Dynamic path planning method and system based on generative adversarial network model

Abstract

The invention provides a dynamic path planning method and system based on a generative adversarial network model, and the method comprises the steps: firstly receiving a flight task instruction of an air traffic control system and an airspace real-time operation data set, inputting the flight task instruction and the airspace real-time operation data set into a generator unit of a generative adversarial network, and generating a plurality of candidate dynamic flight paths; and calling a discriminator unit to evaluate the candidate dynamic flight paths and output comprehensive evaluation values by taking the airspace control constraint rule and the path operation efficiency index as references, feeding back the candidate dynamic flight path with the highest comprehensive evaluation value to the generator unit to adjust parameters, and repeating evaluation and adjustment operations until a preset optimization threshold value is reached. And the final path is sent to an air traffic control command system, so that the airspace utilization rate and the flight operation efficiency can be improved, and the flight safety is guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of air transportation, in particular to a dynamic path planning method and system based on a generative adversarial network model. BACKGROUND

[0002] In the field of air transportation, dynamic path planning is a key link to ensure flight safety, improve airspace utilization rate and enhance operation efficiency. Traditional flight path planning methods mainly rely on pre-set fixed routes and manual planning based on experience. Although the fixed route planning method is simple and easy to implement, it lacks flexibility and cannot be dynamically adjusted according to the real-time operating conditions of the airspace. When unexpected situations occur in the airspace, such as severe weather, aircraft malfunctions or other airspace control requirements, fixed routes may cause flight delays, increased detour distances, and even air traffic congestion, increasing the risk of flight safety.

[0003] Although manual planning paths can be adjusted to some extent in combination with real-time information, it is difficult for planners to comprehensively and accurately plan a large number of flight tasks in a short period of time due to their experience and energy. Moreover, manual planning cannot fully consider the dynamic changes of all aircraft in the airspace and the complex airspace control constraints, which may lead to unreasonable planning, path conflicts or violations of control regulations.

[0004] In addition, existing dynamic path planning methods mostly consider only a single dimension of safety or efficiency, lacking a comprehensive trade-off between safety and efficiency. Under the premise of ensuring flight safety, the potential of airspace resources cannot be fully tapped to improve the operation efficiency of flight paths, and the demand for efficient, safe and flexible path planning in the modern air transportation industry cannot be met. SUMMARY

[0005] In view of the above-mentioned problems, in combination with the first aspect of the present application, the embodiments of the present application provide a dynamic path planning method based on a generative adversarial network model, which comprises: receiving a current flight task instruction and a corresponding airspace real-time operating data set sent by an air traffic control system, the flight task instruction containing a takeoff airport identifier, a landing airport identifier and a flight task priority, and the airspace real-time operating data set containing real-time position information of existing aircraft in the airspace, a time-space range of occupied air routes and airspace control constraint rules; inputting the flight task instruction and the airspace real-time operating data set into a generator unit of a pre-trained generative adversarial network to generate a plurality of candidate dynamic flight paths covering the entire flight, each candidate dynamic flight path containing consecutive waypoint geographic position coordinates, flight segment headings between waypoints and predicted passing times of corresponding flight segments; The discriminator unit of the generative adversarial network is called to perform safety compliance checking and efficiency effectiveness evaluation on each candidate dynamic flight path based on the airspace regulation constraint rule and in combination with a path operation efficiency index, and output a comprehensive evaluation value of each candidate dynamic flight path. The candidate dynamic flight path with the highest comprehensive evaluation value is taken as a feedback path, and is input into a parameter adjustment module of the generator unit to adjust a waypoint distribution probability of path generation and a flight segment time distribution weight in the generator unit. The candidate dynamic flight path is regenerated based on the adjusted generator unit, and the evaluation of the discriminator unit and the parameter adjustment of the generator unit are repeatedly performed until the comprehensive evaluation value of the generated candidate dynamic flight path reaches a preset path optimization threshold, and the candidate dynamic flight path is taken as a final dynamic flight path and is sent to an air traffic control system.

[0006] In still another aspect, the embodiment of the present application further provides a dynamic path planning system based on a generative adversarial network model, comprising a processor and a machine readable storage medium, the machine readable storage medium is connected with the processor, the machine readable storage medium is used for storing programs, instructions or codes, and the processor is used for executing the programs, instructions or codes in the machine readable storage medium to realize the above method.

[0007] Based on the above aspects, the embodiment of the present application comprehensively obtains key information of a flight task and real-time dynamic conditions of an airspace by receiving a current flight task instruction and a corresponding airspace real-time operation data set sent by an air traffic control system, inputs the flight task instruction and the airspace real-time operation data set into a generator unit of a pre-trained generative adversarial network, and can generate a plurality of candidate dynamic flight paths covering the whole flight, fully utilizes the powerful generation capability of the generative adversarial network, improves the diversity and flexibility of path generation, calls a discriminator unit of the generative adversarial network, performs safety compliance checking and efficiency effectiveness evaluation on each candidate dynamic flight path based on an airspace regulation constraint rule and in combination with a path operation efficiency index, and outputs a comprehensive evaluation value, realizes multi-dimensional and comprehensive evaluation of the path, ensures that the generated path meets safety requirements and has high operation efficiency, feeds back the candidate dynamic flight path with the highest comprehensive evaluation value to a parameter adjustment module of the generator unit, adjusts a waypoint distribution probability of path generation and a flight segment time distribution weight, and realizes continuous optimization of the candidate dynamic flight path by repeatedly performing the evaluation of the discriminator unit and the parameter adjustment of the generator unit until a preset path optimization threshold is reached, and finally generates a final dynamic flight path that can better adapt to real-time changes of the airspace, ensures flight safety, and significantly improves airspace utilization and flight operation efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0008] Figure 1 is an execution flow diagram of the dynamic path planning method based on the generative adversarial network model provided by an embodiment of the present application.

[0009] Figure 2 is a schematic diagram of exemplary hardware and software components of the dynamic path planning system based on the generative adversarial network model provided by an embodiment of the present application. DETAILED DESCRIPTION

[0010] The present application will be described in detail below with reference to the accompanying drawings, Figure 1 is a flow diagram of the dynamic path planning method based on the generative adversarial network model provided by an embodiment of the present application, which will be described in detail below.

[0011] Step S110: receiving the current flight task instruction and the corresponding airspace real-time running data set sent by the air traffic control system, wherein the flight task instruction includes the takeoff airport identifier, the landing airport identifier and the flight task priority, and the airspace real-time running data set includes the real-time position information of the existing aircraft in the airspace, the time-space range of the occupied air route and the airspace control constraint rules.

[0012] In this embodiment, the air traffic control system can continuously send various information to the path planning system, and when a new flight task is generated, the path planning system will receive the corresponding current flight task instruction and airspace real-time running data set. The takeoff airport identifier and the landing airport identifier in the flight task instruction are specific strings, which are used to uniquely determine the corresponding takeoff airport and landing airport. The flight task priority is represented by a group of codes, and different codes correspond to different emergency levels and importance.

[0013] The real-time position information of the existing aircraft in the airspace in the airspace real-time running data set is obtained and sent to the air traffic control system by the positioning device on each aircraft, and then forwarded by the air traffic control system. The real-time position information of each aircraft includes its latitude and longitude coordinates and flight altitude layer. The time-space range of the occupied air route refers to the coverage of the air route in time and space that has been used by other aircraft, wherein the spatial range is the latitude and longitude interval passed by the air route, and the time range is the starting time and ending time of the aircraft flying on the air route. The airspace control constraint rules are a series of regulations made by the aviation management department, which cover various restrictions and requirements that need to be followed during the flight of the aircraft.

[0014] Step S120: input the flight task instruction and the airspace real-time running data set into a generator unit of a pre-trained generative adversarial network to generate a plurality of candidate dynamic flight paths covering the entire flight, each candidate dynamic flight path including continuous waypoint geographic position coordinates, flight segment headings between waypoints, and estimated passing time of the corresponding flight segment.

[0015] The generator unit of the generative adversarial network is pre-trained and can generate reasonable flight paths according to the input information. After inputting the flight task instruction and the airspace real-time running data set into the generator unit, the generator unit can process and analyze the flight task instruction and the airspace real-time running data set, combine the learned path generation mode, and generate a plurality of different candidate dynamic flight paths.

[0016] The continuous waypoint geographic position coordinates in each candidate dynamic flight path are a series of longitude and latitude coordinates of points planned by the generator unit according to the positions of the takeoff airport and the landing airport and the conditions in the airspace. The flight segment heading between waypoints refers to the direction from one waypoint to the next waypoint, usually represented by the azimuth angle. The estimated passing time of the corresponding flight segment is estimated according to the length of the flight segment, the estimated flight speed of the aircraft, and the traffic conditions in the airspace.

[0017] Step S121: extract the takeoff airport identifier corresponding to the takeoff airport geographic position coordinates and the landing airport identifier corresponding to the landing airport geographic position coordinates from the flight task instruction as the starting point coordinates and the end point coordinates of the flight path.

[0018] After receiving the flight task instruction, the path planning system can query the pre-stored airport geographic position coordinate database according to the takeoff airport identifier and the landing airport identifier. The airport geographic position coordinate database stores the mapping relationship between all airport identifiers and their corresponding precise longitude and latitude coordinates.

