Terminal Area Traffic Prediction Method and System Based on Multivariate Time Series Analysis

Abstract

This invention relates to the field of traffic control, specifically to a terminal area traffic flow prediction method and system based on multivariate time series analysis. The method includes: acquiring basic data within the terminal area; preprocessing the basic data to construct a multivariate time series database for terminal area traffic flow prediction; segmenting and symbolically representing the data in the multivariate time series database based on trend and mean time series representation methods; using a random forest model to perform feature filtering and dimensionality reduction on the segmented and symbolically represented multivariate time series data to construct a feature matrix; constructing a terminal area traffic flow prediction model based on CNN-GRUA, inputting the feature matrix into the trained terminal area traffic flow prediction model, and outputting the terminal area traffic flow prediction result. This method helps air traffic controllers understand the expected traffic situation, provides important decision support information for traffic allocation, improves flight arrival and departure efficiency, alleviates traffic pressure in airport terminal areas, and thus improves the operational efficiency of the entire air traffic system.

Description

Technical Field

[0001] This invention relates to the field of civil aviation traffic control, specifically to a terminal area traffic prediction method and system based on multivariate time series analysis. Background Technology

[0002] Since the 20th century, my country's GDP has grown rapidly every year. This economic development has spurred the rapid growth of my country's civil aviation industry, leading to the construction of more and more airports across the country to meet the demands of the times. While the rapid development of civil aviation has brought prosperity and convenience to cities, it has also brought a series of problems. Faced with the surge in air traffic, limited airspace resources are insufficient to meet flight demands. Coupled with increased air defense training missions and frequent severe weather, the contradiction between airspace resources and flight demand has become increasingly prominent, ultimately leading to frequent problems such as airspace congestion, flight delays, and even flight conflicts. Therefore, air traffic flow management is an important means of solving these problems.

[0003] Airport terminal areas, typically located near airports, are airspace units where arriving and departing aircraft converge. They are characterized by high flight density, complex route structures, numerous restrictions, and significant conflicts, making them critical nodes in the air traffic network and areas with the most severe airspace congestion, flight delays, and flight conflicts. In the future, with the further increase in air traffic volume, airport terminal areas will only bear increasingly larger flight flows, pushing existing airspace towards saturation and leading to even more severe traffic congestion and flight delays. Because airport terminal areas are crucial nodes in the entire air traffic system, they are bound to become a major bottleneck restricting further improvements in the quality of current and future air transport services. Therefore, real-time, accurate, and efficient short-term traffic forecasting for airport terminal areas is imperative. Summary of the Invention

[0004] The purpose of this invention is to provide a terminal area traffic prediction method and system based on multivariate time series analysis.

[0005] To address the aforementioned technical problems, this invention provides a terminal area traffic prediction method based on multivariate time-series analysis, comprising:

[0006] Step 1: Obtain basic data within the terminal area;

[0007] Step 2: Preprocess the basic data to construct a multivariate time-series database for terminal area traffic prediction;

[0008] Step 3: Based on trend and mean time series representation methods, segment and symbolize the data in the multivariate time series database;

[0009] Step 4: Use the random forest model to identify factors that are highly correlated with terminal area traffic prediction, perform feature filtering and dimensionality reduction on the segmented and symbolically represented multivariate time series data, and construct a feature matrix.

[0010] Step 5: Construct a terminal area traffic prediction model based on CNN-GRUA, input the feature matrix into the trained terminal area traffic prediction model, and output the terminal area traffic prediction result.

[0011] Furthermore, the basic data obtained in step 1 within the terminal area includes: actual flow in the terminal area, planned flow in the terminal area, number of regular flights, number of canceled flights, capacity difference, flight delays, average delay time, number of delayed flights, intensity of flow control strategy issuance, average flow control interval, meteorological characteristics in METAR messages, WAF convective weather, and terminal area weather index.

[0012] Furthermore, step 2 involves preprocessing the basic data to construct a multivariate time-series database for terminal area traffic prediction. This includes: cleaning the collected terminal area traffic data by filling in missing values, smoothing noisy data, and identifying and deleting outliers; constructing multivariate time-series data containing traffic demand characteristics, delay characteristics, strategy status characteristics, and environmental status characteristics from the cleaned basic data to form a multivariate time-series database; wherein the traffic demand characteristics reflect the traffic demand status, including actual terminal area traffic, planned terminal area traffic, on-time flight count, canceled flight count, and capacity difference; the delay characteristics reflect the delay status, including flight delays, average delay time, and number of delayed flights; the strategy status characteristics reflect the terminal area strategy status, including the intensity of flow control strategy issuance and average flow control interval; and the environmental status characteristics reflect the terminal area environmental status, including meteorological characteristics in the METAR report, WAF convective weather, and terminal area weather index.

[0013] Furthermore, the trend and mean-based time series representation method in step 3 segments and symbolically represents the data in the multivariate time series database, including: obtaining all trend segment points on the normalized time series of the multivariate time series data based on the binary iterative algorithm; dividing the entire time series into several subsequences of fixed length using a non-overlapping sliding window mechanism according to the preset subsequence length l and the number of trend segments ω, and adopting corresponding strategies to ensure that the number of trend segment points in each subsequence is (ω-1), forming ω trend segments; and symbolically representing each trend segment using tuples based on trend features and mean statistics.

[0014] Furthermore, the method of employing a corresponding strategy to ensure that the number of trend segmentation points in each subsequence is (ω-1), forming ω trend segments, includes: finding the number γ of trend segmentation points in each subsequence; if γ ≥ ω-1, then selecting the (ω-1) trend segmentation points with the most trend information to form ω trend segments; if... Then, the trend segment with the longest sequence (ω-1-γ) is bisected to form ω trend segments; if Then, the average segmentation method is directly used to form ω trend segments; where Indicates to Round down to the nearest integer.

[0015] Furthermore, the symbolic representation of each trend segment using a pair based on trend characteristics and mean statistics includes: dividing all trend segment points of the entire time series into 3 to 10 equally probable regions, each represented by a different lowercase letter, according to a Gaussian distribution; dividing the entire slope value range into 5 non-overlapping regions, each represented by a different uppercase letter; calculating the slope value of each trend segment based on the trend segment points or endpoints at both ends of each trend segment, and using the uppercase letter corresponding to the region where the slope value is located to represent the trend characteristic information of that trend segment; calculating the average value of each trend segment, and using the lowercase letter corresponding to the equally probable region where the average value is located to represent the mean statistics information of that trend segment; using a pair where uppercase letters represent trend characteristics and lowercase letters represent mean statistics to represent each trend segment; the formula for calculating the average value of the trend segment is expressed as: Where x = [t] i ,…,t i+m-1 ] is a trend segmented sequence of length m.

[0016] Furthermore, in step 4, a random forest model is used to identify factors highly correlated with terminal area traffic prediction. Feature selection and dimensionality reduction are performed on the segmented and symbolically represented multivariate time-series data to construct a feature matrix. This includes: based on the segmented and symbolically represented multivariate time-series data, using mean squared error as the evaluation criterion for feature partitioning in decision tree nodes, n decision trees are trained to jointly construct a random forest model; the constructed random forest model is used to evaluate the importance of all features in the segmented and symbolically represented multivariate time-series data, and the top 8 features with the highest importance scores are selected to form the feature matrix.

[0017] Furthermore, in step 5, a terminal area traffic prediction model based on CNN-GRUA is constructed, and the feature matrix is ​​input into the trained terminal area traffic prediction model to output the terminal area traffic prediction result. This includes: constructing and training a CNN sub-model by setting the number of layers in the CNN network, the size of the convolutional kernels in each convolutional layer, and the pooling method of the pooling layers; constructing and training a GRUA sub-model by setting the number of layers and the number of hidden layer neurons in the GRU network, and introducing an attention mechanism to assign corresponding weights to the output results of each GRU network; fusing the CNN sub-model and the GRUA sub-model using a fully connected layer to construct and train a CNN-GRUA-based terminal area traffic prediction model; and inputting the feature matrix into the trained CNN sub-model and the GRUA sub-model respectively, and obtaining the final terminal area traffic prediction result by fusing their output spatial feature prediction results and temporal feature prediction results.

[0018] In another aspect, the present invention also provides a terminal area traffic prediction system based on multivariate time series analysis, comprising: a data acquisition module for acquiring basic data within the terminal area; a data preprocessing module adapted to preprocess the data acquired by the data acquisition module to construct a multivariate time series database; a terminal area dynamic multivariate time series data representation module adapted to segment and symbolically represent the data in the multivariate time series database; a terminal area traffic flow feature filtering module for performing feature filtering and dimensionality reduction on the segmented and symbolically represented multivariate time series data to construct a feature matrix; and a terminal area traffic prediction module for performing terminal area traffic prediction on the feature matrix and outputting the terminal area traffic prediction result.

[0019] The beneficial effects of this invention are that it first uses a time-series representation method to simplify the representation of multivariate time-series data, uses a random forest model for importance assessment to filter and reduce the dimensionality of features in the multivariate time-series data, and utilizes a terminal area traffic prediction model based on CNN-GRUA to predict terminal area traffic. This helps air traffic controllers understand the expected traffic situation, provides important decision support information for traffic dispatch, improves the efficiency of flight arrivals and departures, alleviates traffic pressure in airport terminal areas, and thus improves the operational efficiency of the entire air traffic system.

[0020] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a flowchart of the terminal area traffic prediction method based on multivariate time series analysis of the present invention;

[0024] Figure 2 This is a schematic diagram of the present invention for obtaining trend segmentation points based on the binary iterative algorithm;

[0025] Figure 3 This is a schematic diagram illustrating the time series representation of a subsequence based on trend and mean, as exemplified in an embodiment of the present invention;

[0026] Figure 4 This is a schematic diagram of the random forest model of the present invention;

[0027] Figure 5 This is a schematic diagram of the terminal area traffic prediction model based on CNN-GRUA of the present invention;

[0028] Figure 6 This is a system block diagram of the terminal area traffic prediction system based on multivariate time series analysis of the present invention. Detailed Implementation

[0029] To make the objectives, 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. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. 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.

[0030] Example 1

[0031] like Figure 1 As shown, this embodiment provides a terminal area traffic prediction method based on multivariate time series analysis, including:

[0032] Step 1: Obtain basic data within the terminal area;

[0033] Step 2: Preprocess the basic data to construct a multivariate time-series database for terminal area traffic prediction;

[0034] Step 3: Based on trend and mean time series representation methods, segment and symbolize the data in the multivariate time series database;

[0035] Step 4: Use the random forest model to identify factors that are highly correlated with terminal area traffic prediction, perform feature filtering and dimensionality reduction on the segmented and symbolically represented multivariate time series data, and construct a feature matrix.

[0036] Step 5: Construct a terminal area traffic prediction model based on CNN-GRUA, input the feature matrix into the trained terminal area traffic prediction model, and output the terminal area traffic prediction result.

[0037] In some embodiments, the basic data obtained in step 1 within the terminal area includes: actual flow in the terminal area, planned flow in the terminal area, number of regular flights, number of canceled flights, capacity difference, flight delays, average delay time, number of delayed flights, intensity of flow control strategy issuance, average flow control interval, meteorological characteristics in METAR messages, WAF convective weather, and terminal area weather index.

[0038] It is understood that step 2 involves preprocessing the basic data to construct a multivariate time-series database for terminal area traffic prediction. This includes: cleaning the collected basic data on terminal area traffic, i.e., filling in missing values, smoothing noisy data, and identifying and deleting outliers; constructing multivariate time-series data containing traffic demand characteristics, delay characteristics, strategy status characteristics, and environmental status characteristics from the cleaned basic data to form a multivariate time-series database; wherein the traffic demand characteristics are used to reflect the traffic demand status, including actual terminal area traffic, planned terminal area traffic, on-time flight count, canceled flight count, and capacity difference; the delay characteristics are used to reflect the delay status, including flight delays, average delay time, and number of delayed flights; the strategy status characteristics are used to reflect the strategy status of the terminal area, including the intensity of flow control strategy issuance and average flow control interval; and the environmental status characteristics are used to reflect the environmental status of the terminal area, including meteorological characteristics in the METAR report, WAF convective weather, and terminal area weather index.

[0039] In this embodiment, step 3, the trend and mean-based time series representation method, segments and symbolically represents the data in the multivariate time series database. This includes: obtaining all trend segmentation points on the normalized time series of the multivariate time series data based on a binary iterative algorithm; dividing the entire time series into several subsequences of fixed length using a non-overlapping sliding window mechanism according to a preset subsequence length l and the number of trend segments ω, and employing a corresponding strategy to ensure that the number of trend segmentation points in each subsequence is (ω-1), forming ω trend segments; and symbolically representing each trend segment using tuples based on trend features and mean statistics.

[0040] The calculation formula for normalizing multivariate time series data in this embodiment is expressed as follows: Where x represents the multivariate time series data to be normalized, y represents the multivariate time series data after normalization, and min and max represent the minimum and maximum values ​​in the multivariate time series data to be normalized, respectively. Normalization of multivariate time series data is performed to rescale the multivariate time series data within the original range so that the values ​​are within the range of 0 to 1, which facilitates subsequent operations and machine learning.

[0041] The binary iterative algorithm described in this embodiment is specifically as follows: Figure 2 As shown, the dashed lines at the top represent the normalized sequences S, S1, and S2. n These are the start and end points of the sequence S, respectively, and the dashed line below is the connecting line L. 1-n Calculate the connection line L from each point in the sequence S. 1-n Find the distance and locate the point S with the maximum distance. i If the maximum distance is greater than the preset threshold ε, then this point S is... i Mark the trend segmentation points and divide the sequence S into two parts with the starting point S1 and the ending point S2. i and the starting point is S i The endpoint is S n The two sequence segments are then subjected to a binary iterative recursion, and so on, until all trend segmentation points are found. If the maximum distance is not greater than the preset threshold ε, the binary iteration ends.

[0042] In a specific implementation, the method of employing a corresponding strategy to ensure that the number of trend segmentation points in each subsequence is (ω-1) to form ω trend segments includes: finding the number γ of trend segmentation points in each subsequence; if γ ≥ ω-1, then selecting the (ω-1) trend segmentation points with the most trend information to form ω trend segments; if Then, the trend segment with the longest sequence (ω-1-γ) is bisected to form ω trend segments; if Then, the average segmentation method is directly used to form ω trend segments; where Indicates to Round down. Trend segments are formed by connecting the start and end points of each subsequence with (ω-1) trend segment points in sequence.

[0043] In a specific implementation, the symbolic representation of each trend segment using a binary tuple based on trend characteristics and mean statistics includes: dividing all trend segment points of the entire time series into 3 to 10 equally probable regions, each represented by a different lowercase letter, according to a Gaussian distribution; dividing the entire slope value range into 5 non-overlapping regions, each represented by a different uppercase letter; calculating the slope value of each trend segment based on the trend segment points or endpoints at both ends of each trend segment, and using the uppercase letter corresponding to the region where the slope value is located to represent the trend characteristic information of the trend segment; calculating the average value of each trend segment, and using the lowercase letter corresponding to the equally probable region where the average value is located to represent the mean statistics information of the trend segment; representing each trend segment using a binary tuple where uppercase letters represent trend characteristics and lowercase letters represent mean statistics; the formula for calculating the average value of the trend segment is expressed as: Where x = [t] i ,…,t i+m-1 ] is a trend segmented sequence of length m.

[0044] In some embodiments, the slope value of the trend segment ranges from (-∞, +∞), and this range is divided into 5 non-overlapping regions. Each region corresponds to an uppercase letter to describe the degree of trend change in this segment: (-∞, -1) corresponds to A, indicating a sharp decline; (-1, -0.1) corresponds to B, indicating a slight decline; (-0.1, 0.1) corresponds to C, indicating slight fluctuations (approximately horizontal); (0.1, 1) corresponds to D, indicating a slight increase; and (1, +∞) corresponds to E, indicating a sharp increase. Assuming all trend segment points in the entire time series are divided into 4 equally probable regions according to a Gaussian distribution, each region corresponds to a lowercase letter to describe the mean statistical information of this segment: (-∞, -1) corresponds to a; (-1, 0) corresponds to b; (0, 1) corresponds to c; and (1, +∞) corresponds to d. For example, ... Figure 3 The subsequence shown can be represented by symbolic pairs of data pairs that represent the trend characteristics and mean statistics of each trend segment. That is, (1.31,0.15) is represented as (E,c), (-0.95,0.02) is represented as (B,c), (1.15,-0.15) is represented as (E,b), (-1.16,0.11) is represented as (A,c), (0.36,0.01) is represented as (D,c), and (-0.18,0.12) is represented as (B,c). Then, the subsequence containing 6 trend segments can be represented as EcBcEbAcDcBc.

[0045] In this embodiment, step 4 utilizes a random forest model to identify factors highly correlated with terminal area traffic prediction, performs feature screening and dimensionality reduction on segmented and symbolically represented multivariate time-series data, and constructs a feature matrix. This includes: based on segmented and symbolically represented multivariate time-series data, using mean squared error as the evaluation criterion for feature partitioning in decision tree nodes, training n decision trees to jointly construct a random forest model; using the constructed random forest model to evaluate the importance of all features in the segmented and symbolically represented multivariate time-series data, and selecting the top 8 features by importance score to form the feature matrix.

[0046] Specifically, such as Figure 4 As shown, the random forest model is composed of multiple decision trees, and the final decision result is obtained by combining the decision results of each classification and regression decision tree. The main steps are as follows: Using the bootstrap method with replacement, select several data samples from the segmented and symbolically represented multivariate time-series data samples; using the sampled data samples and mean squared error as the evaluation criterion for feature splitting when splitting nodes in the decision tree, determine the optimal splitting method and generate decision trees; repeat the above operation n times until n decision trees are obtained, thus constructing the random forest model; use the constructed random forest model to evaluate the importance of all features in the multivariate time-series data, and select the top 8 features with the highest importance scores as the input feature matrix of the prediction model.

[0047] It is understandable that multivariate time series data includes four major categories of features: traffic demand features, delay features, strategy state features, and environmental state features. Each category contains many features. Extensive experiments have shown that the accuracy of terminal area traffic prediction using a feature matrix composed of more than 8 features is similar to that using a feature matrix composed of 8 features. Therefore, to simplify the calculation process, the top 8 features in the importance assessment of the random forest model are selected to form the feature matrix.

[0048] In this embodiment, step 5 involves constructing a CNN-GRUA-based terminal area traffic prediction model and inputting the feature matrix into the trained terminal area traffic prediction model to output the terminal area traffic prediction result. This includes: constructing and training a CNN sub-model by setting the number of layers in the CNN network, the kernel size in each convolutional layer, and the pooling method of the pooling layers; constructing and training a GRUA sub-model by setting the number of layers and the number of hidden layer neurons in the GRU network, and introducing an attention mechanism to assign corresponding weights to the output results of each GRU network; fusing the CNN sub-model and the GRUA sub-model using fully connected layers to construct and train a CNN-GRUA-based terminal area traffic prediction model; and inputting the feature matrix into the trained CNN sub-model and the GRUA sub-model respectively, and obtaining the final terminal area traffic prediction result by fusing their output spatial feature prediction results and temporal feature prediction results.

[0049] In this embodiment, such as Figure 5 As shown, the terminal area traffic prediction model based on CNN-GRUA is suitable for first using a time series representation method to reduce the dimensionality and represent the features of the original data (i.e., multivariate time series data, which is divided into a training set for training the model and a test set for prediction) to facilitate subsequent processing and computation. Then, a random forest model is constructed for importance evaluation, and features with higher importance are selected to form a feature matrix, which is then input into two different sub-models, CNN and GRUA, for prediction. The CNN sub-model requires setting the number of layers in the CNN network and the operations performed in each layer, as well as setting the kernel size and pooling mode in each convolutional layer. The GRUA sub-model also requires setting the number of layers in the GRU network and the number of hidden layer neurons. The CNN sub-model and GRUA sub-model are trained using the training set, and the training results of the CNN sub-model and GRUA sub-model are trained through a fully connected layer with input of 2 and output of 1. The parameters are iteratively updated through backpropagation. The training is complete when the defined loss function converges. The feature matrix of the test set is input into the trained CNN sub-model and GRUA sub-model respectively, and the output spatial feature prediction sequence is obtained. and temporal feature prediction sequence And using a pre-trained fully connected layer with input of 2 and output of 1, spatial features are used to predict sequences. and temporal feature prediction sequence The data is then merged to obtain the final terminal area traffic prediction result.

