A dynamic sequencing method for arriving and departing flights at urban air take-off and landing points

Abstract

The present invention discloses a method for dynamically sorting flights arriving and departing from urban air take-off and landing points. Taking into account the future urban air traffic operation scenarios, the present invention proposes the concept of an aerial apron, provides a method for dividing the corresponding aerial parking spaces, and then designs a take-off and landing program based on the aerial apron. In view of demand-responsive air traffic services, a collaborative sorting strategy for arriving and departing flights is proposed. With the goals of minimizing the waiting time of passengers in the air, the total delay time within a cycle, and the number of empty flights called by other take-off and landing points, an optimization model is established, and a deep neural network (DQN) algorithm is used for multi-objective optimization solution. Based on randomly generated flight dynamic data, an optimized sample data structure is designed, different state features are extracted, and a reasonable reward function is set to train the DQN to obtain a maturely trained DQN, thereby obtaining the final flight sorting that meets multi-objective optimization.

Description

Technical Field

[0001] The present invention relates to a method for dynamically sequencing arrival and departure flights at a city's aerial take-off and landing point. Background Art

[0002] Compared to traditional civil aviation, Didi Chuxing, and traditional public transportation, eVTOL operations differ significantly from traditional civil aviation in the following ways: eVTOL is a vertical take-off and landing aircraft that, through aerial operations, can navigate complex terrain and busy cities, offering greater adaptability, accessibility, efficiency, and safety. Compared to traditional civil aviation, eVTOL's take-off and landing space is more flexible, saving significant time on ground transportation. Compared to taxis, eVTOL can reach its destination more quickly, reducing problems like traffic congestion. Compared to traditional public transportation, eVTOL can go directly to its destination, reducing the time and inconvenience of transfers and transit. However, eVTOL operations also require addressing technical and management issues, such as flight altitude, noise pollution, and air traffic control.

[0003] Due to the operating environment constraints of urban low-altitude airspace, the scheduling of urban air traffic flight services needs to take into account the city's take-off and landing point configurations. Currently available eVTOLs, including Ehang's EH216, Fengfei's V1500M Shengshilong, Germany's Lilium, and the United States' JOBY, lack rolling take-off and landing capabilities. This means that both landing and take-off must be completed from a fixed apron. However, urban parking space capacity is extremely limited, resulting in landing space only available for subsequent eVTOLs after the current eVTOL has landed. Therefore, scheduling must allow for immediate take-off of eVTOLs after landing to free up take-off and parking spaces for subsequent flights. In reality, urban air traffic aircraft typically require very high take-off and landing frequencies, and effective scheduling and parking space management can significantly improve flight efficiency and space utilization. The present invention addresses the problem of rationally scheduling eVTOL arrival and departure flights and efficiently managing parking spaces. Summary of the Invention

[0004] A method for dynamically sequencing arriving and departing flights at a city airborne take-off and landing point, the method comprising the following steps:

[0005] S1. Establish an aerial apron. The aerial apron is established according to the divided airspace range, eVTOL-related parameters and parameters required for approach control, including an approach ring, an aerial parking space, a descent / take-off ring, and a vertical lift airport. The approach ring serves as a sign for the eVTOL to enter the controlled airspace of the vertical lift airport. There are approach points and departure points on it. When the eVTOL reaches the approach ring, it starts to execute the descent procedure. The aerial parking space is located above the vertical lift airport and is divided into multiple layers. It is numbered from the top to the bottom layer, and from the inner layer to the outer layer. The direction from the top to the bottom layer is the descent channel of the eVTOL. The eVTOL lands to the vertical lift airport through the descent channel. The outer area of ​​the aerial parking space and the area between the vertical lift airport are the ascent channel of the eVTOL. The eVTOL takes off from the vertical lift airport and leaves the aerial apron through the ascent channel. The descent channel and the ascent channel constitute the descent / take-off ring. Multiple layers of parking spaces are set from the outside to the inside of the aerial parking space, and the gaps between adjacent parking spaces serve as emergency descent channels.

[0006] Design the air apron: Input the parameters required for airspace design and approach control to design the optimal airspace, including the radius of the waiting circle, the layout of each waiting point, the size of the air parking space, etc. The entire airspace above the eVTOL approach and take-off point can be regarded as the air apron. This area includes the approach ring, emergency corridor, waiting area, descent / take-off ring, and is divided into multiple stands according to pre-numbering to facilitate the eVTOL to hover and wait. For the same numbered stand, the eVTOL will automatically fly to the lowest level that can be reached. The gap corridor between the stands is used as an emergency descent channel to facilitate the rapid descent of aircraft that are forced to land in special circumstances or have insufficient power.

[0007] S2. Establish apron-based arrival and departure procedures, including:

[0008] S201. The approach procedure primarily involves scanning the parking apron from the lowest level upwards for occupancy and allocating available parking spaces. The eVTOL then descends to the numbered parking space on the highest level, determining whether the next level is available. If so, it moves to the next level; otherwise, it hovers. After reaching the lowest level, the eVTOL determines whether the landing platform and descent / takeoff ring are both available. If both are available, the approach procedure continues; otherwise, it hovers.

[0009] S202: The departure procedure primarily involves determining whether the descent / takeoff loop is idle. If so, the drone vertically climbs to the descent / takeoff loop and pans to the target heading. Otherwise, the drone remains in a hovering state. The drone then vertically climbs to the approach loop and determines whether the eVTOL has reached the departure point in the approach loop. If so, the departure procedure is completed and the drone leaves the landing pad. Otherwise, the drone remains in a hovering state.

[0010] S3. Establish a dynamic sorting model for take-off and landing airspace: Taking into account the operator's profits and the service capabilities of the take-off and landing points, select the minimum total delay within the cycle, the minimum number of flights with passengers waiting time exceeding 3 minutes in the air, and the minimum number of empty flights dispatched specifically for passengers as optimization goals, and establish a multi-objective optimization mathematical model.

[0011] S4. Solving Dynamic Flight Ranking: The model is calculated using a Deep Q-Network (DQN) algorithm. The system obtains order demand within a cycle, including information such as the takeoff / landing procedures and scheduled takeoff and landing times for the flight at the departure and landing point, and generates a chronologically ordered flight ranking. Six state eigenvalues ​​are defined to describe the ranking state. These eigenvalues ​​are used as inputs to the DQN, which is trained with a reasonable reward function. A fully trained DQN is then developed to obtain the final flight ranking that satisfies multi-objective optimization requirements.

[0012] Furthermore, the method for dividing the air parking apron parking spaces in S1 is as follows:

[0013] S101. The size of the air parking apron can be obtained by the following formula:

[0014]

[0015]

[0016]

[0017]

[0018]

[0019] Where, Indicates the thrust received by the eVTOL rotor; Indicates the number of rotors of the eVTOL. The UAV studied in this invention has an eight-rotor structure; Indicates the atmospheric density, the present invention Take the sea level atmospheric density at 15°C ; represents the lift coefficient; Indicates the rotation speed of the drone; Indicates the diameter of the propeller; Indicates the yaw angle, roll angle and outer angle of attack of the drone during flight; Indicates the quality of the drone; Represent the maximum acceleration of the UAV in the horizontal and vertical directions respectively; Indicates the maximum cruising speed of the drone; Indicates the descent speed of the drone; Indicates the height threshold; the parking space is long and wide , Gao Wei cube.

[0020] S102. The number of air parking spaces can be obtained from the following formula:

[0021] The design airspace has a radius of , the height is Each floor's airborne parking bays are inscribed within the cylinder, leaving a gap in the corridor, and the bottom floor is left empty for a landing / takeoff loop.

[0022]

[0023] Where, Indicates the number of air parking spaces.

[0024] S103, the code for each floor of the apron is The parking spaces on each floor are coded from the inner layer to the outer layer, and the values ​​increase gradually in a clockwise direction.

[0025] Furthermore, the dynamic sorting model of the take-off and landing point airspace in S3 is as follows:

[0026] The present invention selects three optimization goals:

[0027] (1) When the passenger experience is the best, that is, the total delay in the period is the smallest, and all flights are close to the flight schedule:

[0028] Min

[0029] (2) When the flight regularity rate is the highest, that is, the number of flights with passengers waiting for more than 3 minutes in the air is the least:

[0030]

[0031] In the above formula, when the waiting time of a flight exceeds three minutes, that is, hour, .

[0032] (3) When resource utilization is highest, that is, the number of empty flights dispatched specifically for passengers is the least:

[0033]

[0034] In the above formula, when the current demand is to take off and there are no idle flights at the take-off and landing points, .