[0019] By matching the takeoff airport identifier, the corresponding takeoff airport geographic position coordinates are found in the airport geographic position coordinate database, which are determined as the starting point coordinates of the flight path. Similarly, by matching the landing airport identifier, the corresponding landing airport geographic position coordinates are found, which are used as the end point coordinates of the flight path. These coordinates are longitude and latitude values in degrees, which can accurately locate the position of the airport.

[0020] Step S122: extract the real-time position information of existing aircraft in the airspace and the spatiotemporal range of occupied air routes from the airspace real-time running data set to construct an airspace resource occupation heat map, which is used to represent the air route occupation density of different regions in the airspace at different times.

[0021] The real-time position information of the existing aircraft in the airspace extracted from the airspace real-time operation data set and the time-space range of the occupied air route are the basic data for constructing the airspace resource occupation heat map. The real-time position information of the existing aircraft can reflect the distribution of the aircraft in the airspace at the current time, and the time-space range of the occupied air route can reflect the use of the air route in the past and future period of time.

[0022] When constructing the airspace resource occupation heat map, the airspace needs to be divided first, then the occupation situation of each divided area at different time is counted, the occupation density is determined according to the occupation situation, and finally different colors or gray scales are used to represent different occupation densities to form the heat map.

[0023] Step S1221: Extract the real-time position information of each existing aircraft from the airspace real-time operation data set, the real-time position information including longitude and latitude coordinates and flight altitude layer.

[0024] The information of each existing aircraft is stored in the airspace real-time operation data set according to the unique identifier of the aircraft, and the path planning system can read these information one by one to extract the real-time position information of each aircraft. The longitude and latitude coordinates are represented by two values respectively, and the flight altitude layer is a specific value representing the altitude level of the aircraft at the current time. These extracted real-time position information is stored in a data structure separately for subsequent use when constructing the airspace resource occupation heat map.

[0025] Step S1222: Extract the time-space range of each occupied air route, the time-space range of each occupied air route including the start time, end time and longitude and latitude range covered by the air route.

[0026] In the airspace real-time operation data set, the information of the occupied air route is also stored in a set format, and the path planning system can traverse all the records of the occupied air route to extract the start time, end time and longitude and latitude range covered by each air route. The start time and end time are represented by time stamps, accurate to seconds. The longitude and latitude range covered by the air route is represented by two longitude and latitude intervals, corresponding to the start and end longitude and latitude and the longitude and latitude range passed through.

[0027] Step S1223: Divide the airspace into a plurality of equal-area grid cells, the longitude and latitude range of each grid cell being the same.

[0028] When dividing the airspace, the size of the grid cell needs to be determined according to the longitude and latitude span of the entire airspace. First, the maximum longitude, minimum longitude, maximum latitude and minimum latitude of the airspace are determined, and then the longitude span and latitude span are calculated. According to the requirement of equal area, the longitude range and latitude range of each grid cell are calculated so that the area of each grid cell is equal. After the division is completed, each grid cell has its unique identification, and the corresponding longitude and latitude range is clear.

[0029] Step S1224: For each grid cell, the number of times that the real-time position information of the existing aircraft falls into the grid cell within the preset time window is counted as the real-time occupation number of the grid cell.

[0030] The preset time window is a pre-set time interval, for example, the past 30 minutes to the current time. For each grid cell, the path planning system can check the number of times that the longitude and latitude coordinates of all existing aircrafts' real-time position information fall within the longitude and latitude range of the grid cell within the time window.

[0031] When counting, the real-time position information of each aircraft at each time within the preset time window can be viewed one by one to determine whether it is within the grid cell, and if so, the real-time occupation number is incremented by one. After the counting is completed, each grid cell has a corresponding real-time occupation number.

[0032] Step S1225: The length of the time period covered by the occupied route in the grid cell within the preset time window is counted as the historical occupation duration of the grid cell.

[0033] For each grid cell, the path planning system can check the spatiotemporal range of all occupied routes to determine whether it covers the grid cell. If the longitude and latitude range of a certain occupied route contains the longitude and latitude range of the grid cell, and its time range overlaps with the preset time window, then the length of the overlapping time period is calculated. The sum of the overlapping time period lengths of all occupied routes that cover the grid cell is the historical occupation duration of the grid cell.

[0034] Step S1226: The resource occupation density value of each grid cell is obtained by combining the real-time occupation number with the preset real-time weight and the historical occupation duration with the preset historical weight, and the sum of the real-time weight and the historical weight is 1.

[0035] The preset real-time weight and historical weight are set according to the experience and actual situation of airspace management, and their values are determined through multiple tests and adjustments. When calculating the resource occupation density value of each grid cell, the real-time occupation number is multiplied by the real-time weight, the historical occupation duration is multiplied by the historical weight, and then the two results are added to obtain the resource occupation density value of the grid cell.

[0036] Since the sum of the real-time weight and the historical weight is 1, the resource occupation density value can comprehensively reflect the real-time and historical occupation of the grid cell, and the size of the value is related to the busy degree of occupation.

[0037] Step S1227: According to the size of the resource occupation density value, a corresponding color or gray value is assigned to each grid cell, and the larger the resource occupation density value, the deeper the color or gray value, and a spatial resource occupation heat map is generated.

[0038] After obtaining the resource occupation density value of each grid cell, a mapping range of color or gray can be set. The minimum value to the maximum value of the resource occupation density value is divided into multiple intervals, and each interval corresponds to a set color or gray value. The larger the grid cell of the resource occupation density value, the deeper the corresponding color or gray value.

[0039] According to the mapping relationship, a corresponding color or gray value is assigned to each grid cell, and then all grid cells are combined to generate a spatial resource occupation heat map. The spatial resource occupation heat map can intuitively show the route occupation density of different regions in the airspace at different times.

[0040] Step S123: Input the start point coordinates, end point coordinates and spatial resource occupation heat map into the path skeleton generation layer of the generator unit, and generate a plurality of initial waypoint sequences from the start point coordinates to the end point coordinates based on the preset waypoint spacing constraint. Each of the initial waypoint sequences contains at least three consecutive waypoint geographic position coordinates.

[0041] After the path skeleton generation layer of the generator unit receives the start point coordinates, the end point coordinates and the spatial resource occupation heat map, the positions of the waypoints can be planned according to the preset waypoint spacing constraint. The preset waypoint spacing constraint refers to the distance range between adjacent two waypoints, which cannot be too large or too small.

[0042] The path skeleton generation layer will avoid the area with high occupation density according to the airspace occupation shown in the heat map, and plan a plurality of possible paths between the start point coordinates and the end point coordinates, and at least three consecutive waypoints are set on each candidate dynamic flight path. The geographic position coordinates of these waypoints are recorded to form an initial waypoint sequence.

[0043] Each initial waypoint sequence starts from the start point coordinates, passes through a plurality of waypoints, and finally reaches the end point coordinates, and the distance between the waypoints meets the preset waypoint spacing constraint.

[0044] Step S124: Determine the heading of each flight segment between adjacent waypoints in the initial waypoint sequence, and assign an estimated passing time to each flight segment based on the route occupancy density at the corresponding time in the airspace resource occupancy heat map, to generate an initial path containing waypoint coordinates, flight segment headings, and estimated passing times.

[0045] For each initial waypoint sequence, the flight segment heading between adjacent waypoints needs to be determined, which is obtained by calculating the azimuth angle between the two waypoint coordinates. Then, according to the route occupancy density at different times in the airspace resource occupancy heat map, a suitable estimated passing time is selected for each flight segment, so that the aircraft encounters less airspace occupancy pressure when passing through the flight segment at that time.

[0046] Combining waypoint coordinates, flight segment headings, and estimated passing times forms an initial path. Each initial path completely describes the flight route and time arrangement from the starting point to the ending point.

[0047] Step S1241: For each pair of adjacent waypoint coordinates in the initial waypoint sequence, determine the azimuth angle from the previous waypoint coordinate to the next waypoint coordinate as the heading of the flight segment.

[0048] When calculating the azimuth angle, the latitude and longitude coordinates of the two adjacent waypoints are used to determine the angle through the conventional geographic calculation method in the related art. With the previous waypoint coordinate as the reference, the angle of the next waypoint coordinate relative to it is calculated, which is the heading of the flight segment.

[0049] The value of the heading ranges from 0 degrees to 360 degrees, which can accurately indicate the direction of the aircraft flight. The heading of each flight segment is accurately calculated and recorded.

[0050] Step S1242: Determine the distance between the two adjacent waypoint coordinates as the length of the flight segment, and determine the flight time of the flight segment based on the preset aircraft cruising speed.

[0051] When calculating the distance between the two adjacent waypoint coordinates, the geographic distance calculation method is used to calculate the straight-line distance between the two points based on the latitude and longitude coordinates, which is the length of the flight segment. The preset aircraft cruising speed is a fixed value determined according to the type and performance of the aircraft.

[0052] The flight time is calculated by dividing the length of the flight segment by the preset aircraft cruising speed, and the result is the flight time required for the flight segment, in hours or minutes.

[0053] Step S1243: Extract the resource occupation density values of the airspace grid cells corresponding to the flight segment at different time points from the airspace resource occupation heat map, and select the time point with the minimum resource occupation density value as the predicted start time point of the flight segment.

[0054] The airspace grid cells corresponding to the flight segment refer to all the grid cells passed by the flight segment. From the airspace resource occupation heat map, the resource occupation density values of these grid cells at different time points are extracted to form a data list.

[0055] Then, the time point with the minimum resource occupation density value in the data list is found, which is determined as the predicted start time point of the flight segment. In this way, it can be ensured that the occupation of the airspace is relatively loose when the aircraft passes through the flight segment, reducing the possibility of conflict.

[0056] Step S1244: Determine the predicted end time point of the flight segment according to the predicted start time point and the flight time.

[0057] The calculation method of the predicted end time point can be to add the flight time to the predicted start time point. For example, if the predicted start time point is a certain time point and the flight time is 20 minutes, then the predicted end time point is the time point after adding 20 minutes to the time point.

[0058] Step S1245: Store the predicted start time point as the predicted passing time point of the flight segment in the predicted passing time point sequence of the initial path.

[0059] The predicted passing time point of each flight segment is its predicted start time point. These predicted passing time points are arranged in the order of the flight segments to form the predicted passing time point sequence of the initial path. The predicted passing time point sequence, together with the waypoint coordinates and the flight segment headings, constitutes the complete initial path information.

[0060] Step S125: Input the initial path into the conflict pre-avoidance layer of the generator unit, compare the spatiotemporal range of the flight segments of the initial path with the occupied air route spatiotemporal range in the airspace resource occupation heat map, adjust the waypoint coordinates and the corresponding predicted passing time points, and generate multiple candidate dynamic flight paths.

[0061] After receiving the initial path, the conflict pre-avoidance layer of the generator unit can perform conflict detection on each initial path. The spatiotemporal range of each flight segment in the initial path is compared with the occupied air route spatiotemporal range in the airspace resource occupation heat map. If overlap is found, it indicates that there is a potential conflict.

[0062] At this time, the conflict pre-avoidance layer adjusts the waypoint coordinates, changes the direction of the flight segment, or adjusts the estimated passing time to avoid the use time of the occupied route, so as to eliminate the conflict. After multiple adjustments, multiple candidate dynamic flight paths without conflict or with low conflict risk are generated.

[0063] Step S1251: Extract the spatiotemporal range of each flight segment from the initial path, the spatiotemporal range of each flight segment including the latitude and longitude range of the flight segment, the estimated starting time and the estimated ending time.

[0064] For each flight segment in the initial path, its latitude and longitude range is extracted, which is determined by the latitude and longitude coordinates of the starting and ending waypoints of the flight segment, covering all areas passed by the flight segment. At the same time, the estimated starting time and the estimated ending time of the flight segment are extracted, which define the time range of the aircraft flying on the flight segment.

[0065] Step S1252: Extract the spatiotemporal range of the occupied route from the airspace resource occupation heat map, the spatiotemporal range of the occupied route including the latitude and longitude range of the occupied route, the starting time and the ending time.

[0066] The airspace resource occupation heat map stores the relevant information of the occupied route, and the conflict pre-avoidance layer extracts the latitude and longitude range, the starting time and the ending time of each occupied route from it. These information are consistent with the spatiotemporal range information of the flight segment in the initial path extracted before, facilitating comparison operation.

[0067] Step S1253: Compare the spatiotemporal range of the flight segment of the initial path with the spatiotemporal range of the occupied route. If there is overlap, mark the flight segment as a conflict flight segment.

[0068] In comparison, first check whether the latitude and longitude range of the flight segment overlaps with the latitude and longitude range of the occupied route. If the latitude and longitude range does not overlap, the above two flight paths have no intersection in space and will not conflict. If the latitude and longitude range overlaps, check whether the estimated starting time and the estimated ending time of the flight segment overlap with the starting time and the ending time of the occupied route.

[0069] If the time range also overlaps, it means that the flight segment conflicts with the occupied route, and it is marked as a conflict flight segment. The marking method is to add a conflict identifier in the information of the flight segment.

[0070] Step S1254: For the conflict flight segment, move the ending waypoint coordinates in the direction where the resource occupation density value does not exceed the preset threshold by a preset distance to generate new waypoint coordinates.

[0071] The direction in which the resource occupation density value does not exceed the preset threshold refers to a direction in which the resource occupation density value is smaller in the airspace around the conflict flight segment. After the direction is determined, the end waypoint coordinate of the conflict flight segment is moved by a preset distance along the direction.

[0072] The preset distance is a fixed value determined according to the size of the airspace and flight safety requirements. The new waypoint coordinate generated after the movement is recorded as the adjusted waypoint coordinate.

[0073] Step S1255: The heading and length of the adjusted flight segment are re-determined, the heading being the azimuth angle of the adjusted adjacent waypoint coordinates, and the length being the distance between the adjusted adjacent waypoint coordinates.

[0074] The start waypoint coordinate of the adjusted flight segment remains unchanged, and the end waypoint coordinate becomes the newly generated waypoint coordinate. The azimuth angle between the two waypoint coordinates is re-calculated as the heading of the adjusted flight segment.

[0075] Meanwhile, the distance between the adjusted adjacent waypoint coordinates is calculated as the new flight segment length. These new heading and length information will replace the original information for subsequent flight time calculation and other operations.

[0076] Step S1256: The flight time is re-determined according to the length of the adjusted flight segment and the preset aircraft cruising speed.

[0077] The length of the adjusted flight segment has been determined, and the preset aircraft cruising speed remains unchanged. The length of the adjusted flight segment is divided by the preset aircraft cruising speed to obtain the flight time of the adjusted flight segment. The unit of the flight time matches the units of the aircraft cruising speed and the flight segment length, ensuring the reasonableness of the calculation result.

[0078] For example, if the unit of the length of the adjusted flight segment is kilometers and the unit of the preset aircraft cruising speed is kilometers per hour, the calculated flight time unit is hours.

[0079] Step S1257: The resource occupation density values of the airspace grid cells corresponding to the adjusted flight segment at different times are extracted from the airspace resource occupation heat map, and the time with the smallest resource occupation density value is selected as the new estimated start time.

[0080] The airspace grid cells corresponding to the adjusted flight segment may be different from the airspace grid cells corresponding to the original flight segment because the waypoint coordinates have changed. The path planning system can re-determine all the grid cells passed by the flight segment, and then extract the resource occupation density values of these grid cells at different times from the airspace resource occupation heat map to form a new data list.

[0081] In the new data list, find the time point with the minimum resource occupation density value, and determine it as the new predicted starting time. In this way, the adjusted flight segment can also avoid the busy period of airspace occupation in terms of time arrangement, further reducing the conflict risk.

[0082] Step S1258: Determine the new predicted end time, which is the new predicted starting time plus the newly determined flight time.

[0083] The new predicted starting time has been determined, and the newly determined flight time has been obtained. Adding these two values together gives the new predicted end time. The time format of the new predicted end time is consistent with that of the new predicted starting time, both of which are time stamps accurate to the second.

[0084] The new predicted end time will replace the original predicted end time and become the time range endpoint of the adjusted flight segment, which will be used for subsequent conflict detection.

[0085] Step S1259: Compare the space-time range of the adjusted flight segment with the space-time range of the occupied air route. If there is still overlap, repeat the steps of adjusting the waypoint coordinates and the predicted passing time until there is no overlap, and generate an adjusted initial path as a candidate dynamic flight path.

[0086] Compare the space-time range of the adjusted flight segment, which includes the latitude and longitude range, the new predicted starting time, and the new predicted end time, with the space-time range of all occupied air routes. The comparison method is the same as before, first checking whether the latitude and longitude range overlaps, and then checking whether the time range overlaps.

[0087] If it is found through comparison that there is still overlap, it means that the adjusted flight segment still has a conflict risk and needs to be adjusted again. Re-execute the operations of steps S1254 to S1258, that is, move the end waypoint coordinates again, recalculate the heading, length, flight time, predicted starting time, and predicted end time, and then compare again.

[0088] This process is repeated until there is no overlap between the space-time range of the adjusted flight segment and the space-time range of all occupied air routes. At this time, this adjusted initial path no longer has a conflict risk and is stored as a candidate dynamic flight path.

[0089] Step S130: Call the discriminator unit of the generative adversarial network, use the airspace regulation constraint rules as the benchmark, combine the path operation efficiency indicators, perform safety compliance verification and efficiency effectiveness evaluation on each candidate dynamic flight path, and output the comprehensive evaluation value of each candidate dynamic flight path.

[0090] The discriminator unit of the generative adversarial network can comprehensively evaluate each generated candidate dynamic flight path. The evaluation process takes airspace regulation constraint rules as the basic criteria to ensure that the path meets the requirements of aviation safety management, while combining path operation efficiency indicators to measure the operation efficiency of the path.

[0091] Through safety compliance verification and efficiency effectiveness evaluation, each candidate dynamic flight path is assigned a comprehensive evaluation value that can comprehensively reflect the safety and efficiency of the path.

[0092] Step S131: Extract the minimum aircraft separation requirement, no-fly area range, and flight altitude layer restriction condition from the airspace regulation constraint rules.

[0093] The airspace regulation constraint rules are a set of regulations that the discriminator unit can screen for key information related to safety compliance verification. The minimum aircraft separation requirement refers to the minimum distance that must be maintained between two aircraft in the horizontal and vertical directions to ensure flight safety.

[0094] The no-fly area range refers to an area that aircraft are not allowed to enter due to certain reasons, usually defined by latitude and longitude ranges. The flight altitude layer restriction condition specifies the altitude layer range that aircraft can fly in different airspace and different stages to prevent aircraft from colliding in altitude.

[0095] Step S132: For each candidate dynamic flight path, check whether all the waypoint geographic location coordinates it contains fall within the no-fly area range. If at least one waypoint coordinate falls within the no-fly area range, mark the candidate dynamic flight path as a violation path.

[0096] The discriminator unit can check all the waypoint geographic location coordinates in each candidate dynamic flight path one by one. For each waypoint coordinate, compare it with the latitude and longitude range of the no-fly area range to determine whether the waypoint coordinate is within the no-fly area.

[0097] If it is found that at least one waypoint coordinate in a candidate dynamic flight path falls within the no-fly area range, then this path violates the airspace regulation constraint rules and is marked as a violation path. The marking method is to add a violation identifier to the path's attribute information and note that the violation reason is that the waypoint enters the no-fly area.

[0098] Step S133: Compare the spatial distance between each flight segment of each candidate dynamic flight path at the predicted passing time and the real-time position information of existing aircraft in the airspace. If the spatial distance of at least one flight segment is less than the minimum aircraft separation requirement, mark the candidate dynamic flight path as a conflict path.

[0099] For each candidate dynamic flight path that is not marked as a violation path, the discriminator unit can check each flight segment thereof. At the predicted passing time of each flight segment, the real-time position information of all existing aircrafts in the airspace at that time is obtained, and then the spatial distance between the aircraft on the flight segment and the existing aircrafts is calculated.

[0100] When calculating the spatial distance, both the horizontal distance and the vertical distance are considered, and a three-dimensional spatial distance calculation method is adopted. The calculated spatial distance is compared with the minimum aircraft separation requirement. If the spatial distance of at least one flight segment is less than the minimum aircraft separation requirement, the path has a collision risk and is marked as a conflict path.

[0101] The marking method is to add a conflict identifier in the attribute information of the path and specify the flight segment where the conflict occurs and the corresponding existing aircraft.

[0102] Step S134: Extract the total flight time requirement and the fuel consumption estimation coefficient from the path operation efficiency indicators.

[0103] The path operation efficiency indicators include various parameters for measuring the efficiency of the flight path. The discriminator unit can extract information related to efficiency effectiveness evaluation from the indicators. The total flight time requirement refers to the maximum flight time allowed to complete the flight task. If the time exceeds the requirement, the path efficiency is considered to be low.

[0104] The fuel consumption estimation coefficient is a coefficient related to the type of aircraft, flight speed, flight altitude, and other factors, which is used to estimate the fuel consumption based on the flight time.

[0105] Step S135: Determine the total flight time of each candidate dynamic flight path that is not marked as a violation path or a conflict path. The total flight time is the sum of the flight times of all flight segments, and the flight time of each flight segment is determined by the preset aircraft cruising speed.

[0106] For the candidate dynamic flight path that is not marked as a violation path or a conflict path, the discriminator unit can summarize the flight times of all flight segments. The flight time of each flight segment has been determined in the previous step and is calculated based on the length of the flight segment and the preset aircraft cruising speed.

[0107] The sum of the flight times of all flight segments is the total flight time of the candidate dynamic flight path. The unit of the total flight time is consistent with the unit of the flight time of each flight segment, which is usually hours or minutes.

[0108] Step S136: Determine the fuel consumption estimation of each candidate dynamic flight path based on the fuel consumption estimation coefficient and the total flight time.

[0109] The fuel consumption estimation is calculated by multiplying the total flight time by a fuel consumption estimation coefficient. The unit of the fuel consumption estimation coefficient is fuel consumption per unit time, so the result of the multiplication of the total flight time and the fuel consumption estimation coefficient is the fuel consumption estimation of the candidate dynamic flight path, in tons or kilograms.

[0110] Step S137: Set a control priority weight according to the airspace control priority requirement, and set a task efficiency weight according to the flight task efficiency requirement. The sum of the control priority weight and the task efficiency weight is 1.

[0111] The airspace control priority requirement reflects the importance of airspace safety and order in aviation management, and the task efficiency requirement reflects the importance of timeliness and economy in completing the flight task. According to the two requirements, the control priority weight and the task efficiency weight are set respectively.

[0112] For example, if the airspace control priority requirement is high, the control priority weight can be set to be large, and the task efficiency weight is correspondingly small; on the contrary, if the flight task efficiency requirement is high, the task efficiency weight can be set to be large. However, no matter how it is set, the sum of the control priority weight and the task efficiency weight is always 1, to ensure that the calculation of the comprehensive evaluation value is on a unified scale.

[0113] Step S138: For the violation path or the conflict path, the safety compliance score is set to 0, and for the path without violation and conflict, the safety compliance score is determined based on the correspondence between the average value of the separation margin of each flight segment and the preset maximum allowed separation margin.

[0114] The violation path and the conflict path have a high safety risk because they violate the airspace control constraint rules, so the safety compliance score is directly set to 0. For the path without violation and conflict, the degree of safety compliance needs to be further evaluated.

[0115] The separation margin of each flight segment refers to the difference between the spatial distance of the flight segment at the predicted passing time and the existing aircraft and the minimum aircraft separation requirement. The average value of the separation margin of all flight segments is calculated to obtain the average value of the separation margin.

[0116] The preset maximum allowed separation margin is a pre-set value representing the maximum separation margin that can be maintained between aircrafts under the premise of ensuring safety. According to the proportional relationship between the average value of the separation margin and the preset maximum allowed separation margin, the safety compliance score is determined. For example, if the average value of the separation margin is equal to the preset maximum allowed separation margin, the safety compliance score is set to full score; if the average value of the separation margin is half of the preset maximum allowed separation margin, the safety compliance score is set to half of the full score.

[0117] Step S139: Based on the correspondence between the total flight time and the preset maximum allowed flight time, the correspondence between the fuel consumption estimate and the preset maximum allowed fuel consumption, combined with the preset time weight and fuel weight, determine the efficiency effectiveness score.

[0118] The preset maximum allowed flight time is the longest flight time specified in the total flight time requirement, and the preset maximum allowed fuel consumption is the maximum fuel consumption limit set according to the fuel carrying capacity and economic requirements of the flight mission.

[0119] First, according to the proportional relationship between the total flight time and the preset maximum allowed flight time, the time efficiency score is determined. For example, if the total flight time is equal to half of the preset maximum allowed flight time, the time efficiency score is set to half of the full score; if the total flight time is equal to the preset maximum allowed flight time, the time efficiency score is set to 0.

[0120] Second, according to the proportional relationship between the fuel consumption estimate and the preset maximum allowed fuel consumption, the fuel efficiency score is determined. The calculation method is similar to the time efficiency score, and the smaller the fuel consumption estimate, the higher the fuel efficiency score.

[0121] The preset time weight and fuel weight are set according to the importance of time and fuel in the flight mission, and the sum of the two is 1. Multiply the time efficiency score by the time weight, multiply the fuel efficiency score by the fuel weight, and then add the two results to get the efficiency effectiveness score.

[0122] Step S1310: Multiply the safety compliance score by the regulation priority weight, and multiply the efficiency effectiveness score by the mission efficiency weight, then add them together to get the comprehensive evaluation value of each candidate dynamic flight path.

[0123] The safety compliance score reflects the safety of the path, and the efficiency effectiveness score reflects the efficiency of the path. By weighting the two scores through the regulation priority weight and the mission efficiency weight, the importance of safety and efficiency in the flight mission can be considered comprehensively.

[0124] Multiply the safety compliance score by the regulation priority weight to get the contribution value of safety in the comprehensive evaluation; multiply the efficiency effectiveness score by the mission efficiency weight to get the contribution value of efficiency in the comprehensive evaluation. Add the two contribution values to get the comprehensive evaluation value of the candidate dynamic flight path. The range of the comprehensive evaluation value is between 0 and the full score, and the higher the score, the better the path.

[0125] Step S140: the candidate dynamic flight path with the highest comprehensive evaluation value is input into the parameter adjustment module of the generator unit as a feedback path, to adjust the waypoint distribution probability and flight segment time assignment weight used by the generator unit in path generation.

[0126] Among all candidate dynamic flight paths, the path with the highest comprehensive evaluation value is considered as the current optimal path, which is input into the parameter adjustment module of the generator unit as a feedback path. The parameter adjustment module adjusts the parameters used by the generator unit in path generation according to the characteristics of the feedback path, to improve the quality of the subsequent generated paths.

[0127] The waypoint distribution probability determines the possibility of each location appearing as a waypoint when the generator unit plans waypoints, and the flight segment time assignment weight affects the preference for different times when the generator unit assigns the estimated passing time to a flight segment. By adjusting the above parameters, the generator unit is more inclined to generate high-quality paths similar to the feedback path when generating paths next time.

[0128] Step S141: extracting a waypoint geographical position coordinate sequence, a flight segment heading sequence, and an estimated passing time sequence from the candidate dynamic flight path with the highest comprehensive evaluation value.

[0129] The parameter adjustment module can analyze the candidate dynamic flight path with the highest comprehensive evaluation value and extract key information therefrom. The waypoint geographical position coordinate sequence is a sequence composed of the longitude and latitude coordinates of all waypoints in the path in chronological order, reflecting the spatial trend of the path.

[0130] The flight segment heading sequence is a sequence composed of the headings of each flight segment in chronological order, representing the flight direction of the aircraft in each flight segment. The estimated passing time sequence is a sequence composed of the estimated passing times of each flight segment in chronological order, reflecting the time arrangement of the candidate dynamic flight path.

[0131] Step S142: comparing the waypoint geographical position coordinate sequence with the waypoint coordinates output by the current path skeleton generation layer of the generator unit to obtain a coordinate deviation sequence.

[0132] The waypoint coordinates output by the current path skeleton generation layer of the generator unit refer to the waypoint coordinates generated by the path skeleton generation layer when generating the feedback path. The waypoint geographical position coordinate sequence of the feedback path is compared with these waypoint coordinates one by one to calculate the deviation between each corresponding waypoint.

[0133] The calculation of coordinate deviation is to calculate the longitude coordinate deviation and the latitude coordinate deviation respectively, and then to compose a coordinate deviation sequence according to the order of waypoints. The coordinate deviation sequence reflects the difference between the waypoint coordinates currently generated by the generator unit and the optimal waypoint coordinates.

[0134] Step S143: Adjust the waypoint distribution probability of the path skeleton generation layer of the generator unit according to the coordinate deviation sequence.

[0135] Each deviation value in the coordinate deviation sequence reflects the degree of deviation of the corresponding waypoint position. The parameter adjustment module can adjust the waypoint distribution probability of each position in the path skeleton generation layer according to the above-mentioned deviation value. For positions with smaller deviations, the waypoint distribution probability is increased, so that the generator unit is more likely to plan waypoints at these positions when generating the path next time; for positions with larger deviations, the waypoint distribution probability is reduced, reducing the possibility of waypoints appearing at these positions.

[0136] wherein the adjustment amplitude is related to the size of the coordinate deviation, and the larger the deviation, the larger the adjustment amplitude, so as to quickly correct the deviation of the generator unit in the waypoint planning.

[0137] Step S144: Compare the flight segment expected passing time sequence with the time points output by the current conflict pre-avoidance layer of the generator unit, and obtain a time deviation sequence.

[0138] The time points output by the current conflict pre-avoidance layer of the generator unit refer to the expected passing time points allocated by the conflict pre-avoidance layer for each flight segment when generating the feedback path. The expected passing time sequence of the flight segment of the feedback path is compared with these time points one by one, and the deviation between each corresponding time point is calculated to obtain a time deviation sequence.

[0139] The calculation of the time deviation is to subtract the time output by the conflict pre-avoidance layer from the expected passing time of the feedback path, and the difference is the time deviation. The time deviation sequence reflects the difference between the expected passing time allocated by the generator unit and the optimal expected passing time.

[0140] Step S145: Adjust the flight segment time allocation weight of the conflict pre-avoidance layer of the generator unit according to the time deviation sequence.

[0141] Each deviation value in the time deviation sequence reflects the degree of deviation of the corresponding flight segment expected passing time. The parameter adjustment module can adjust the flight segment time allocation weight in the conflict pre-avoidance layer according to these deviation values. For time points with smaller deviations, the allocation weight is increased, so that the generator unit is more likely to select these time points when allocating the expected passing time for the flight segment next time; for time points with larger deviations, the allocation weight is reduced, reducing the selection of these time points.

[0142] The adjustment method is to adjust the allocation weight of the corresponding time point in proportion to the size of the time deviation, so as to ensure that the sum of the weights remains unchanged.

[0143] Step S146: Store the adjusted waypoint distribution probability and flight segment time distribution weight to the parameter library of the generator unit, overwrite the original parameters, and use for the generation of the next round of candidate dynamic flight paths.

[0144] After the parameter adjustment module completes the adjustment of the waypoint distribution probability and the flight segment time distribution weight, it can store these adjusted parameters into the parameter library of the generator unit. The parameter library is where the generator unit stores various generation parameters, and new parameters will overwrite the original parameters to become the parameters used by the generator unit when generating paths next time.

[0145] In this way, when the generator unit receives input information to generate a candidate dynamic flight path next time, it can generate according to the adjusted parameters, thereby improving the comprehensive evaluation value of the generated path.

[0146] Step S150: Regenerate the candidate dynamic flight path based on the adjusted generator unit, repeat the evaluation of the discriminator unit and the parameter adjustment of the generator unit until the comprehensive evaluation value of the generated candidate dynamic flight path reaches the preset path optimization threshold, and send the candidate dynamic flight path as the final dynamic flight path to the air traffic control command system.

[0147] After the parameter adjustment of the generator unit is completed, the flight mission instruction and the airspace real-time operation data set can be received again, and a plurality of new candidate dynamic flight paths can be generated according to the new parameters. The discriminator unit can evaluate the above new candidate dynamic flight paths to obtain their comprehensive evaluation values.

[0148] Then, the path with the highest comprehensive evaluation value is taken as a new feedback path, and the parameters of the generator unit are adjusted again. This process will be repeated until a path with a comprehensive evaluation value reaching the preset path optimization threshold appears in the generated candidate dynamic flight paths. At this time, this path is considered to be the optimal final dynamic flight path, which is sent to the air traffic control command system for guiding the flight of the aircraft.

[0149] Step S151: Call the adjusted generator unit, input the flight mission instruction and the airspace real-time operation data set, and generate a plurality of new candidate dynamic flight paths.

[0150] After the adjusted generator unit receives the flight mission instruction and the airspace real-time operation data set, it can generate paths according to the new waypoint distribution probability and the flight segment time distribution weight. The generation process is similar to the previous one, including extracting the starting point coordinates and the ending point coordinates, constructing the airspace resource occupation heat map, generating the initial waypoint sequence, determining the flight segment heading and the expected passing time, and performing conflict pre-avoidance adjustment.

[0151] When generating the initial waypoint sequence, the generated waypoint positions will be more inclined to the waypoint distribution characteristics in the feedback path with high comprehensive evaluation values, due to the adjusted waypoint distribution probability. For example, if the waypoint distribution in a certain area in the feedback path is relatively dense, the adjusted waypoint distribution probability will make the newly generated initial waypoint sequence more likely to generate more waypoints in that area to adapt to the airspace environment and path efficiency requirements.

[0152] When determining the flight segment predicted passing time, the adjustment of the flight segment time allocation weight will affect the consideration of resource occupation density values at different times in the airspace resource occupation heat map. The change of the weight may make the generator unit focus more on the resource occupation of certain time periods when selecting the predicted starting time, thereby generating a time allocation that meets the efficiency and safety requirements.

[0153] After the above process, the adjusted generator unit can output multiple new candidate dynamic flight paths, which incorporate the optimization information of the previous round of feedback paths in terms of waypoint distribution and time allocation.

[0154] Step S152: Perform safety compliance verification and efficiency effectiveness evaluation on each new candidate dynamic flight path, and output the comprehensive evaluation value of each new candidate dynamic flight path.

[0155] For each newly generated candidate dynamic flight path, the discriminator unit can process it according to the same safety compliance verification and efficiency effectiveness evaluation standards as before.

[0156] In terms of safety compliance verification, each candidate dynamic flight path is checked one by one to determine whether the waypoints fall into the no-fly area, and whether the spatial distance between each flight segment at the predicted passing time and the existing aircraft meets the minimum separation requirement. For paths that violate the no-fly area regulations or have insufficient separation, they are marked as illegal paths or conflict paths.

[0157] In terms of efficiency effectiveness evaluation, the total flight time and fuel consumption estimate of each compliant path are calculated. The total flight time is the sum of the flight times of each flight segment, and the fuel consumption estimate is determined by combining the fuel consumption estimate coefficient and the total flight time. Then, according to the safety compliance score and the efficiency effectiveness score, as well as the set regulatory priority weight and task efficiency weight, the comprehensive evaluation value of each new candidate dynamic flight path is calculated.

[0158] Step S153: Extract the comprehensive evaluation values of all new candidate dynamic flight paths, and check whether there is a candidate dynamic flight path whose comprehensive evaluation value reaches the preset path optimization threshold.

[0159] The comprehensive evaluation values of all new candidate dynamic flight paths are collected to form an evaluation value list. The preset path optimization threshold is a preset value representing the minimum acceptable comprehensive evaluation level for path optimization.

[0160] Each comprehensive evaluation value in the evaluation value list is compared with the preset path optimization threshold one by one to determine whether at least one comprehensive evaluation value reaches or exceeds the preset path optimization threshold. This step aims to determine whether the currently generated candidate dynamic flight path has met the optimization requirement.

[0161] Step S154: If so, the candidate dynamic flight path with the highest comprehensive evaluation value that reaches the path optimization threshold is selected as the final dynamic flight path.

[0162] When the check finds that there are candidate dynamic flight paths with comprehensive evaluation values reaching the preset path optimization threshold, the candidate dynamic flight path with the highest comprehensive evaluation value is selected from these qualified candidate dynamic flight paths. This candidate dynamic flight path performs best in terms of safety compliance and efficiency effectiveness, and can best meet the requirements of the current flight task. Therefore, this candidate dynamic flight path is determined as the final dynamic flight path and is prepared to be sent to the air traffic control system as the execution route of the flight task.

[0163] Step S155: If not, the candidate dynamic flight path with the highest comprehensive evaluation value in the new candidate dynamic flight paths is selected as the new feedback path, and the waypoint distribution probability and flight segment time allocation weight of the generator unit are adjusted.

[0164] If the comprehensive evaluation values of all new candidate dynamic flight paths do not reach the preset path optimization threshold, it means that there is still room for optimization for the currently generated path. At this time, the candidate dynamic flight path with the highest comprehensive evaluation value is selected as the new feedback path and is input into the parameter adjustment module of the generator unit.

[0165] The parameter adjustment module can compare the waypoint coordinate sequence and the expected passing time sequence of the new feedback path with the corresponding sequences currently output by the generator unit, calculate the coordinate deviation and the time deviation, and then adjust the waypoint distribution probability and the flight segment time allocation weight. This process is consistent with the parameter adjustment method in step S140, and aims to make the generation parameters of the generator unit further approach the optimization direction.

[0166] Step S156: Repeat the above steps until the comprehensive evaluation value of the generated candidate dynamic flight path reaches the preset path optimization threshold.

[0167] After adjusting the parameters of the generator unit, return to step S151 to regenerate a new candidate dynamic flight path and perform evaluation and checking again. This cycle is repeated to continuously optimize the parameters of the generator unit and the generated path until a path with a comprehensive evaluation value reaching a preset path optimization threshold appears in the generated candidate dynamic flight path, and then step S154 is performed to determine the final dynamic flight path.

[0168] Step S210: Obtain a historical flight task data set and a corresponding airspace historical operation data set, the historical flight task data set containing a historical take-off airport identifier, a historical landing airport identifier, a historical flight task priority, and a corresponding historical optimal flight path, and the airspace historical operation data set containing real-time position information of existing aircraft in the historical airspace, a time-space range of a historical occupied air route, and a historical airspace control constraint rule.

[0169] Before training the generative adversarial network model, a large amount of historical data needs to be collected as a training basis. The historical flight task data set covers relevant information of various types of flight tasks completed in the past, among which the historical take-off airport identifier and the historical landing airport identifier are used to determine the starting point and the ending point of the flight, the historical flight task priority reflects the importance of the task at that time, and the corresponding historical optimal flight path is a flight route that has been actually verified and performed well.

[0170] The airspace historical operation data set records the airspace operation situation during the same period as the historical flight task, including the position information of existing aircraft in the airspace at that time, the time-space range of the occupied air route, and the applicable airspace control constraint rule.

[0171] Step S220: Divide the historical flight task data set and the airspace historical operation data set into a training set and a validation set, and the proportion of the training set data amount to the total data amount is a preset training set proportion.

[0172] In order to ensure the effect and generalization ability of model training, the collected historical data needs to be divided into a training set and a validation set. The preset training set proportion can be set according to the total data amount and the model training requirement, for example, it can be set to 80% of the total data amount.

[0173] The division process adopts a random sampling method to ensure that the data distribution in the training set and the validation set is representative and can reflect the path planning situation under different airspace conditions and different flight task types. The training set is used for parameter learning of the model, and the validation set is used for evaluating the effect of model training.

[0174] Step S230: initialize the parameters of the generator unit and the parameters of the discriminator unit, wherein the parameters of the generator unit include the initial value of the waypoint distribution probability of the path skeleton generation layer and the initial value of the flight segment time allocation weight of the conflict pre-avoidance layer, and the parameters of the discriminator unit include the initial value of the regulation priority weight of the safety compliance verification and the initial value of the task efficiency weight of the efficiency effectiveness evaluation.

[0175] When initializing the parameters of the generator unit, the initial value of the waypoint distribution probability of the path skeleton generation layer can be set to be uniformly distributed in the airspace range from the starting point to the ending point, that is, each possible waypoint position has the same initial probability. The initial value of the flight segment time allocation weight of the conflict pre-avoidance layer can be set to assign the same weight to the resource occupation density values at different times, so that the initial time selection is more random.

[0176] When initializing the parameters of the discriminator unit, the initial value of the regulation priority weight of the safety compliance verification and the initial value of the task efficiency weight of the efficiency effectiveness evaluation can be set to be equal, for example, each accounting for fifty percent, so that the consideration of safety and efficiency is equally important when the initial evaluation is performed. These initial values provide a starting point for the training of the model.

[0177] Step S240: for each piece of historical flight task data and corresponding airspace historical operation data in the training set: extract the historical takeoff airport coordinates and the historical landing airport coordinates from the historical flight task data, and extract the historical airspace resource occupation heat map from the airspace historical operation data.

[0178] For each piece of data in the training set, data extraction is performed first. According to the historical takeoff airport identifier and the historical landing airport identifier, the corresponding geographical position coordinates are queried to obtain the historical takeoff airport coordinates and the historical landing airport coordinates.

[0179] At the same time, the real-time position information of existing aircraft and the occupied air route space-time range at that time are extracted from the airspace historical operation data, and the historical airspace resource occupation heat map is generated in the same way as the construction of the airspace resource occupation heat map, which is used to simulate the airspace operation environment at that time.

[0180] Step S241: input the historical takeoff airport coordinates, the historical landing airport coordinates, and the historical airspace resource occupation heat map into the generator unit to generate a plurality of historical candidate flight paths.

[0181] After the generator unit receives the historical takeoff airport coordinates, the historical landing airport coordinates, and the historical airspace resource occupation heat map, it generates a plurality of historical candidate flight paths in the same process as generating the current candidate dynamic flight path. These paths are the path planning attempts of the model based on the initial parameters for historical tasks.

[0182] Each historical candidate flight path contains information such as continuous waypoint coordinates, flight segment headings, and estimated passing times.

[0183] Step S242: input the historical candidate flight path and the corresponding historical optimal flight path into the discriminator unit to determine the comprehensive evaluation value of the historical candidate flight path, and set the comprehensive evaluation value of the historical optimal flight path as a preset maximum value.

[0184] The discriminator unit simultaneously receives the historical candidate flight path and the corresponding historical optimal flight path. For the historical candidate flight path, its comprehensive evaluation value is calculated according to the process of safety compliance verification and efficiency effectiveness evaluation.

[0185] The historical optimal flight path is the known optimal solution, and its comprehensive evaluation value is set as a preset maximum value, which is higher than the comprehensive evaluation value of all possible generated historical candidate flight paths, and is used as the target reference for model training.

[0186] Step S243: determine the loss value of the generator unit and the loss value of the discriminator unit, the loss value of the generator unit being the difference between the comprehensive evaluation value of the historical candidate flight path and the comprehensive evaluation value of the historical optimal flight path, and the loss value of the discriminator unit being the difference between the comprehensive evaluation value output by the discriminator unit and the comprehensive evaluation value of the historical optimal flight path.

[0187] The loss value of the generator unit reflects the gap between the historical candidate flight path generated by it and the historical optimal flight path, and the larger the difference, the less ideal the generated path, and the higher the loss value.

[0188] The loss value of the discriminator unit reflects the gap between the evaluation result of the historical candidate flight path by it and the comprehensive evaluation value of the historical optimal flight path, and the larger the difference, the lower the evaluation accuracy of the discriminator, and the higher the loss value.

[0189] Step S244: adjust the parameters of the generator unit and the discriminator unit using the back propagation algorithm, and after minimizing the loss values of the generator unit and the discriminator unit, repeat the training steps in the training set until the loss value of the training set reaches a preset training termination threshold.

[0190] The back propagation algorithm will propagate errors from the output layer to the input layer according to the loss values of the generator unit and the discriminator unit, and adjust the parameters of each layer of the model. For the generator unit, adjust the waypoint distribution probability of the path skeleton generation layer and the flight segment time allocation weight of the conflict pre-avoidance layer; for the discriminator unit, adjust the control priority weight and the task efficiency weight and other parameters.

[0191] Through continuous iteration adjustment, the path generated by the generator unit is closer and closer to the historical optimal flight path, and the evaluation of the discriminator unit is more and more accurate, until the loss value of the training set is reduced below the preset training termination threshold, indicating that the performance of the model on the training set reaches the expectation.

[0192] Step S250: verifying the trained generator unit and discriminator unit using the verification set, generating candidate flight paths of the verification set, and determining the deviation of the comprehensive evaluation value of the candidate flight paths from the comprehensive evaluation value of the historical optimal flight path.

[0193] After training, the historical flight task data and airspace historical operation data of the verification set are input into the trained generator unit and discriminator unit. The generator unit generates candidate flight paths of the verification set, and the discriminator unit evaluates these paths to obtain their comprehensive evaluation values.

[0194] The deviation between these comprehensive evaluation values and the corresponding historical optimal flight path comprehensive evaluation value (preset maximum value) is calculated. The smaller the deviation, the better the generalization ability of the model, which can better handle data not appearing in the training set.

[0195] Step S260: If the deviation exceeds the preset deviation range, adjust the training set proportion or parameter initial value and re-execute the training step.

[0196] When the deviation of the verification set exceeds the preset deviation range, it indicates that the generative adversarial network model may have overfitting or underfitting problems, and the generalization ability is insufficient. At this time, the training set proportion needs to be adjusted, for example, the data amount proportion of the training set is increased, so that the generative adversarial network model can learn more data features.

[0197] Alternatively, the parameter initial value of the generator unit and the discriminator unit is reset, the initial state of the generative adversarial network model is changed, and then the training step is re-executed until the deviation of the verification set is within the preset range.

[0198] Step S270: If the deviation does not exceed the preset deviation range, save the parameters of the generator unit and the discriminator unit as a pre-trained generative adversarial network model.

[0199] When the deviation of the verification set is within the preset deviation range, it indicates that the training effect of the generative adversarial network model is good, and the generalization ability can meet the requirements. At this time, the current parameters of the generator unit and the discriminator unit are saved to form a pre-trained generative adversarial network model.

[0200] The generative adversarial network model can be used for actual dynamic path planning tasks, and after receiving new flight task instructions and real-time airspace operation data sets, it can quickly generate flight paths that meet the requirements.

[0201] Figure 2 An exemplary hardware and software components of the dynamic path planning system 100 based on the generative adversarial network model that can implement the idea of the present application are shown in the schematic diagram. For example, the processor 120 can be used in the dynamic path planning system 100 based on the generative adversarial network model and used to perform the functions in the present application.

[0202] The dynamic path planning system 100 based on the generative adversarial network model can be a general server or a special-purpose server, both of which can be used to implement the dynamic path planning method based on the generative adversarial network model of the present application. Although only one server is shown in the present application, for the sake of convenience, the functions described in the present application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0203] For example, the dynamic path planning system 100 based on the generative adversarial network model can include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and different forms of storage media 140, such as a disk, a ROM, or a RAM, or any combination thereof. Exemplarily, the dynamic path planning system 100 based on the generative adversarial network model can also include program instructions stored in a ROM, a RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present application can be implemented according to these program instructions. The dynamic path planning system 100 based on the generative adversarial network model also includes an I / O interface 150 between the computer and other input / output devices.

[0204] For the sake of illustration, only one processor is described in the dynamic path planning system 100 based on the generative adversarial network model. However, it should be noted that the dynamic path planning system 100 based on the generative adversarial network model in the present application can also include multiple processors, so the steps performed by one processor described in the present application can also be jointly performed or individually performed by multiple processors. For example, if the processor of the dynamic path planning system 100 based on the generative adversarial network model performs steps A and B, it should be understood that steps A and B can also be jointly performed by two different processors or individually performed in one processor. For example, a first processor performs step A, a second processor performs step B, or a first processor and a second processor jointly perform steps A and B.

[0205] In addition, the present application also provides a readable storage medium, wherein computer executable instructions are pre-stored in the readable storage medium, and when the processor executes the computer executable instructions, the dynamic path planning method based on the generative adversarial network model is implemented.

[0206] It should be noted that the foregoing description of embodiments of the application has been presented for the purposes of illustration and description. It is not intended to be exhaustive or to limit the application to the precise form described, and many modifications, variations, and alternatives are possible.

Claims

1. A dynamic path planning method based on a generative adversarial network model, characterized in that, The method comprises: receiving a current flight task instruction and a corresponding airspace real-time operation data set sent by an air traffic control system, wherein the flight task instruction comprises a take-off airport identifier and a landing airport identifier, and the airspace real-time operation data set comprises real-time position information of existing aircraft in the airspace, a time-space range of occupied air routes, and airspace control constraint rules; inputting the flight task instruction and the airspace real-time operation data set into a generator unit of a pre-trained generative adversarial network to generate a plurality of candidate dynamic flight paths covering the entire flight, wherein each candidate dynamic flight path comprises continuous waypoint geographic position coordinates, flight segment headings between waypoints, and estimated passing times of corresponding flight segments; calling a discriminator unit of the generative adversarial network to perform safety compliance checking and efficiency effectiveness evaluation on each candidate dynamic flight path based on the airspace control constraint rules and in combination with a path operation efficiency index, and outputting a comprehensive evaluation value of each candidate dynamic flight path; inputting a candidate dynamic flight path with the highest comprehensive evaluation value as a feedback path into a parameter adjustment module of the generator unit to adjust waypoint distribution probability and flight segment time allocation weight in path generation of the generator unit; re-generating a candidate dynamic flight path based on the adjusted generator unit, and repeatedly performing evaluation of the discriminator unit and parameter adjustment of the generator unit until the comprehensive evaluation value of the generated candidate dynamic flight path reaches a preset path optimization threshold, and then sending the candidate dynamic flight path as a final dynamic flight path to an air traffic control command system.

2. The dynamic path planning method based on the generative adversarial network model according to claim 1, characterized in that, The method comprises: extracting take-off airport geographic position coordinates corresponding to the take-off airport identifier and landing airport geographic position coordinates corresponding to the landing airport identifier from the flight task instruction as a starting point coordinate and an end point coordinate of the flight path; extracting real-time position information of existing aircraft in the airspace and a time-space range of occupied air routes from the airspace real-time operation data set to construct an airspace resource occupation heat map, wherein the airspace resource occupation heat map is used to represent air route occupation density of different regions in the airspace at different times; inputting the starting point coordinate, the end point coordinate, and the airspace resource occupation heat map into a path skeleton generation layer of the generator unit to generate a plurality of initial waypoint sequences from the starting point coordinate to the end point coordinate based on a preset waypoint distance constraint, wherein each initial waypoint sequence comprises at least three continuous waypoint geographic position coordinates; determining headings of adjacent waypoint coordinates in each initial waypoint sequence to obtain flight segment headings between waypoints, and allocating an estimated passing time to each flight segment in combination with air route occupation density at a corresponding time in the airspace resource occupation heat map to generate an initial path comprising waypoint coordinates, flight segment headings, and estimated passing times; The initial path is input into a conflict pre-avoidance layer of the generator unit, a flight segment space-time range of the initial path is compared with an occupied air route space-time range in the airspace resource occupation heat map, waypoint coordinates and corresponding expected passing time are adjusted, and a plurality of candidate dynamic flight paths are generated.

3. The dynamic path planning method based on the generative adversarial network model according to claim 2, characterized in that, The airspace resource occupation heat map is constructed by extracting real-time position information of existing aircrafts in the airspace and space-time ranges of occupied air routes from the airspace real-time operation data set, and includes: Real-time position information of each existing aircraft is extracted from the airspace real-time operation data set, and the real-time position information includes longitude and latitude coordinates and a flight height layer; Space-time ranges of each occupied air route are extracted, and the space-time range of each occupied air route includes a starting time of the air route, an ending time of the air route, and a longitude and latitude range covered by the air route; The airspace is divided into a plurality of equal-area grid cells, and the longitude and latitude ranges of each grid cell are the same; For each grid cell, the number of times that the real-time position information of the existing aircraft falls into the grid cell within a preset time window is counted as a real-time occupation number of the grid cell; The length of the time period in which the space-time range of the occupied air route covers the grid cell within the preset time window is counted as a historical occupation length of the grid cell; The resource occupation density value of each grid cell is obtained by combining the real-time occupation number with a preset real-time weight and the historical occupation length with a preset historical weight, and the sum of the real-time weight and the historical weight is 1; According to the size of the resource occupation density value, each grid cell is assigned a corresponding color or gray value, and the larger the resource occupation density value, the deeper the color or gray value, and the airspace resource occupation heat map is generated.

4. The dynamic path planning method based on the generative adversarial network model according to claim 2, characterized in that, The heading of each adjacent waypoint coordinate in the initial waypoint sequence is determined to obtain the flight segment heading between the waypoints, and the expected passing time of each flight segment is assigned in combination with the air route occupation density at the corresponding time in the airspace resource occupation heat map, and includes: For each of the two adjacent waypoint coordinates in the initial waypoint sequence, the azimuth angle from the previous waypoint coordinate to the next waypoint coordinate is determined as the heading of the flight segment; The distance between the two adjacent waypoint coordinates is determined as the length of the flight segment, and the flight time of the flight segment is determined according to the preset aircraft cruising speed; The resource occupation density values of the airspace grid cells corresponding to the flight segment at different times are extracted from the airspace resource occupation heat map, and the time with the smallest resource occupation density value is selected as the expected starting time of the flight segment; The expected ending time of the flight segment is determined according to the expected starting time and the flight time; The expected starting time is taken as the expected passing time of the flight segment, and is stored in the expected passing time sequence of the initial path.

5. The dynamic path planning method based on the generative adversarial network model according to claim 2, characterized in that, The initial path is input into a conflict pre-avoidance layer of the generator unit, a flight segment space-time range of the initial path is compared with an occupied air route space-time range in the airspace resource occupation heat map, waypoint coordinates and corresponding expected passing time are adjusted, and a plurality of candidate dynamic flight paths are generated. extracting a time-space range of each flight segment from the initial path, the time-space range of each flight segment including a latitude-longitude range of the flight segment, a predicted start time and a predicted end time; extracting a time-space range of the occupied air route from the air resource occupation heat map, the time-space range of the occupied air route including a latitude-longitude range of the occupied air route, a start time and an end time; comparing the time-space range of the flight segment of the initial path with the time-space range of the occupied air route, and marking the flight segment as a conflict flight segment if there is an overlap; for the conflict flight segment, moving a terminal waypoint coordinate of the conflict flight segment by a preset distance in a direction in which a resource occupation density value does not exceed a preset threshold to generate a new waypoint coordinate; redetermining a heading and a length of the adjusted flight segment, the heading being a bearing angle of the adjusted adjacent waypoint coordinates, and the length being a distance between the adjusted adjacent waypoint coordinates; redetermining a flight time according to the length of the adjusted flight segment and a preset aircraft cruising speed; extracting resource occupation density values of air space grid cells corresponding to the adjusted flight segment at different times from the air resource occupation heat map, and selecting a time at which the resource occupation density value is the smallest as a new predicted start time; determining a new predicted end time, and implementing the new predicted end time as the new predicted start time plus the redetermined flight time; comparing the time-space range of the adjusted flight segment with the time-space range of the occupied air route, and repeating the steps of adjusting the waypoint coordinate and the predicted passing time until there is no overlap, to generate an adjusted initial path as a candidate dynamic flight path.

6. The dynamic path planning method based on the generative adversarial network model according to claim 1, wherein, The calling of the discriminator unit of the generative adversarial network is based on the air space regulation constraint rule and combined with a path operation efficiency index to perform safety compliance verification and efficiency effectiveness evaluation on each candidate dynamic flight path, and an integrated evaluation value of each candidate dynamic flight path is output, including: extracting a minimum aircraft separation requirement, a no-fly area range and a flight altitude layer limitation condition from the air space regulation constraint rule; checking whether all waypoint geographical position coordinates included in each candidate dynamic flight path fall within the no-fly area range, and marking the candidate dynamic flight path as a rule violation path if at least one waypoint coordinate falls within the no-fly area range; comparing a spatial distance of each flight segment of each candidate dynamic flight path at a predicted passing time with real-time position information of existing aircraft in the air space, and marking the candidate dynamic flight path as a conflict path if the spatial distance of at least one flight segment is less than the minimum aircraft separation requirement; extracting a total flight time requirement and a fuel consumption estimation coefficient from the path operation efficiency index; determining a total flight time of each candidate dynamic flight path that is not marked as a rule violation path or a conflict path, the total flight time being a sum of flight times of all flight segments, and the flight time of each flight segment being determined by a preset aircraft cruising speed; determining a fuel consumption estimation amount of each candidate dynamic flight path in combination with the fuel consumption estimation coefficient and the total flight time; The control priority weight is set according to the airspace control priority requirement, and the task efficiency weight is set according to the flight task efficiency requirement, and the sum of the control priority weight and the task efficiency weight is 1; For the illegal path or the conflict path, the safety compliance score is set to 0, and for the path without violation and conflict, the safety compliance score is determined based on the correspondence between the average value of the interval margin of each flight segment and the preset maximum allowed interval margin; The efficiency effectiveness score is determined based on the correspondence between the total flight time and the preset maximum allowed flight time, the correspondence between the fuel consumption estimation and the preset maximum allowed fuel consumption, and the preset time weight and fuel weight; The safety compliance score is multiplied by the control priority weight, and the efficiency effectiveness score is multiplied by the task efficiency weight, and then added to obtain the comprehensive evaluation value of each candidate dynamic flight path.

7. The dynamic path planning method based on the generative adversarial network model according to claim 1, wherein, The candidate dynamic flight path with the highest comprehensive evaluation value is taken as the feedback path, and the parameter adjustment module of the generator unit is input to adjust the waypoint distribution probability and flight segment time allocation weight of the path generation in the generator unit, including: The waypoint geographical position coordinate sequence, the flight segment heading sequence and the expected passing time sequence are extracted from the candidate dynamic flight path with the highest comprehensive evaluation value; The waypoint geographical position coordinate sequence is compared with the waypoint coordinates output by the current path skeleton generation layer of the generator unit to obtain a coordinate deviation sequence; The waypoint distribution probability of the path skeleton generation layer of the generator unit is adjusted according to the coordinate deviation sequence; The flight segment expected passing time sequence is compared with the time output by the current conflict pre-avoidance layer of the generator unit to obtain a time deviation sequence; The flight segment time allocation weight of the conflict pre-avoidance layer of the generator unit is adjusted according to the time deviation sequence; The adjusted waypoint distribution probability and flight segment time allocation weight are stored in the parameter library of the generator unit to overwrite the original parameters and be used for the generation of the next round of candidate dynamic flight paths.

8. The dynamic path planning method based on the generative adversarial network model according to claim 1, characterized in that, The generator unit is called to input the flight task instruction and the real-time running data set of the airspace to generate a plurality of new candidate dynamic flight paths, the safety compliance verification and the efficiency effectiveness evaluation are performed on each new candidate dynamic flight path, the comprehensive evaluation value of each new candidate dynamic flight path is output, the comprehensive evaluation values of all new candidate dynamic flight paths are extracted, and it is checked whether there is a candidate dynamic flight path with a comprehensive evaluation value reaching the preset path optimization threshold value; If there is, the candidate dynamic flight path with the highest comprehensive evaluation value and reaching the path optimization threshold value is selected as the final dynamic flight path; If not, the path with the highest comprehensive evaluation value in the new candidate dynamic flight path is taken as a new feedback path, and the waypoint distribution probability and flight segment time allocation weight of the generator unit are adjusted. ​ ​ ​ The above steps are repeatedly performed until the comprehensive evaluation value of the generated candidate dynamic flight path reaches a preset path optimization threshold. 9.The dynamic path planning method based on the generative adversarial network model according to claim 1, wherein, The training process of the generator unit and the discriminator unit of the pre-trained generative adversarial network includes: obtaining a historical flight task data set and a corresponding airspace historical operation data set, the historical flight task data set containing historical take-off airport identifiers, historical landing airport identifiers and corresponding historical optimal flight paths, the airspace historical operation data set containing real-time position information of existing aircraft in the historical airspace, time-space ranges of historical occupied air routes and historical airspace control constraint rules; dividing the historical flight task data set and the airspace historical operation data set into a training set and a validation set, the proportion of the training set data volume to the total data volume being a preset training set proportion; initializing parameters of the generator unit and parameters of the discriminator unit, the parameters of the generator unit including initial values of waypoint distribution probabilities of a path skeleton generation layer and initial values of flight segment time distribution weights of a conflict pre-avoidance layer, and the parameters of the discriminator unit including initial values of control priority weights for safety compliance verification and initial values of task efficiency weights for efficiency effectiveness evaluation; for each historical flight task data and corresponding airspace historical operation data in the training set: extracting historical take-off airport coordinates and historical landing airport coordinates from the historical flight task data, and extracting a historical airspace resource occupation heat map from the airspace historical operation data; inputting the historical take-off airport coordinates, the historical landing airport coordinates and the historical airspace resource occupation heat map into the generator unit to generate a plurality of historical candidate flight paths; inputting the historical candidate flight paths and the corresponding historical optimal flight paths into the discriminator unit to determine comprehensive evaluation values of the historical candidate flight paths, the comprehensive evaluation value of the historical optimal flight path being set as a preset maximum value; determining a loss value of the generator unit and a loss value of the discriminator unit, the loss value of the generator unit being a difference between the comprehensive evaluation values of the historical candidate flight paths and the comprehensive evaluation value of the historical optimal flight path, and the loss value of the discriminator unit being a difference between the comprehensive evaluation values output by the discriminator unit and the comprehensive evaluation value of the historical optimal flight path; adjusting the parameters of the generator unit and the discriminator unit using a back propagation algorithm, and repeating the training steps in the training set until the loss values of the training set reach a preset training termination threshold after minimizing the loss values of the generator unit and the discriminator unit; verifying the trained generator unit and the discriminator unit using the validation set to generate candidate flight paths of the validation set and determine a deviation between the comprehensive evaluation values of the candidate flight paths and the comprehensive evaluation value of the historical optimal flight path; if the deviation exceeds a preset deviation range, adjusting the training set proportion or the parameter initial values and re-executing the training steps, and if the deviation does not exceed the preset deviation range, saving the parameters of the generator unit and the discriminator unit as a pre-trained generative adversarial network model. 10.A dynamic path planning system based on a generative adversarial network model, characterized in that, The application relates to a dynamic path planning method based on a generative adversarial network model, comprising a processor and a memory, wherein the memory is connected with the processor, the memory is used for storing programs, instructions or codes, and the processor is used for executing the programs, instructions or codes in the memory to realize the dynamic path planning method based on the generative adversarial network model in any one of claims 1-9.

Images