2018-05-29 | 南京航空航天大学 | A kind of approach control sector ATC controller workload Forecasting Methodology |
| CN109886352A (en) * | 2019-03-04 | 2019-06-14 | 北京航空航天大学 | A kind of unsupervised appraisal procedure of airspace complexity |
| CN109993225A (en) * | 2019-03-29 | 2019-07-09 | 北京航空航天大学 | A kind of airspace complexity classification method and device based on unsupervised learning |
| CN111009155A (en) * | 2019-12-06 | 2020-04-14 | 南京莱斯信息技术股份有限公司 | Air traffic flow complexity quantitative analysis method based on airspace structure and flight flow |
| CN111047182A (en) * | 2019-12-10 | 2020-04-21 | 北京航空航天大学 | Airspace complexity evaluation method based on deep unsupervised learning |
| CN111461482A (en) * | 2020-02-25 | 2020-07-28 | 北京航空航天大学 | Airspace dynamic management method and device |
Non-Patent Citations (2)
| Title |
|---|
| XI ZHU ET.: "A SEMI-SUPERVISED LEARNING METHOD FOR AIR TRAFFIC", 《2017 INTEGRATED COMMUNICATIONS, NAVIGATION AND SURVEILLANCE CONFERENCE (ICNS)》 * |
| 丛玮，等: "基于指标体系的扇区复杂性评估方法", 《交通运输系统工程与信息》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113139576A (en) * | 2021-03-22 | 2021-07-20 | 广东省科学院智能制造研究所 | Deep learning image classification method and system combining image complexity |
| CN113139576B (en) * | 2021-03-22 | 2024-03-12 | 广东省科学院智能制造研究所 | Deep learning image classification method and system combining image complexity |
| CN112989287A (en) * | 2021-04-27 | 2021-06-18 | 北京航空航天大学 | Traffic situation real-time calculation method based on streaming big data |
| CN115223402A (en) * | 2022-06-29 | 2022-10-21 | 北京航空航天大学 | Space domain sector complexity prediction method based on space-time graph convolutional network |
| CN115223402B (en) * | 2022-06-29 | 2023-05-26 | 北京航空航天大学 | Airspace sector complexity prediction method based on space-time diagram convolutional network |
| CN115527397A (en) * | 2022-09-30 | 2022-12-27 | 中国民用航空飞行学院 | Air traffic control situation feature extraction method and device based on multimode neural network |
| CN116312072A (en) * | 2023-03-21 | 2023-06-23 | 中国人民解放军93209部队 | Flight path operation conflict decoupling control method based on airspace grids |
| CN116312072B (en) * | 2023-03-21 | 2024-01-26 | 中国人民解放军93209部队 | Flight path operation conflict decoupling control method based on airspace grids |
| CN116935700A (en) * | 2023-09-18 | 2023-10-24 | 四川大学 | Sector traffic situation prediction method based on multi-source features |
| CN116935700B (en) * | 2023-09-18 | 2023-12-05 | 四川大学 | Sector traffic situation prediction method based on multi-source features |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112489497B (en) | 2022-03-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112489497B (en) | Airspace operation complexity evaluation method based on deep convolutional neural network | |
| CN112465199B (en) | Airspace situation assessment system | |
| CN107909206B (en) | PM2.5 prediction method based on deep structure recurrent neural network | |
| CN109508360B (en) | Geographical multivariate stream data space-time autocorrelation analysis method based on cellular automaton | |
| CN112766549A (en) | Air pollutant concentration forecasting method and device and storage medium | |
| CN109145983A (en) | A kind of real-time scene image, semantic dividing method based on lightweight network | |
| CN114092832B (en) | High-resolution remote sensing image classification method based on parallel hybrid convolutional network | |
| CN113808396B (en) | Traffic speed prediction method and system based on traffic flow data fusion | |
| CN111310965A (en) | Aircraft track prediction method based on LSTM network | |
| CN111695731A (en) | Load prediction method, system and equipment based on multi-source data and hybrid neural network | |
| CN112990222B (en) | Image boundary knowledge migration-based guided semantic segmentation method | |
| CN112180471B (en) | Weather forecasting method, device, equipment and storage medium | |
| CN112560967A (en) | Multi-source remote sensing image classification method, storage medium and computing device | |
| US20230222768A1 (en) | Multiscale point cloud classification method and system | |
| CN115271181A (en) | Typhoon probability forecasting intelligent method and device based on multi-mode data fusion | |
| CN115862324A (en) | Space-time synchronization graph convolution neural network for intelligent traffic and traffic prediction method | |
| WO2020013726A1 (en) | Method for interpreting artificial neural networks | |
| CN113128769A (en) | Intelligent flight delay prediction method based on deep learning | |
| CN116151479B (en) | Flight delay prediction method and prediction system | |
| CN115907079B (en) | Airspace traffic flow prediction method based on attention space-time diagram convolutional network | |
| CN115063972A (en) | Traffic speed prediction method and system based on graph convolution and gate control cyclic unit | |
| CN109726690B (en) | Multi-region description method for learner behavior image based on DenseCap network | |
| CN113591608A (en) | High-resolution remote sensing image impervious surface extraction method based on deep learning | |
| Toubeh et al. | Risk-aware planning by confidence estimation using deep learning-based perception | |
| Dupree et al. | Time forecasting satellite light curve patterns using Neural Networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |