An LSTM track prediction system embedded with environmental attention coding

Abstract

The invention relates to the technical field of track prediction, and discloses an LSTM track prediction system embedded with environmental attention coding; the system comprises an airspace planning module, a data acquisition module, a model training module, a model prediction module and a track optimization module; an LSTM track prediction model for predicting the airspace track of the observation point at a future time is trained by collecting environmental data and track data of the track prediction airspace; the airspace track of the observation point at a future time is predicted by the LSTM track prediction model, and a corresponding track control instruction is formulated; the invention can capture the long-term dependency relationship between the environmental data and the track data by embedding the environmental data and the track data, obtain the LSTM track prediction model capable of predicting the airspace track at a future time, and predict the specific track position of an aircraft at a future time in advance, so as to know the change of the airspace track in advance, and effectively avoid the lag existing in the real-time track prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of track prediction, and more specifically, to an LSTM track prediction system embedded with environmental attention coding. Background Art

[0002] As the core technology of airspace situational awareness in the terminal area, trajectory prediction has a direct impact on the dynamic allocation efficiency of airspace resources and the level of flight safety operations in the terminal area. High-density traffic flow in the terminal area will significantly increase the probability of track intersection conflicts, and complex environmental disturbances and aircraft maneuvers will form a strong coupling effect, further increasing the unpredictability of aircraft tracks. In order to avoid uncontrollable aircraft tracks, it is necessary to accurately predict the aircraft's track position.

[0003] The patent application with reference publication number CN116894158A discloses a track prediction method based on LSTM, including step 1: acquisition and preprocessing of track sequence data set, wherein the acquisition and preprocessing of track sequence data set is to obtain The original data is processed into the required stable, uniform-length multivariate time series data set. Step 2: Construct an LSTM track prediction model. The purpose of constructing an LSTM track prediction model is to obtain the The data is trained and learned to realize the prediction of the track. Step 3: The track to be detected is input into the model to obtain the predicted track. The purpose of inputting the track to be detected into the model to obtain the predicted track is to use the model to realize the function of track prediction;

[0004] In the existing aircraft trajectory prediction, the aircraft's own navigation attitude data during the flight is collected and combined with the intelligent model to perform a single-dimensional trajectory prediction operation. For example, in the above patent application, The LSTM trajectory prediction model is trained with the data to achieve the effect of trajectory prediction through the aircraft's own navigation attitude data. However, in the actual navigation process of the aircraft, the aircraft's trajectory at future times will not only be affected by its own navigation attitude, but also by the changes in the external meteorological environment. This makes the trajectory prediction method based only on the single dimension of its own navigation attitude have limitations, and will also reduce the prediction accuracy of the aircraft's future trajectory, which is not conducive to the accurate prediction and control of the aircraft's future trajectory.

[0005] In view of this, the present invention proposes an LSTM track prediction system embedded with environmental attention coding to solve the above problems. Summary of the invention

[0006] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: an LSTM track prediction system embedded with environmental attention coding, applied to a track prediction server, comprising:

[0007] The airspace planning module is used to query the airspace parameters of the target airport and plan the track prediction airspace of the target airport based on the airspace planning criteria;

[0008] A data collection module is used to set a time period with an observation point and collect environmental data and track data of the track prediction airspace at the observation point;

[0009] The model training module is used to convert environmental data and track data into airspace data and airspace tracks, and train an LSTM track prediction model to predict the airspace track of the observation point at the future moment;

[0010] The model prediction module is used to collect the airspace data and airspace track of the target aircraft, predict the airspace track of the observation point at the future time through the LSTM track prediction model, and determine whether to issue a track abnormality warning information;

[0011] The track optimization module is used to formulate track control instructions for the target aircraft and send track control instructions to the target aircraft through the air traffic control center to control the target aircraft to optimize the track.

[0012] Furthermore, the airspace parameters include endpoint coordinates and control altitude values, and the endpoint coordinates include southeast coordinates, southwest coordinates, northwest coordinates, and northeast coordinates.

[0013] Furthermore, the airspace planning criterion is: the maximum value of the mid-range distance is used as the planning radius of the track prediction airspace.

[0014] Furthermore, the planning method of the trajectory prediction airspace is:

[0015] An electronic map of the target airport is queried from a database, and the southeast endpoint, southwest endpoint, northwest endpoint, and northeast endpoint of the target airport are marked on the electronic map, and lines are connected between the southeast endpoint and the northwest endpoint, and between the southwest endpoint and the northeast endpoint, to obtain a first diagonal line and a second diagonal line;

[0016] The lengths of the first diagonal and the second diagonal are measured by a scale, recorded as the first distance value and the second distance value, half of the first distance value and half of the second distance value are compared, and the maximum value after comparison is recorded as the extension value;

[0017] Take the southeast end point, southwest end point, northwest end point and northeast end point of the target airport as the starting point, take an extension value as the extension amplitude, draw extension lines in the southeast direction, southwest direction, northwest direction and northeast direction respectively, and record the endpoints on the four extension lines far away from the target airport as far points;

[0018] Measure the distances from the intersection of the first diagonal and the second diagonal to the far points on the four extension lines respectively, obtain four mid-extension distance values, use the maximum value of the mid-extension distance value as the radius, draw a circle with the intersection of the first diagonal and the second diagonal as the center, and draw the basic airspace;

[0019] In the basic airspace, the altitude of the target airport is taken as the base altitude, and an airspace corresponding to a controlled altitude value is expanded upward on the basic altitude to plan the predicted trajectory airspace.

[0020] Furthermore, the marking method of the observation points is:

[0021] The aircraft's navigation position is monitored in real time through a satellite positioning system, and the time when the navigation position is first located within the track prediction airspace is recorded as the starting time;

[0022] After the start time, if the aircraft's navigation position does not change within a calibrated navigation time, the time when the aircraft's navigation position last changes is recorded as the end time, and the period from the start time to the end time is recorded as the time period;

[0023] The start time is taken as the starting point of the time period, the end time is taken as the end point of the time period, and the preset observation time is used as the interval standard to mark A observation points between the start time and the end point.

[0024] Furthermore, the environmental data include wind direction azimuth, temperature value, precipitation, wind speed transverse component and wind speed longitudinal component;

[0025] The collection method of the transverse component and the longitudinal component of wind speed is:

[0026] The two endpoints of the target airport runway are marked by the satellite positioning system, and the distances from the two endpoints to the observation point are measured respectively, and the endpoint corresponding to the minimum distance value is recorded as the starting point;

[0027] Taking the starting point as the origin of the coordinate system, draw the X-axis along the horizontal direction of the target airport runway, and draw the Y-axis perpendicular to the target airport runway through the origin to construct the runway coordinate system;

[0028] The airflow velocity and airflow direction of A observation points are queried one by one through the meteorological automatic observation system, and A wind speed values ​​and A wind direction azimuths are obtained;

[0029] The runway azimuth of the target airport runway is inquired through the runway coordinate system, and after subtracting A wind direction azimuths from the runway azimuths one by one, the A differences are respectively subjected to vector decomposition calculation with A wind speed values ​​to obtain A wind speed transverse components and A wind speed longitudinal components;

[0030] The expression of the lateral component of wind speed is:

[0031] ;

[0032] In the formula, For the The lateral component of wind speed at each observation point is =1,2,...,A, For the The wind speed value at each observation point, For the The wind direction angle of each observation point, is the runway azimuth;

[0033] The expression of the longitudinal component of wind speed is:

[0034] ;

[0035] In the formula, For the The longitudinal component of wind speed at each observation point.

[0036] Furthermore, the track data includes real-time speed value, speed decay value and track coordinates;

[0037] The method for obtaining the speed attenuation value is:

[0038] The speed sensor collects the flight speed of the aircraft when entering the track prediction airspace and at A observation points in real time to obtain the initial speed value and A real-time speed values;

[0039] Subtract the initial speed value from the real-time speed value of the first observation point to obtain the speed attenuation value of the first observation point;

[0040] Subtract the real-time speed value of the remaining A-1 observation points from the next real-time speed value in turn to obtain A-1 speed attenuation values;

[0041] After summing up the speed attenuation value of the first observation point and A-1 speed attenuation values, A speed attenuation values ​​are obtained.

[0042] Furthermore, the training method of the LSTM track prediction model is:

[0043] The multiple sets of airspace data are converted into multiple feature vectors using the sliding window method. The airspace tracks are converted into labels corresponding to the airspace data according to the sliding step size. One feature vector corresponds to one label and constitutes a set of training data. Multiple sets of training data constitute a training set. The airspace data are arranged in the order of collection time.

[0044] The feature vector is used as the input of the LSTM trajectory prediction model, the airspace trajectory at the future moment after the predicted time step is used as the output, the subsequent airspace trajectory of each training set is used as the prediction target, and the sum of the minimized prediction errors is used as the training target. The LSTM trajectory prediction model is trained to predict the airspace trajectory at the future moment based on the airspace data.

[0045] Furthermore, the method for determining whether to issue a track abnormality warning message is as follows:

[0046] Split the track coordinates corresponding to the airspace track of the predicted observation point at the future moment and the airspace track of the previous observation point to obtain a first coordinate group and a second coordinate group respectively;

[0047] Subtract the X coordinate value, Y coordinate value, and Z coordinate value of the second coordinate group from the X coordinate value, Y coordinate value, and Z coordinate value of the first coordinate group to obtain an X change value, a Y change value, and a Z change value;

[0048] When the X change value is greater than the X change threshold, the X change value is recorded as an abnormal value;

[0049] When the Y change value is greater than the Y change threshold, the Y change value is recorded as an abnormal value;

[0050] When the Z change value is greater than the Z change threshold, the Z change value is recorded as an abnormal value;

[0051] Counting the number of abnormal values ​​in the first coordinate group to obtain abnormal values, and when the number of abnormal values ​​is 0, determining not to issue a track abnormality warning message;

[0052] When the number of abnormal values ​​is 1, 2 or 3, it is determined that a track abnormality warning message is issued.

[0053] Furthermore, the track control instruction includes an instruction for correcting a horizontal deviation, an instruction for correcting a lateral deviation, and an instruction for correcting a vertical deviation;

[0054] The method for formulating the corrected horizontal offset instruction, the corrected lateral offset instruction and the corrected vertical offset instruction is as follows:

[0055] When the abnormal value is the X change value, an instruction to correct the horizontal deviation is formulated;

[0056] When the abnormal value is the Y change value, an instruction to correct the lateral deviation is formulated;

[0057] When the abnormal value is the Z change value, an instruction to correct the vertical deviation is formulated.

[0058] The technical effects and advantages of the LSTM track prediction system embedded with environmental attention coding of the present invention are as follows:

[0059] (1): By setting a time period with observation points and collecting environmental data and track data of the track prediction airspace at the observation points, it is possible to effectively limit the collection time of the aircraft's track prediction related data and ensure the continuity of each data collection time, while also achieving the multi-dimensional collection effect of navigation position data and navigation environment data.

[0060] (2): By converting environmental data and trajectory data into airspace data and airspace trajectory, and training an LSTM trajectory prediction model that predicts the airspace trajectory of observation points at future times, the airspace trajectory of observation points at future times can be predicted through the LSTM trajectory prediction model. The environmental data and trajectory data can be embedded in the model so that the environmental data and trajectory data can be integrated and transformed with each other, and the long-term dependency between the environmental data and trajectory data can be captured. In this way, an LSTM trajectory prediction model that can predict the airspace trajectory at future times can be obtained, avoiding the limitations and inaccuracies of single-dimensional data training, and predicting the specific trajectory position of the aircraft at future times in advance, so that the changes in the airspace trajectory of the aircraft can be known in advance before the airspace trajectory of the aircraft changes abnormally, thereby effectively avoiding the lag in real-time trajectory prediction.

[0061] (3): By formulating the target aircraft's track control instructions and sending them to the target aircraft through the air traffic control center, the target aircraft can be controlled to optimize its track. Before predicting abnormal track phenomena in the future, corresponding track optimization control measures can be made in advance to avoid the aircraft's unpredictable track loss of control in the future, thereby effectively improving the aircraft's track accuracy prediction and control effect in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 A schematic diagram of the architecture of an LSTM track prediction system embedded with environmental attention coding provided in the first embodiment of the present invention;

[0063] Figure 2 A schematic diagram of a module of a track prediction server provided in Example 1 of the present invention;

[0064] Figure 3 A flowchart of an LSTM track prediction method with embedded environmental attention coding provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0065] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0066] Example 1: Please refer to Figure 1 and Figure 2 As shown, the LSTM track prediction system embedded with environmental attention coding described in this embodiment is applied to a track prediction server, including:

[0067] The airspace planning module queries the airspace parameters of the target airport and plans the track prediction airspace of the target airport based on the airspace planning criteria;

[0068] The target airport refers to the airport on the ground that can provide data collection and command control for the take-off and landing process of the aircraft, and is the object of the aircraft trajectory prediction this time. The airspace parameters refer to the diversified data that can represent the take-off and landing process of the target airport and the air area corresponding to the command control, so as to achieve the overall representation of the corresponding control area of ​​the target airport;

[0069] Airspace parameters include endpoint coordinates and control altitude values;

[0070] Endpoint coordinates refer to the specific coordinate data of the four boundary endpoints of the target airport's geographical location in the satellite positioning system, which can represent the specific location information of the four endpoints of the area covered by the target airport. Therefore, the endpoint coordinates include four independent position coordinates, namely the southeast coordinate, southwest coordinate, northwest coordinate and northeast coordinate; so that the southeast coordinate, southwest coordinate, northwest coordinate and northeast coordinate can correspond one by one to the southeast endpoint, southwest endpoint, northwest endpoint and northeast endpoint of the target airport respectively; the southeast coordinate, southwest coordinate, northwest coordinate and northeast coordinate are all obtained by querying the coordinate information of the southeast endpoint, southwest endpoint, northwest endpoint and northeast endpoint of the target airport one by one through the satellite positioning system.

[0071] The control altitude value refers to the altitude above the ground at which the target airport can ensure that the aircraft data is effectively collected and responds to instructions during the take-off and landing process and command control of the target airport, thereby serving as a basis for determining whether the aircraft can enter the control range of the target airport; the control altitude value is obtained by querying the database of the air traffic control department.

[0072] After the airspace parameters of the target airport are queried, the aircraft track control altitude area of ​​the target airport can be planned based on the airspace parameters and recorded as the track prediction airspace, so that the navigation data of the aircraft in the track prediction airspace can be used as the data collection object. Therefore, the track prediction airspace can also be used as the spatial limitation for subsequent aircraft track prediction;

[0073] In order to accurately construct the trajectory prediction airspace, it is necessary to plan based on airspace parameters and in combination with airspace planning criteria, so as to ensure that the planned trajectory prediction airspace can accurately represent the control area of ​​the target airport and avoid the interference of irrelevant and useless negative interference data on the trajectory prediction results;

[0074] The airspace planning principle is: the maximum value of the mid-range distance is used as the planning radius of the trajectory prediction airspace; this ensures that the planned trajectory prediction airspace can effectively control the airspace at all boundary locations of the target airport, thereby raising the threshold of trajectory prediction;

[0075] The planning method of trajectory prediction airspace is:

[0076] An electronic map of the target airport is queried from a database, and the southeast endpoint, southwest endpoint, northwest endpoint, and northeast endpoint of the target airport are marked on the electronic map, and lines are connected between the southeast endpoint and the northwest endpoint, and between the southwest endpoint and the northeast endpoint, to obtain a first diagonal line and a second diagonal line;

[0077] The lengths of the first diagonal and the second diagonal are measured by a scale, recorded as the first distance value and the second distance value, half of the first distance value and half of the second distance value are compared, and the maximum value after comparison is recorded as the extension value;

[0078] Take the southeast end point, southwest end point, northwest end point and northeast end point of the target airport as the starting point, take an extension value as the extension amplitude, draw extension lines in the southeast direction, southwest direction, northwest direction and northeast direction respectively, and record the endpoints on the four extension lines far away from the target airport as far points;

[0079] Measure the distance from the intersection of the first diagonal and the second diagonal to the far point on the four extension lines respectively, obtain four mid-range distance values, take the maximum value of the mid-range distance value as the radius, draw a circle with the intersection of the first diagonal and the second diagonal as the center, and draw the basic airspace; by drawing a circle, all the boundary positions of the target airport can be wrapped, so as to ensure that the planned track prediction airspace can fully represent the air above the target base;

[0080] In the basic airspace, the altitude of the target airport is taken as the base altitude, and an airspace corresponding to a controlled altitude value is expanded upward on the basic altitude to plan the predicted trajectory airspace.

[0081] It should be noted that the planned trajectory prediction airspace is an airspace with boundaries located above the target airport. Only aircraft entering the trajectory prediction airspace can perform a series of operations such as subsequent data collection and trajectory prediction.

[0082] The data acquisition module sets the time period of the track prediction airspace, marks the observation points of the time period, and collects the environmental data and track data of the aircraft at the observation points. The environmental data includes wind direction azimuth, temperature value, precipitation, wind speed lateral component and wind speed longitudinal component; the track data includes real-time speed value, speed attenuation value and track coordinates;

[0083] The time period refers to the time between the first time the aircraft generates relevant data in the track prediction airspace of the target airport and the last time the aircraft generates relevant data in the track prediction airspace of the target airport. It can be used to represent the flight time range of the aircraft in the track prediction airspace as a whole, and provide a time basis for data collection and analysis of the aircraft in the track prediction airspace.

[0084] An observation point refers to a moment in a time period used to collect and predict the change in the aircraft's track position in the track prediction airspace. It also serves as an accurate time limit for collecting relevant data of the aircraft in the track prediction airspace. In order to ensure that the aircraft can collect data and predict the track at multiple time points and continuously in the track prediction airspace, it is necessary to ensure that the number of observation points is sufficient and the time interval between two adjacent observation points is consistent.

[0085] The marking method of observation points is:

[0086] The aircraft's navigation position is monitored in real time through a satellite positioning system, and the time when the navigation position is first located within the track prediction airspace is recorded as the starting time;

[0087] After the start time, if the aircraft's navigation position does not change within a calibrated navigation time, the time when the aircraft's navigation position last changed is recorded as the end time, and the period from the start time to the end time is recorded as the time period; the calibrated navigation time refers to the maximum time when the aircraft's navigation position does not change, which can be used as the time basis for judging whether the aircraft has landed and is stationary, ensuring that the navigation position of an aircraft in navigation will change within a calibrated navigation time;

[0088] The start time is taken as the starting point of the time period, the end time is taken as the end point of the time period, and the preset observation time is used as the interval standard, and A observation points are marked between the start and end points. The preset observation time is used to represent the interval between two adjacent observation points to ensure that the time interval between any two adjacent observation points is consistent. At the same time, in order to ensure that the two adjacent observation points can collect refined data on the navigation position of the aircraft, the preset observation time should be shorter. For example, the preset observation time is 1 second or 2 seconds.

[0089] In this embodiment, the time period can be finely divided into time periods through A observation points, so that each observation point can be used as the data collection moment of the aircraft in the track prediction airspace, and also as the minimum time length for the aircraft's track to change.

[0090] Environmental data refers to the data of the air environment of an aircraft in the trajectory prediction airspace at each observation point in the past historical time, which can be used to represent the dynamic changes of the aircraft's air environment at each observation point in a diversified manner, and provide data support in the environmental dimension for the trajectory prediction and management of the aircraft in the trajectory prediction airspace;

[0091] Environmental data include wind direction azimuth, temperature, precipitation, transverse component of wind speed and longitudinal component of wind speed;

[0092] The wind direction azimuth refers to the direction of airflow in the airspace predicted by the aircraft's track when it is at the observation point, which can be used to indicate the blowing direction of the disturbed airflow that the aircraft is subject to at the observation point. The wind direction azimuth is obtained by querying the wind directions of A observation points one by one through the meteorological automatic observation system.

[0093] The temperature value refers to the ambient air temperature within the predicted airspace of the aircraft's flight path when the aircraft is at the observation point, which can be used to numerically represent the high and low air temperatures experienced by the aircraft at the observation point; the temperature value is obtained by querying the temperatures of A observation points one by one through the automatic meteorological observation system.

[0094] Precipitation refers to the amount of aerial precipitation in the airspace predicted by the aircraft's track when it is at the observation point, which can be used to numerically represent the magnitude of precipitation received by the aircraft at the observation point; the precipitation is obtained by querying the precipitation at A observation points one by one through the meteorological automatic observation system.

[0095] The transverse component of wind speed and the longitudinal component of wind speed are used to express the wind speed in the airspace of the aircraft's track prediction when it is at A observation points and the wind speed in the parallel and vertical directions of the target airport runway, respectively, which can numerically express the magnitude of the airflow disturbance to which the aircraft is subjected at the observation point;

[0096] The collection method of the transverse component and the longitudinal component of wind speed is:

[0097] The two endpoints of the target airport runway are marked by the satellite positioning system, and the distances from the two endpoints to the observation point are measured respectively, and the endpoint corresponding to the minimum distance value is recorded as the starting point;

[0098] Taking the starting point as the origin of the coordinate system, draw the X-axis along the horizontal direction of the target airport runway, and draw the Y-axis perpendicular to the target airport runway through the origin to construct the runway coordinate system; by constructing the runway coordinate system, it can provide data reference basis for the subsequent calculation of the horizontal and vertical components of the wind speed, thereby achieving the size conversion effect between the wind speed and the runway position;

[0099] The airflow velocity and airflow direction of A observation points are queried one by one through the meteorological automatic observation system, and A wind speed values ​​and A wind direction azimuths are obtained;

[0100] The runway azimuth of the target airport runway is inquired through the runway coordinate system, and after subtracting A wind direction azimuths from the runway azimuths one by one, the A differences are respectively subjected to vector decomposition calculation with A wind speed values ​​to obtain A wind speed transverse components and A wind speed longitudinal components;

[0101] The expression of the lateral component of wind speed is:

[0102] ;

[0103] In the formula, For the The lateral component of wind speed at each observation point is =1,2,...,A, For the The wind speed value at each observation point, For the The wind direction angle of each observation point, is the runway azimuth;

[0104] The expression of the longitudinal component of wind speed is:

[0105] ;

[0106] In the formula, For the The longitudinal component of wind speed at each observation point.

[0107] Track data refers to the data of the aircraft's own track at each observation point in the past historical time within the track prediction airspace, which can represent the dynamic changes of the aircraft's track at each observation point and provide data support in the navigation dimension for the aircraft's track prediction and management within the track prediction airspace;

[0108] Track data includes real-time speed value, speed decay value and track coordinates;

[0109] The real-time speed value refers to the real-time flight speed of the aircraft in the track prediction airspace at the observation point, which can be used to numerically represent the real-time flight speed of the aircraft at the observation point; the real-time speed value is obtained by monitoring the speed of A observation points in real time through a speed sensor.

[0110] The speed attenuation value refers to the decrease in the flight speed of the aircraft in the track prediction airspace when it is at two adjacent observation points, which can be used to numerically represent the degree of attenuation of the aircraft's real-time flight speed at the observation points;

[0111] The method for obtaining the speed attenuation value is:

[0112] The speed sensor collects the flight speed of the aircraft when entering the track prediction airspace and at A observation points in real time to obtain the initial speed value and A real-time speed values;

[0113] Subtract the initial speed value from the real-time speed value of the first observation point to obtain the speed attenuation value of the first observation point;

[0114] The expression of the speed attenuation value at the first observation point is:

[0115] ;

[0116] In the formula, is the speed attenuation value of the first observation point, is the initial speed value, is the real-time speed value of the first observation point;

[0117] Subtract the real-time speed value of the remaining A-1 observation points from the next real-time speed value in turn to obtain A-1 speed attenuation values;

[0118] The expression of speed attenuation is:

[0119] ;

[0120] In the formula, For the The speed attenuation value of each observation point, =1,2,...,A-1, For the The real-time speed value of each observation point, For the Real-time speed value of each observation point;

[0121] After summing up the speed attenuation value of the first observation point and A-1 speed attenuation values, A speed attenuation values ​​are obtained.

[0122] The track coordinates refer to the three-dimensional coordinates of the three-dimensional space of the track prediction airspace when the aircraft is at the observation point, which can be used to numerically represent the real-time position of the aircraft at the observation point; the track coordinates are obtained by querying the three-dimensional coordinates of A observation points one by one through the satellite positioning system.

[0123] The model training module converts environmental data and track data into airspace data and airspace tracks, and trains an LSTM track prediction model to predict the airspace track of the observation point at the future moment;

[0124] Airspace data refers to the comprehensive data of each observation point of the aircraft in the track prediction airspace, and serves as the input data for training the LSTM track prediction model. It can comprehensively represent the data that affects the aircraft's navigation trajectory in the track prediction airspace. The airspace track refers to the real-time navigation position of the aircraft at each observation point in the track prediction airspace, and serves as the output data of the LSTM track prediction model. It can also represent the specific position of the aircraft.

[0125] Since airspace data is used to represent the comprehensive data of each observation point, the airspace data includes wind direction azimuth, temperature value, precipitation, lateral component of wind speed, longitudinal component of wind speed, lateral component of speed and longitudinal component of speed, so as to realize the embedded effect of environmental data and location data, so as to achieve the technical effect of environmental attention coding embedding. At this time, the wind direction azimuth, temperature value, precipitation, lateral component of wind speed, longitudinal component of wind speed, lateral component of speed and longitudinal component of speed corresponding to A observation points can be summarized one by one to obtain the airspace data of A observation points, and at the same time, the track coordinates of A observation points are used as airspace tracks, so that A observation point can obtain the corresponding A airspace data and A airspace tracks, and A airspace data and A airspace tracks are arranged in chronological order.

[0126] After obtaining the airspace data and airspace tracks of A observation points of the aircraft, the airspace data and airspace tracks of each observation point of the aircraft can be used as the training data of the LSTM track prediction model, and the LSTM track prediction model that can predict the airspace track of the observation point at the future time is obtained by training, so as to ensure that the LSTM track prediction model can predict the airspace track of the observation point at the future time in advance based on the airspace data of the known observation point;

[0127] Before training the LSTM track prediction model, it is necessary to arrange the collected historical airspace data and airspace tracks of a large number of aircraft at A observation points in chronological order so that each airspace data can match the airspace track at the corresponding moment. At this time, the number of airspace data and airspace tracks are multiple groups to ensure that the LSTM track prediction model can be trained efficiently and accurately.

[0128] The training method of the LSTM track prediction model is:

[0129] The multiple sets of airspace data are converted into multiple feature vectors using the sliding window method. The airspace tracks are converted into labels corresponding to the airspace data according to the sliding step size. One feature vector corresponds to one label and constitutes a set of training data. Multiple sets of training data constitute a training set. The airspace data are arranged in the order of collection time, and the prediction time step size T, sliding step size Z and sliding window length N are preset.

[0130] The feature vector is used as the input of the LSTM trajectory prediction model, and the airspace trajectory at the future moment after the predicted time step T is used as the output. The subsequent airspace trajectory of each training set is used as the prediction target. The LSTM trajectory prediction model is trained with the minimized sum of prediction errors as the training target. An LSTM trajectory prediction model is trained to predict the airspace trajectory at the future moment based on the airspace data.

[0131] Specifically, an example of the sliding window method is as follows: If you want to use airspace data (A1, A2, A3, A4, A5, A6) to train an LSTM track prediction model to predict the value of one time step in the future, you can use a sliding window of length 4 and a sliding step of length 1 to generate a predicted future training set and prediction target. For example, the training sets are (A1, A2, A3, A4) and (A2, A3, A4, A5), and the prediction targets are (B5) and (B6). B5 is the airspace track corresponding to the airspace data of the next observation point of the observation point where A4 is located, and B6 is the airspace track corresponding to the airspace data of the next observation point of the observation point where A5 is located.

[0132] The LSTM trajectory prediction model using the sliding window method can accurately and in advance predict the airspace trajectory of observation points at future times based on the airspace data of the aircraft's existing observation points, thereby knowing the changes in the airspace trajectory of the aircraft in advance before the airspace trajectory of the aircraft changes abnormally, thereby effectively avoiding the lag in real-time trajectory prediction and achieving a super-timeline trajectory prediction effect.