[0035] set up is the weighted value of the optimization objective:

[0036]

[0037] The multi-objective optimization function of the present invention is:

[0038]

[0039]

[0040]

[0041]

[0042]

[0043]

[0044]

[0045]

[0046]

[0047]

[0048] Where: The set of arrival flight requirements is , the set of departure flight demands is , the take-off and landing point parking stand collection is ; 、 Indicates arriving flights Planned arrival time and actual arrival time; 、 Departure flights Planned departure time and actual departure time; Indicates the total number of flights in the period; Indicates the number of flights that waited to hover for more than 3 minutes; Indicates dispatching an empty flight from another location. 、 represents a decision variable, which takes a value of 0 or 1; Indicates incoming flights Occupied a landing pad at the take-off point, Indicates departing flights A landing pad is occupied, and 0 means no landing pad is occupied. Represents a time period, Indicates the time within a cycle, and its value is ;

[0049] express The number of arriving flights within the take-off and landing points at the time, express The number of departure flights within the time slot, express ; Indicates the maximum time offset for a single flight, limiting flight adjustments within a certain range of the planned time; Indicates incoming flights The time of landing on the ground, Indicates the minimum time interval between flights, Indicates the minimum time the drone can stay on the ground.

[0050] Formulas (2) and (3) represent capacity constraints. Formula (2) indicates that only one parking space can be assigned to an incoming and outgoing flight. Formula (3) indicates that the number of flights at a take-off and landing point cannot be greater than the number of parking spaces. Formulas (4) and (5) represent maximum constraint time conversion. Formula (4) indicates that a flight must arrive within a specified time window. Formula (5) indicates that a flight must depart within a specified time window. Formulas (6), (7), and (8) indicate that only one aircraft can be allowed to perform take-off and landing operations at the same time during a single take-off and landing process. At the same time, the approach and departure procedures do not interfere with each other, and the time interval between take-off or landing flights must be greater than the threshold. Formula (9) indicates that the same parking space at the take-off and landing point The take-off and landing procedure intervals on the aircraft must be greater than the minimum ground stay time.

[0051] Furthermore, the process of solving the flight dynamic sorting model based on the DQN algorithm in S4 is as follows:

[0052] S401. Define six state feature values ​​to describe the sorting state and serve as the input of the DQN. The feature values ​​are as follows:

[0053] Average arrival delay time of flights: ;

[0054] The standard deviation of the average arrival delay time of flights is: ;

[0055] Average departure delay time for flights: ;

[0056] The standard deviation of the average departure delay time for flights is: ;

[0057] In addition, to describe the overall flight regularity, the ratio of the number of flights with a waiting hover time of more than 3 minutes to the total number of flights in the period is: ;

[0058] To describe resource utilization, the ratio of empty flights dispatched from other locations to the total number of flights in the period is: .

[0059] Among them, the set of arrival flight requirements is , the set of departure flight demands is , the take-off and landing point parking stand collection is ; Indicates the number of arriving flights; Indicates the number of departing flights; 、 Indicates arriving flights Planned arrival time and actual arrival time; 、 Respectively indicate the position of flights Planned departure time and actual departure time; Indicates the total number of flights in the period; Indicates the number of flights that waited to hover for more than 3 minutes; Indicates dispatching an empty flight from another location.

[0060] S402: Optimize the sample data and design a sample data structure storage method based on the right subtree to reduce the dimension of the flight schedule and improve training efficiency. The root node is the full flight ranking, which is equal to the sum of the rankings of all leaf nodes.

[0061] Initial observation data It is arranged in chronological order within the cycle After obtaining the initial observation data, start traversing the flights from left to right to determine whether the time difference between adjacent flights exceeds the set threshold. If it exceeds the limit, it will be split into two child nodes. Then continue to judge the right child node until all flights have been judged. Finally, all flights are stored according to the child node structure.

[0062] S403: Set a reasonable reward function to train the DQN to obtain a fully trained DQN.

[0063] The present invention designs the following reward mechanism for state feature values:

[0064] Located in The action selected from the decision network at time , is the reward value required by the agent, is the DQN attenuation coefficient, so the actual DQN target value is: ,in, represents the expected return at time t.

[0065] Step 1: Calculate the optimization target at time t+1 ;

[0066] Step 2: If the target value decreases, , then set the reward ;

[0067] Step 3: If the total delay time is reduced, ,but ;

[0068] Step 4: If the proportion of irregular flights decreases, ,but ;

[0069] Step 5: If the proportion of empty flights decreases, ,but ;

[0070] Step 6: If the flight sorting is completed, ;

[0071] Step 7: If the flight sorting fails and cannot be continued, then .

[0072] The present invention provides corresponding rewards according to the changes in the optimization target, total delay time, abnormal flight ratio, and empty flight ratio. In addition, when the flight sorting is completed, positive rewards must be given according to the final optimized sorting; when there is no feasible solution at the current moment, negative rewards are given.

[0073] S404. Within period T, the system obtains order requirements within the period, including information such as the takeoff / landing procedures and planned takeoff and landing times of the flights at the takeoff and landing points; then, the system selects an appropriate sorting optimization method through the DQN algorithm to obtain the final flight sorting that meets the multi-objective optimization requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 This is a bird's-eye view of the airfield;

[0075] Figure 2 This is the overall schematic diagram of the air apron;

[0076] Figure 3 This is a schematic diagram of the distribution and number calculation of air parking aprons;

[0077] Figure 4 Flowchart for arrival and departure procedures;

[0078] Figure 5 Schematic diagram of the take-off and landing intervals of aircraft in different scenarios;

[0079] Figure 6 Optimize the design diagram for sample data;

[0080] Figure 7 Flight ranking table after DQN optimization. DETAILED DESCRIPTION

[0081] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with examples and drawings. The exemplary embodiments of the present invention and their descriptions are only used to explain the present invention and are not intended to limit the present invention.

[0082] Example:

[0083] This method includes the following steps:

[0084] S1. Design the air apron: Input the parameters required for airspace design and approach control, and design the optimal airspace, including the radius of the waiting circle, the layout of each waiting point, the size of the air parking space, etc. The entire airspace above the eVTOL approach and take-off point can be regarded as the air apron. This area includes an approach ring, an emergency corridor, a waiting area, and a descent / take-off ring. It is divided into multiple stands according to pre-numbering to facilitate the eVTOL to hover and wait. For the same numbered stand, the eVTOL will automatically fly to the lowest level that can be reached. The gap corridor between the stands is used as an emergency descent channel to facilitate the rapid descent of aircraft that are forced to land in special circumstances or have insufficient power.

[0085] Figure 1 This is an overhead view of the airfield.

[0086] Figure 2 This is an overall schematic diagram of the air apron.

[0087] S101. The size of the air parking apron can be obtained by the following formula:

[0088]

[0089]

[0090]

[0091]

[0092]

[0093] Where, Indicates the thrust received by the eVTOL rotor; Indicates the number of rotors of the eVTOL. The UAV studied in this invention has an eight-rotor structure; Indicates the atmospheric density, the present invention Take the sea level atmospheric density at 15°C ; represents the lift coefficient; Indicates the rotation speed of the drone, the value ; Indicates the diameter of the propeller, the value is ; Indicates the yaw angle, roll angle and outer angle of attack of the drone during flight, and their values ​​are °; Indicates the quality of the drone, and its value is ; Represent the maximum acceleration of the UAV in the horizontal and vertical directions respectively; Indicates the maximum cruising speed of the drone, the value is ; Indicates the descent speed of the drone, the value is ; Indicates the height threshold, the value is ; The parking space is long and wide , Gao Wei cube.

[0094] S102. The number of air parking spaces can be obtained from the following formula:

[0095] The design airspace has a radius of , the height is of a cylinder, The values ​​are The air parking spaces on each floor are inscribed in the cylinder, leaving a gap corridor, and the bottom floor is left empty for landing / take-off ring.

[0096]

[0097] Where, Indicates the number of air parking spaces.

[0098] S103, the code for each floor of the apron is The parking spaces on each floor are coded from the inner layer to the outer layer, and the values ​​increase gradually in a clockwise direction.

[0099] Figure 3 This is a schematic diagram of the distribution and number calculation of air parking aprons.

[0100] S2. Take-off and landing procedure design: The design is based on the take-off and landing procedures and sequencing update rules of the air parking apron.

[0101] S201. The approach procedure primarily involves scanning the parking apron from the lowest level upwards for occupancy and allocating available parking spaces. The eVTOL then descends to the numbered parking space on the highest level, determining whether the next level is available. If so, it moves to the next level; otherwise, it hovers. After reaching the lowest level, the eVTOL determines whether the landing platform and descent / takeoff ring are both available. If both are available, the approach procedure continues; otherwise, it hovers.

[0102] S202: The departure procedure primarily involves determining whether the descent / takeoff loop is idle. If so, the drone vertically climbs to the descent / takeoff loop and pans to the target heading. Otherwise, the drone remains in a hovering state. The drone then vertically climbs to the approach loop and determines whether the eVTOL has reached the departure point in the approach loop. If so, the departure procedure is completed and the drone leaves the landing pad. Otherwise, the drone remains in a hovering state.

[0103] Figure 4 Flowchart for entry and exit procedures.

[0104] S3. Establish a dynamic sorting model for take-off and landing airspace: Taking into account passenger comfort, operator profits, and take-off and landing service capabilities, select the flight with the smallest total delay within the cycle, the fewest number of flights with passengers waiting time exceeding 3 minutes in the air, and the fewest number of empty flights dispatched specifically for passengers as optimization objectives, and establish a multi-objective optimization mathematical model.

[0105] The dynamic sorting model of the S3 take-off and landing point airspace is as follows:

[0106] The present invention selects three optimization goals:

[0107] (1) When the passenger experience is the best, that is, the total delay in the period is the smallest, and all flights are close to the flight schedule:

[0108] Min

[0109] (2) When the flight regularity rate is the highest, that is, the number of flights with passengers waiting for more than 3 minutes in the air is the least:

[0110]

[0111] In the above formula, when the waiting time of a flight exceeds three minutes, that is, hour, .

[0112] (3) When resource utilization is highest, that is, the number of empty flights dispatched specifically for passengers is the least:

[0113]

[0114] In the above formula, when the current demand is to take off and there are no idle flights at the take-off and landing points, .

[0115] set up is the weighted value of the optimization objective:

[0116]

[0117] The multi-objective optimization function of the present invention is:

[0118]

[0119]

[0120]

[0121]

[0122]

[0123]

[0124]

[0125]

[0126]

[0127]

[0128] Where: The set of arrival flight requirements is , the set of departure flight demands is , the take-off and landing point parking stand collection is ; 、 Indicates arriving flights Planned arrival time and actual arrival time; 、 Departure flights Planned departure time and actual departure time; Indicates the total number of flights in the period; Indicates the number of flights that waited to hover for more than 3 minutes; Indicates dispatching an empty flight from another location. 、 represents a decision variable, which takes a value of 0 or 1; Indicates incoming flights Occupied a landing pad at the take-off point, Indicates departing flights A landing pad is occupied, and 0 means no landing pad is occupied. Represents a time period, Indicates the time within a cycle, and its value is ;

[0129] express The number of arriving flights within the take-off and landing points at the time, express The number of departure flights within the time slot, express ; Indicates the maximum time offset for a single flight, limiting flight adjustments within a certain range of the planned time; Indicates incoming flights The time of landing on the ground, Indicates the minimum time interval between flights, Indicates the minimum time the drone can stay on the ground.

[0130] Formulas (2) and (3) represent capacity constraints. Formula (2) indicates that only one parking space can be assigned to an incoming and outgoing flight. Formula (3) indicates that the number of flights at a take-off and landing point cannot be greater than the number of parking spaces. Formulas (4) and (5) represent maximum constraint time conversion. Formula (4) indicates that a flight must arrive within a specified time window. Formula (5) indicates that a flight must depart within a specified time window. Formulas (6), (7), and (8) indicate that only one aircraft can be allowed to perform take-off and landing operations at the same time during a single take-off and landing process. At the same time, the approach and departure procedures do not interfere with each other, and the time interval between take-off or landing flights must be greater than the threshold. Formula (9) indicates that the same parking space at the take-off and landing point The take-off and landing procedure intervals on the aircraft must be greater than the minimum ground stay time.

[0131] Figure 5 Schematic diagram of the take-off and landing intervals of aircraft in different scenarios.

[0132] S4. Flight dynamic sorting solution: The Deep Q Network (DQN) algorithm is used to calculate the model.

[0133] S401. Define six state feature values ​​to describe the sorting state and serve as the input of the DQN. The feature values ​​are as follows:

[0134] Average arrival delay time of flights: ;

[0135] The standard deviation of the average arrival delay time of flights is: ;

[0136] Average departure delay time for flights: ;

[0137] The standard deviation of the average departure delay time for flights is: ;

[0138] In addition, to describe the overall flight regularity, the ratio of the number of flights with a waiting hover time of more than 3 minutes to the total number of flights in the period is: ;

[0139] To describe resource utilization, the ratio of empty flights dispatched from other locations to the total number of flights in the period is: .