[0050] Understandably, the CNN sub-model fully exploits the interrelationships between the multidimensional data of the input feature matrix through convolutional operations, extracts local spatial features using the CNN network, reduces data dimensionality through pooling operations, and extracts spatial features from the data, which is the prediction result of the CNN sub-model. The unique gate structure of the GRU sub-model gives it a memory function. Each GRU unit is input into the data sequentially according to the temporal relationship, which can fully consider the sequential relationship of the data and use each GRU unit to extract temporal features from the data. The idea of ​​the attention mechanism is to highlight important information by allocating sufficient attention to key information, while reducing or even ignoring some unimportant information. The attention mechanism assigns different weights to the output of each GRU unit and then passes it through a fully connected layer to obtain the prediction result of the GRUA sub-model.

[0051] Example 2

[0052] Based on the above embodiment 1, as follows Figure 6 As shown in the figure, this embodiment also provides a terminal area traffic prediction system based on multivariate time series analysis, including: a data acquisition module for acquiring basic data within the terminal area; a data preprocessing module adapted to preprocess the data acquired by the data acquisition module to construct a multivariate time series database; a terminal area dynamic multivariate time series data representation module adapted to segment and symbolically represent the data in the multivariate time series database; a terminal area traffic flow feature filtering module for performing feature filtering and dimensionality reduction on the segmented and symbolically represented multivariate time series data to construct a feature matrix; and a terminal area traffic prediction module for performing terminal area traffic prediction on the feature matrix and outputting the terminal area traffic prediction results, providing more reliable support for formulating reasonable control strategies.

[0053] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A terminal area traffic prediction method based on multivariate time series analysis, characterized by, The method comprises the following steps: Step 1, obtaining basic data in the terminal area; Step 2, preprocessing the basic data to construct a multivariate time series database for terminal area traffic prediction; Step 3, based on the trend and mean time series representation method, the data in the multivariate time series database is segmented and symbolically represented; Step 4, using a random forest model to determine the factors with high correlation with terminal area traffic prediction, and performing feature selection and dimension reduction on the segmented and symbolically represented multivariate time series data to construct a feature matrix; Step 5, constructing a terminal area traffic prediction model based on CNN-GRUA, and inputting the feature matrix into the trained terminal area traffic prediction model to output the terminal area traffic prediction result; The basic data obtained in step 1 includes: terminal area actual traffic, terminal area planned traffic, normal flight frequency, canceled flight frequency, flow capacity difference, flight delay, average delay time, flight delay frequency, flow control policy release intensity, average flow control interval, weather features in METAR messages, WAF convective weather and terminal area weather index; In step 2, the basic data is preprocessed to construct a multivariate time series database for terminal area traffic prediction, which comprises: The collected basic data of terminal area traffic is cleaned, that is, missing values are filled, noise data is smoothed, and outliers are identified and deleted; The basic data after data cleaning is constructed into multivariate time series data containing traffic demand features, delay features, strategy state features and environment state features to form a multivariate time series database; wherein The traffic demand features are used to reflect the traffic demand state, including terminal area actual traffic, terminal area planned traffic, normal flight frequency, canceled flight frequency and flow capacity difference; The delay features are used to reflect the delay state, including flight delay, average delay time and flight delay frequency; The strategy state features are used to reflect the terminal area strategy state, including flow control policy release intensity and average flow control interval; The environment state features are used to reflect the terminal area environment state, including weather features in METAR messages, WAF convective weather and terminal area weather index.

2. The terminal area traffic prediction method based on multivariate time series analysis according to claim 1, wherein In step 3, based on the trend and mean time series representation method, the data in the multivariate time series database is segmented and symbolically represented, which comprises: Based on the bisection iteration algorithm, all trend segmentation points on the normalized time series of the multivariate time series data are obtained; According to the preset subsequence length And the number of trend segmentation , using the non-overlapping sliding window mechanism to segment the entire time series into several sub-sequences, and using the corresponding strategy to make the number of trend segmentation points in each sub-sequence , forming Trend segmentation; Using a binary tuple based on trend features and mean statistics, each trend segment is symbolically represented.