[0133] The model prediction module collects the airspace data and airspace track of the target aircraft, predicts the airspace track of the observation point at the future time through the LSTM track prediction model, and determines whether to issue a track abnormality warning message;

[0134] The target aircraft refers to the aircraft for which trajectory prediction is required this time. By collecting the airspace data and airspace trajectory of the target aircraft, it can be imported into the trained LSTM trajectory prediction model, so that the LSTM trajectory prediction model can predict the airspace trajectory of the target aircraft at the observation point in the future in advance based on the airspace data and airspace trajectory of the target aircraft's existing observation points.

[0135] After predicting the airspace track of the target aircraft at the observation point at a future time, the predicted airspace track can be identified for track anomalies to determine whether the target aircraft will have track anomalies at a future time, thereby determining whether to issue track anomaly warning information;

[0136] The method for determining whether to issue a track abnormality warning message is as follows:

[0137] Split the track coordinates corresponding to the airspace track of the predicted observation point at the future moment and the airspace track of the previous observation point to obtain a first coordinate group and a second coordinate group respectively;

[0138] Subtract the X coordinate value, Y coordinate value, and Z coordinate value of the second coordinate group from the X coordinate value, Y coordinate value, and Z coordinate value of the first coordinate group to obtain an X change value, a Y change value, and a Z change value;

[0139] The X change value, Y change value and Z change value are compared with the X change threshold value, Y change threshold value and Z change threshold value respectively one by one; the X change threshold value, Y change threshold value and Z change threshold value refer to the maximum values ​​when the X change value, Y change value and Z change value are recorded as abnormal values, so as to serve as a numerical basis for judging whether the X change value, Y change value and Z change value are abnormal values. When the X change value, Y change value and Z change value exceed the corresponding change threshold value, it means that the X change value, Y change value and Z change value have a phenomenon of excessive change amplitude;

[0140] When the X change value is greater than the X change threshold, it means that the track change amplitude of the target aircraft in the X-axis direction at the observation point at the future time is too large, and the X change value is recorded as an abnormal value;

[0141] When the Y change value is greater than the Y change threshold, it means that the track change amplitude of the target aircraft in the Y-axis direction at the observation point at the future time is too large, and the Y change value is recorded as an abnormal value;

[0142] When the Z change value is greater than the Z change threshold, it means that the track change amplitude of the target aircraft in the Z-axis direction at the observation point at the future time is too large, and the Z change value is recorded as an abnormal value;

[0143] The number of outliers in the first coordinate group is counted to obtain an outlier value. When the number of outliers is 0, it means that the airspace track of the observation point at the predicted future time is normal, and it is determined that no track abnormality warning information is issued;

[0144] When the number of outliers is 1, 2 or 3, it indicates that the airspace track of the observation point at the predicted future time is abnormal, and it is determined that a track abnormality warning message is issued.

[0145] The track optimization module formulates the track control instructions of the target aircraft and sends the track control instructions to the target aircraft through the air traffic control center to control the target aircraft to optimize the track;

[0146] When it is determined that an abnormal track warning message is issued, it means that the target aircraft's track in the airspace at the observation point at a future moment will be abnormal. If the target aircraft continues to follow the existing flight attitude, unpredictable changes in the track may occur within the track prediction airspace, resulting in the flight trajectory of the target aircraft being unable to be accurately collected and controlled by the air traffic control center, which may easily lead to serious navigation accidents. Therefore, it is necessary to carry out corresponding track optimization control for the target aircraft that issues the abnormal track warning message.

[0147] Track control instructions are specific measures used to optimize the target aircraft's track to ensure that the target aircraft can optimize the airspace track at the future observation point according to the track control instructions, and prevent the occurrence of airspace track anomalies at the future observation point, so as to facilitate the air traffic control center to accurately control the flight trajectory of the target aircraft;

[0148] The track control instruction includes a horizontal deviation correction instruction, a lateral deviation correction instruction and a vertical deviation correction instruction; wherein, the horizontal deviation correction instruction refers to adjusting the pitch angle between the target aircraft and the target airport runway in the track prediction airspace, the lateral deviation correction instruction refers to adjusting the horizontal angle between the target aircraft and the target airport runway in the track prediction airspace, and the vertical deviation correction instruction refers to adjusting the descent height of the target aircraft in the track prediction airspace;

[0149] The method for formulating the corrected horizontal offset instruction, the corrected lateral offset instruction and the corrected vertical offset instruction is as follows:

[0150] When the abnormal value is the X change value, it means that the pitch angle between the target aircraft and the target airport runway in the track prediction airspace needs to be adjusted, and a correction horizontal deviation instruction is formulated;

[0151] When the abnormal value is the Y change value, it means that the horizontal angle between the target aircraft and the target airport runway in the track prediction airspace needs to be adjusted, and a correction lateral deviation instruction is formulated;

[0152] When the abnormal value is the Z change value, it means that the descent altitude of the target aircraft in the track prediction airspace needs to be adjusted, and a correction vertical deviation instruction is formulated.

[0153] When the corresponding track control instructions are formulated, the air traffic control center can send them to the target aircraft in a timely manner according to the specific instructions formulated, so that the target aircraft can receive the corresponding track control instructions in advance and optimize and adjust its own flight status according to the track control instructions, so as to avoid the abnormal airspace track of the target aircraft at the observation point in the future;

[0154] Specifically, optimizing and adjusting its own flight status includes but is not limited to increasing or decreasing the target aircraft's cruising speed, increasing or decreasing the target aircraft's pitch angle, and other operations, so as to achieve the effect of controlling the target aircraft to optimize the airspace track, thereby preventing the target aircraft from having airspace track abnormalities at future observation points.

[0155] In this embodiment, by querying the airspace parameters of the target airport and planning the track prediction airspace of the target airport based on the airspace planning criteria, the area corresponding to the aircraft's track prediction can be accurately planned and limited, ensuring that the planned track prediction airspace can effectively control all boundary locations of the target airport, thereby raising the threshold for track prediction.

[0156] By setting a time period with observation points and collecting the environmental data and track data of the track prediction airspace at the observation points, it is possible to effectively limit the collection time of the aircraft's track prediction related data and ensure the continuity of each data collection time, while also achieving the multi-dimensional collection effect of navigation position data and navigation environment data, providing comprehensive and diversified data support for subsequent track prediction.

[0157] By converting environmental data and track data into airspace data and airspace track, and training an LSTM track prediction model for predicting the airspace track of observation points at future times, the airspace track of observation points at future times can be predicted by the LSTM track prediction model, and it can be determined whether to issue a track abnormality warning message, so that the environmental data and track data can be embedded, so that the environmental data and track data can be mutually integrated and transformed, and based on the data basis of embedded environmental attention encoding, combined with the order of acquisition time series, the long-term dependency between environmental data and track data can be captured, and the LSTM track prediction model that can predict the airspace track at future times can be obtained, avoiding the limitations and inaccuracies in single-dimensional data training, and using the LSTM track prediction model to predict the specific track position of the aircraft at future times in advance, so that the changes in the airspace track can be known in advance before the abnormal changes in the airspace track of the aircraft occur, thereby effectively avoiding the lag in real-time track prediction and achieving the track prediction effect of the super timeline;

[0158] By formulating track control instructions for the target aircraft and sending them to the target aircraft through the air traffic control center, the target aircraft can be controlled to optimize its track. Before predicting abnormal track phenomena in the future, corresponding track optimization control measures can be made in advance to avoid the aircraft's unpredictable track loss of control in the future, thereby effectively improving the aircraft's track accuracy prediction and control effects in complex environments.

[0159] Example 2: Please refer to Figure 3 As shown, the part not described in detail in this embodiment is described in the first embodiment, and a LSTM track prediction method embedded with environmental attention coding is provided, which is applied to a track prediction server and is implemented based on an LSTM track prediction system embedded with environmental attention coding, including:

[0160] S1: Query the airspace parameters of the target airport, and plan the track prediction airspace of the target airport based on the airspace planning criteria;

[0161] S2: Set a time period with observation points, and collect environmental data and track data of the track prediction airspace at the observation points;

[0162] S3: Convert environmental data and track data into airspace data and airspace track, and train an LSTM track prediction model to predict the airspace track of the observation point at the future moment;

[0163] S4: Collect the airspace data and airspace track of the target aircraft, predict the airspace track of the observation point at the future time through the LSTM track prediction model, and determine whether to issue a track abnormality warning information;

[0164] S5: If a track abnormality warning message is issued, a track control instruction for the target aircraft is formulated and sent to the target aircraft through the air traffic control center to control the target aircraft to optimize its track.

[0165] In S1, airspace parameters include endpoint coordinates and control altitude values. The endpoint coordinates include southeast coordinates, southwest coordinates, northwest coordinates, and northeast coordinates. The airspace planning criteria are: the maximum value of the mid-range distance value is the planning radius of the track prediction airspace;

[0166] In S2, the environmental data include wind direction azimuth, temperature value, precipitation, wind speed transverse component and wind speed longitudinal component, and the track data include real-time speed value, speed attenuation value and track coordinates;

[0167] In S5, the track control instruction includes a corrected horizontal deviation instruction, a corrected lateral deviation instruction and a corrected vertical deviation instruction.

[0168] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An LSTM track prediction system embedded with environmental attention coding, applied to a track prediction server, characterized in that: include: The airspace planning module is used to query the airspace parameters of the target airport and plan the track prediction airspace of the target airport based on the airspace planning criteria; A data collection module is used to set a time period with an observation point and collect environmental data and track data of the track prediction airspace at the observation point; The model training module is used to convert environmental data and track data into airspace data and airspace tracks, and train an LSTM track prediction model to predict the airspace track of the observation point at the future moment; Environmental data include wind direction azimuth, temperature, precipitation, wind speed lateral component and wind speed longitudinal component; track data include real-time speed value, speed attenuation value and track coordinates; Airspace data include wind direction azimuth, temperature, precipitation, wind speed lateral component, wind speed longitudinal component, ship speed lateral component and ship speed longitudinal component; The training method of the LSTM track prediction model is: The multiple sets of airspace data are converted into multiple feature vectors using the sliding window method. The airspace tracks are converted into labels corresponding to the airspace data according to the sliding step size. One feature vector corresponds to one label and constitutes a set of training data. Multiple sets of training data constitute a training set. The airspace data are arranged in the order of collection time. The feature vector is used as the input of the LSTM track prediction model, and the airspace track at the future moment after the predicted time step is used as the output. The subsequent airspace track of each training set is used as the prediction target. The sum of the minimized prediction errors is used as the training target. The LSTM track prediction model is trained to predict the airspace track at the future moment based on the airspace data. The model prediction module is used to collect the airspace data and airspace track of the target aircraft, predict the airspace track of the observation point at the future time through the LSTM track prediction model, and determine whether to issue a track abnormality warning information; The track optimization module is used to formulate track control instructions for the target aircraft and send track control instructions to the target aircraft through the air traffic control center to control the target aircraft to optimize the track.

2. The LSTM track prediction system embedded with environmental attention coding according to claim 1 is characterized in that: The airspace parameters include endpoint coordinates and control altitude values. The endpoint coordinates include southeast coordinates, southwest coordinates, northwest coordinates and northeast coordinates.

3. The LSTM track prediction system embedded with environmental attention coding according to claim 2 is characterized in that: The airspace planning principle is: the maximum value of the mid-range distance is the planning radius of the trajectory prediction airspace.

4. The LSTM track prediction system embedded with environmental attention coding according to claim 3 is characterized in that: The planning method of trajectory prediction airspace is: An electronic map of the target airport is queried from a database, and the southeast endpoint, southwest endpoint, northwest endpoint, and northeast endpoint of the target airport are marked on the electronic map, and lines are connected between the southeast endpoint and the northwest endpoint, and between the southwest endpoint and the northeast endpoint, to obtain a first diagonal line and a second diagonal line; The lengths of the first diagonal and the second diagonal are measured by a scale, recorded as the first distance value and the second distance value, half of the first distance value and half of the second distance value are compared, and the maximum value after comparison is recorded as the extension value; Take the southeast end point, southwest end point, northwest end point and northeast end point of the target airport as the starting point, take an extension value as the extension amplitude, draw extension lines in the southeast direction, southwest direction, northwest direction and northeast direction respectively, and record the endpoints on the four extension lines far away from the target airport as far points; Measure the distances from the intersection of the first diagonal and the second diagonal to the far points on the four extension lines respectively, obtain four mid-extension distance values, use the maximum value of the mid-extension distance value as the radius, draw a circle with the intersection of the first diagonal and the second diagonal as the center, and draw the basic airspace; In the basic airspace, the altitude of the target airport is taken as the base altitude, and an airspace corresponding to a controlled altitude value is expanded upward on the basic altitude to plan the predicted trajectory airspace.

5. The LSTM track prediction system embedded with environmental attention coding according to claim 4 is characterized in that: The marking method of observation points is: The aircraft's navigation position is monitored in real time through a satellite positioning system, and the time when the navigation position is first located within the track prediction airspace is recorded as the starting time; After the start time, if the aircraft's navigation position does not change within a calibrated navigation time, the time when the aircraft's navigation position last changes is recorded as the end time, and the period from the start time to the end time is recorded as the time period; The start time is taken as the starting point of the time period, the end time is taken as the end point of the time period, and the preset observation time is used as the interval standard to mark A observation points between the start time and the end point.

6. The LSTM track prediction system embedded with environmental attention coding according to claim 5, characterized in that: The collection method of the transverse component and the longitudinal component of wind speed is: The two endpoints of the target airport runway are marked by the satellite positioning system, and the distances from the two endpoints to the observation point are measured respectively, and the endpoint corresponding to the minimum distance value is recorded as the starting point; Taking the starting point as the origin of the coordinate system, draw the X-axis along the horizontal direction of the target airport runway, and draw the Y-axis perpendicular to the target airport runway through the origin to construct the runway coordinate system; The airflow velocity and airflow direction of A observation points are queried one by one through the meteorological automatic observation system, and A wind speed values ​​and A wind direction azimuths are obtained; The runway azimuth of the target airport runway is inquired through the runway coordinate system, and after subtracting A wind direction azimuths from the runway azimuths one by one, the A differences are respectively subjected to vector decomposition calculation with A wind speed values ​​to obtain A wind speed transverse components and A wind speed longitudinal components; The expression of the lateral component of wind speed is: ； In the formula, For the The lateral component of wind speed at each observation point is =1,2,...,A, For the The wind speed value at each observation point, For the The wind direction angle of each observation point, is the runway azimuth; The expression of the longitudinal component of wind speed is: ； In the formula, For the The longitudinal component of wind speed at each observation point.

7. The LSTM track prediction system embedded with environmental attention coding according to claim 6, characterized in that: The method for obtaining the speed attenuation value is: The speed sensor collects the flight speed of the aircraft when entering the track prediction airspace and at A observation points in real time to obtain the initial speed value and A real-time speed values; Subtract the initial speed value from the real-time speed value of the first observation point to obtain the speed attenuation value of the first observation point; Subtract the real-time speed value of the remaining A-1 observation points from the next real-time speed value in turn to obtain A-1 speed attenuation values; After summing up the speed attenuation value of the first observation point and A-1 speed attenuation values, A speed attenuation values ​​are obtained.

8. The LSTM track prediction system embedded with environmental attention coding according to claim 7, characterized in that: The method for determining whether to issue a track abnormality warning message is as follows: Split the track coordinates corresponding to the airspace track of the predicted observation point at the future moment and the airspace track of the previous observation point to obtain a first coordinate group and a second coordinate group respectively; Subtract the X coordinate value, Y coordinate value, and Z coordinate value of the second coordinate group from the X coordinate value, Y coordinate value, and Z coordinate value of the first coordinate group to obtain an X change value, a Y change value, and a Z change value; When the X change value is greater than the X change threshold, the X change value is recorded as an abnormal value; When the Y change value is greater than the Y change threshold, the Y change value is recorded as an abnormal value; When the Z change value is greater than the Z change threshold, the Z change value is recorded as an abnormal value; Counting the number of abnormal values ​​in the first coordinate group to obtain abnormal values, and when the number of abnormal values ​​is 0, determining not to issue a track abnormality warning message; When the number of abnormal values ​​is 1, 2 or 3, it is determined that a track abnormality warning message is issued.

9. The LSTM track prediction system embedded with environmental attention coding according to claim 8, characterized in that: The track control instructions include the instructions for correcting the horizontal deviation, the instructions for correcting the lateral deviation and the instructions for correcting the vertical deviation; The method for formulating the corrected horizontal offset instruction, the corrected lateral offset instruction and the corrected vertical offset instruction is as follows: When the abnormal value is the X change value, an instruction to correct the horizontal deviation is formulated; When the abnormal value is the Y change value, an instruction to correct the lateral deviation is formulated; When the abnormal value is the Z change value, an instruction to correct the vertical deviation is formulated.

Images