[0140] Among them, the set of arrival flight requirements is , the set of departure flight demands is , the take-off and landing point parking stand collection is ; Indicates the number of arriving flights; Indicates the number of departing flights; 、 Indicates arriving flights Planned arrival time and actual arrival time; 、 Respectively indicate the position of flights Planned departure time and actual departure time; Indicates the total number of flights in the period; Indicates the number of flights that waited to hover for more than 3 minutes; Indicates dispatching an empty flight from another location.

[0141] S402: Optimize the sample data and design a sample data structure storage method based on the right subtree to reduce the dimension of the flight schedule and improve training efficiency. The root node is the full flight ranking, which is equal to the sum of the rankings of all leaf nodes.

[0142] Initial observation data It is arranged in chronological order within the cycle After obtaining the initial observation data, start traversing the flights from left to right to determine whether the time difference between adjacent flights exceeds the set threshold. If it exceeds the limit, it will be split into two child nodes. Then continue to judge the right child node until all flights have been judged. Finally, all flights are stored according to the child node structure.

[0143] Figure 6 Optimize the design diagram for sample data.

[0144] S403: Set a reasonable reward function to train the DQN to obtain a fully trained DQN.

[0145] The present invention designs the following reward mechanism for state feature values:

[0146] Located in The action selected from the decision network at time , is the reward value required by the agent, is the DQN attenuation coefficient, so the actual DQN target value is: ,in, represents the expected return at time t.

[0147] Step 1: Calculate the optimization target at time t+1 ;

[0148] Step 2: If the target value decreases, , then set the reward ;

[0149] Step 3: If the total delay time is reduced, ,but ;

[0150] Step 4: If the proportion of irregular flights decreases, ,but ;

[0151] Step 5: If the proportion of empty flights decreases, ,but ;

[0152] Step 6: If the flight sorting is completed, ;

[0153] Step 7: If the flight sorting fails and cannot be continued, then .

[0154] The present invention provides corresponding rewards according to the changes in the optimization target, total delay time, abnormal flight ratio, and empty flight ratio. In addition, when the flight sorting is completed, positive rewards must be given according to the final optimized sorting; when there is no feasible solution at the current moment, negative rewards are given.

[0155] S404. Within period T, the system obtains order requirements within the period, including information such as the takeoff / landing procedures and planned takeoff and landing times of the flights at the takeoff and landing points; then, the system selects an appropriate sorting optimization method through the DQN algorithm to obtain the final flight sorting that meets the multi-objective optimization requirements.

[0156] Figure 7 Flight ranking table after DQN optimization.

[0157] Considering future urban air traffic scenarios, this paper designs a takeoff and landing program based on an airfield. Targeting demand-responsive air traffic services, a coordinated scheduling strategy for arrival and departure flights is proposed. This strategy targets minimizing passenger waiting time in the air, total delay time within a cycle, and the number of empty flights called from other takeoff and landing points. An optimization model is established, employing a deep-learning neural network (DQN) algorithm for multi-objective optimization. Based on randomly generated flight dynamics data, an optimized sample data structure is designed to extract different state features, improving model training efficiency.

Claims

1. A method for dynamically sequencing flights arriving and departing from a city airborne take-off and landing point, characterized in that: The following steps are involved: S1. Establish an air apron. According to the divided airspace range, eVTOL-related parameters and parameters required for approach control, establish an air apron, including an approach ring, an air parking stand, a descent / take-off ring, and a vertical lift airport. The air parking stand is located above the vertical lift airport and is divided into multiple layers, numbered from the top to the bottom layer, and numbered from the inner layer to the outer layer. The direction from the top to the bottom layer is the descent channel of the eVTOL, and the eVTOL lands at the vertical lift airport through the descent channel. The area outside the airborne parking stand and between it and the vertical take-off and landing area serves as the eVTOL's ascent channel. The eVTOL takes off from the vertical take-off and landing area via the ascent channel, and the descent channel and the ascent channel form a descent / take-off loop. Multiple layers of parking stands are arranged from the outside to the inside of the airborne parking stand, and the gaps between adjacent parking stands serve as emergency descent channels. S2. Establish apron-based arrival and departure procedures, including: S201. Establish an approach procedure: Scan the occupancy of the airborne parking spaces from the lowest level upwards and allocate a vacant parking space. The eVTOL lands at the parking space with the same number on the highest level according to the allocated vacant parking space number and determines whether the next level is vacant. If so, it moves to the next level; otherwise, it hovers. After the eVTOL reaches the lowest level, it determines whether the take-off and landing platform and the descent / take-off ring are vacant. If both are vacant, it continues the approach procedure; otherwise, it hovers. S202: Establish a departure procedure: Determine whether the descent / takeoff loop is in an idle state. If so, vertically climb to the descent / takeoff loop and translate to the target heading. Otherwise, the aircraft remains in a hovering state. Based on the target heading, vertically climb to the approach loop and determine whether the eVTOL has reached the departure point in the approach loop. If so, the departure procedure is completed and the eVTOL leaves the apron. Otherwise, the aircraft remains in a hovering state. S3. Establish a dynamic sorting model for take-off and landing point airspace: Taking into account passenger comfort, operator profits, and take-off and landing point service capabilities, the optimization objectives are to minimize total delays, minimize the number of flights with passenger waiting times exceeding 3 minutes in the air, and minimize the number of empty flights dispatched specifically for passengers. A multi-objective optimization mathematical model is then established. S4. Solving the dynamic sorting of flights: The deep Q-network algorithm is used to calculate the model established in S3. The system obtains order demand within the cycle, including information related to the take-off / landing procedures and planned take-off and landing times of the flights at the take-off and landing points, and generates a flight sorting based on the chronological sequence. Six state feature values ​​are defined to describe the sorting state: the average arrival delay time of the flight, the standard deviation of the average arrival delay time of the flight, the average departure delay time of the flight, the standard deviation of the average departure delay time of the flight, the ratio of the number of flights with a waiting hovering time of more than 3 minutes to the total number of flights in the cycle, and the ratio of flights dispatched from other places with no passengers to the total number of flights in the cycle. The state feature values ​​are used as the input of the DQN, and the reward function is set to train the DQN. The trained DQN is obtained, and the final flight sorting that meets multi-objective optimization is obtained.

2. A method for dynamically sequencing flights arriving and departing from a city airborne take-off and landing point according to claim 1, characterized in that: The method for establishing an airborne parking stand in S1 is as follows: S101. The size of a single airborne parking stand is obtained from the following formula: ， ， ， ， ， Where, Indicates the thrust received by the eVTOL rotor; Indicates the number of rotors of the eVTOL; represents the atmospheric density; represents the lift coefficient; Indicates the rotational speed of the eVTOL; Indicates the diameter of the propeller; Indicates the yaw angle, roll angle, and outside angle of attack during eVTOL flight; Indicates the mass of the eVTOL; Represent the maximum acceleration of eVTOL in the horizontal and vertical directions respectively; Indicates the maximum cruising speed of the eVTOL; Indicates the descent speed of the eVTOL; Indicates the height threshold; a single parking space is long and wide , Gao Wei cube; S102. The number of air parking spaces is obtained from the following formula: Set the airspace to have a radius of , height is The cylinder is inscribed in each layer of air parking spaces, with gaps between adjacent parking spaces. A layer of space is left between the bottom layer and the vertical lift airport to serve as the takeoff channel of the descent / takeoff loop, resulting in: ， Where, Indicates the number of air parking spaces; S103, encode the air parking spaces from the top to the bottom Each parking space on each floor is coded with Arabic numerals from the inner layer to the outer layer, and the values ​​increase gradually in a clockwise direction.