3. The terminal area traffic prediction method based on multivariate time series analysis according to claim 2, wherein The method for adopting the corresponding strategy makes the number of trend segmentation points in each sub-sequence is The method includes: Finding the number of trend break points in each sub-sequence ; If , then select the most trend information from the trend segmentation points, form trend segmentation; like Then for the sequence length in the first position The trend is divided into two equal parts to form Each trend segment; If then directly adopt the average segmentation method to form trend segments; wherein denotes a rounding down. rounding down.

4. The terminal area traffic prediction method based on multivariate time series analysis according to claim 3, wherein The binary tuple based on trend features and mean statistics is used to symbolically represent each trend segment, which comprises: All trend segmentation points of the whole time series are divided into 3-10 equal-probability regions represented by different lowercase letters according to Gaussian distribution, and the whole range of slope values is divided into 5 non-overlapping regions represented by different capital letters; The slope value of each trend segmentation is calculated based on the trend segmentation points or endpoints at both ends of the trend segmentation, and the trend feature information of the trend segmentation is represented by the capital letter corresponding to the region where the slope value is located; The average value of each trend segmentation is calculated, and the mean statistical information of the trend segmentation is represented by the lowercase letter corresponding to the equal-probability region where the average value is located; Each trend segmentation is represented by a binary tuple of capital letter representing trend feature and lowercase letter representing mean statistics; The calculation formula of the average value of the trend segmentation is: ； wherein is a sequence of length of trend segments.

5. The terminal area traffic flow prediction method based on multivariate time series analysis according to claim 2, characterized in that, In step 4, the random forest model is used to determine the factors with high correlation with terminal area traffic flow prediction, and the segmented and symbolized multivariate time series data is subjected to feature selection and dimension reduction to construct a feature matrix, including: Based on the segmented and symbolized representation of the multivariate time series data, using mean square error as the evaluation criterion of features when dividing the nodes of decision tree, a random forest model is constructed by training a plurality of decision trees respectively The plurality of decision trees are trained respectively to construct a random forest model. The importance of all features in the segmented and symbolized multivariate time series data is evaluated using the constructed random forest model, and the top 8 features with the highest importance scores are selected to form the feature matrix.

6. The terminal area traffic flow prediction method based on multivariate time series analysis according to claim 5, characterized in that, In step 5, the CNN-GRUA-based terminal area traffic flow prediction model is constructed, and the feature matrix is input into the trained terminal area traffic flow prediction model to output the terminal area traffic flow prediction result, including: A CNN sub-model is constructed and trained by setting the number of layers, the size of the convolution kernel in each convolution layer, and the pooling method of the pooling layer; A GRUA sub-model is constructed and trained by setting the number of layers and the number of hidden layer neurons of the GRU network, and introducing an attention mechanism to assign appropriate weights to the output results of each GRU network; The CNN sub-model and the GRUA sub-model are fused by using a fully connected layer to construct and train a CNN-GRUA-based terminal area traffic flow prediction model; The feature matrix is input into the trained CNN sub-model and GRUA sub-model respectively, and the final terminal area traffic flow prediction result is obtained by fusing the spatial feature prediction result and the time series feature prediction result output by the two models.

7. A multivariate time series analysis based terminal area traffic prediction system employing the multivariate time series analysis based terminal area traffic prediction method according to any one of claims 1 to 6, characterized by Including: A data acquisition module for acquiring basic data in the terminal area; A data preprocessing module adapted to preprocess the data acquired by the data acquisition module to construct a multivariate time series database; A terminal area dynamic multivariate time series data representation module adapted to segment and symbolize the data in the multivariate time series database; A terminal area traffic flow feature selection module for feature selection and dimension reduction of the segmented and symbolized multivariate time series data to construct a feature matrix; A terminal area traffic flow prediction module for terminal area traffic flow prediction of the feature matrix to output a terminal area traffic flow prediction result.

Images