3. The method for dynamically sequencing flights arriving and departing from a city airborne take-off and landing point according to claim 1, characterized in that: The specific method for establishing a dynamic sorting model for take-off and landing point airspace in S3 is as follows: Select three optimization objectives: (1) With passenger comfort as the goal, the total delay within the period is minimized and all flights are close to the flight schedule: Min , (2) The number of flights with passengers waiting for more than 3 minutes in the air is the least: ， In the above formula, when The waiting time for flights exceeds three minutes. hour, ; (3) When resource utilization is highest, the number of empty flights dispatched specifically for passengers is the least: ， In the above formula, when the current demand is to take off and there are no idle flights at the take-off and landing points, ; set up is the weighted value of the optimization objective: ， The multi-objective optimization function is established as: ， ， ， ， ， ， ， ， ， ； Among them, the set of arrival flight requirements is , the set of departure flight demands is , the take-off and landing point parking stand collection is ; 、 Indicates arriving flights Planned arrival time and actual arrival time; 、 Departure flights Planned departure time and actual departure time; Indicates the total number of flights in the period; Indicates the number of flights that waited to hover for more than 3 minutes; It means dispatching empty flights from other places; 、 represents a decision variable, which takes a value of 0 or 1; Indicates incoming flights Occupied a landing pad at the take-off point, Indicates departing flights One apron at the take-off and landing point is occupied, and 0 means no apron is occupied; Represents a time period, Indicates the time within a cycle, and its value is ; express The number of arriving flights within the take-off and landing points at the time, express The number of departure flights within the time slot, express ; Indicates the maximum time offset for a single flight, limiting flight adjustments within a certain range of the planned time; Indicates incoming flights The time of landing on the ground, Indicates the minimum time interval between flights, Indicates the minimum ground dwell time of eVTOL; Formulas (2) and (3) represent capacity constraints. Formula (2) indicates that only one parking space can be assigned to an approaching or departing flight. Formula (3) indicates that the number of flights at a take-off and landing point cannot be greater than the number of parking spaces. Formulas (4) and (5) represent maximum constraint time conversion. Formula (4) indicates that a flight must arrive within the specified time window. Formula (5) indicates that a flight must depart within the specified time window. Formulas (6), (7), and (8) indicate that only one aircraft can be allowed to perform take-off and landing operations at the same time during a single take-off and landing process. At the same time, the approach and departure procedures do not interfere with each other, and the time interval between take-off or landing flights must be greater than the threshold. , formula (9) indicates that the same parking space at the take-off and landing point The take-off and landing procedure intervals on the aircraft must be greater than the minimum ground stay time.

4. The method for dynamically sequencing flights arriving and departing from a city airborne take-off and landing point according to claim 1, characterized in that: The specific process of using the DQN deep Q network algorithm in S4 to solve the spatial dynamic sorting model is as follows: S401. Define six state feature values ​​to describe the sorting state and serve as the input of DQN. The feature values ​​are as follows: Average arrival delay time of flights: ; The standard deviation of the average arrival delay time of flights is: ; Average departure delay time for flights: ; The standard deviation of the average departure delay time for flights is: ; In addition, to describe the overall flight regularity, the ratio of the number of flights with a waiting hover time of more than 3 minutes to the total number of flights in the period is: ; To describe resource utilization, the ratio of empty flights dispatched from other locations to the total number of flights in the period is: ; Among them, the set of arrival flight requirements is , the set of departure flight demands is , the take-off and landing point parking stand collection is ; Indicates the number of arriving flights; Indicates the number of departing flights; 、 Indicates arriving flights Planned arrival time and actual arrival time; 、 Respectively indicate the position of flights Planned departure time and actual departure time; Indicates the total number of flights in the period; Indicates the number of flights that waited to hover for more than 3 minutes; It means dispatching empty flights from other places; S402: Optimize the sample data and design a sample data structure storage method based on the right subtree to reduce the dimension of the flight schedule and improve training efficiency. The root node is the full flight ranking, which is equal to the sum of the rankings of all leaf nodes. Initial observation data It is arranged in chronological order within the cycle After obtaining the initial observation data, start traversing the flights from left to right to determine whether the time difference between adjacent flights exceeds the set threshold If it exceeds the limit, it will be divided into two child nodes, and then the right child node will be judged until all flights are judged. Finally, all flights are stored according to the structure of the child nodes. S403. Set the reward function to train the DQN to obtain a fully trained DQN: Suppose the action selected from the decision network at time t is , is the reward value required by the agent, is the DQN attenuation coefficient. The actual DQN target value is: ,in, represents the expected return at time t; the training process is: Step 1: Calculate the optimization target at time t+1 ; Step 2: If the target value decreases, , then set the reward ; Step 3: If the total delay time is reduced, ,but ; Step 4: If the proportion of irregular flights decreases, ,but ; Step 5: If the proportion of empty flights decreases, ,but ; Step 6: If the flight sorting is completed, ; Step 7: If the flight sorting fails and cannot be continued, then ; Rewards are awarded based on the changes in the optimization objective, total delay time, the proportion of irregular flights, and the proportion of empty flights. When the flight sorting is complete, positive rewards are awarded based on the final optimized sorting. Negative rewards are awarded when there is no feasible solution at the current moment. S404. Within period T, the system obtains order requirements within the period, including information such as the takeoff / landing procedures and planned takeoff and landing times of the flights at the takeoff and landing points; then, the system selects an appropriate sorting optimization method through the DQN algorithm to obtain the final flight sorting that meets the multi-objective optimization requirements.

Images