An intelligent scheduling method and system for an aircraft to jump into an aircraft hangar

Abstract

The present invention discloses an intelligent scheduling method and system for an aircraft jump hangar, including the hangar obtaining local aircraft flight operation data in real time; the hangar obtaining the parking request information of the jump aircraft in real time, performing data communication with the aircraft control center in real time according to the parking request information, obtaining the flight operation data of the aircraft, and synchronizing the state of the aircraft with the hangar; obtaining the real-time scene operation data of the hangar, and performing aircraft cluster scheduling according to the real-time scene operation data of the hangar; the present invention realizes dynamic adjustment of the scheduling strategy through real-time acquisition and synchronization of the state data of the hangar and the aircraft, combines deep reinforcement learning and mixed integer programming, realizes load balancing and optimal utilization of resources, solves the problems of scheduling chaos and errors caused by communication failures in the existing jump hangar scheduling system, and the DRL model optimizes task allocation by learning the real-time states of the aircraft and the hangar, ensures load balancing, and reduces waiting time.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent scheduling, and specifically to an intelligent scheduling method and system for aircraft skip hangars. Background Art

[0002] The current skip hangar scheduling system involves multi-aircraft cluster scheduling of local aircraft and roaming aircraft, requiring efficient communication and coordination between the aircraft cluster and multiple hangars. However, in the existing system, due to the easy occurrence of faults in the communication link between the aircraft cluster and the hangar, the data transmission is unstable, leading to scheduling chaos and errors. Such communication problems will seriously affect the accuracy of cluster scheduling and the reliability of the system, thus hindering the orderly operation of the hangar, increasing the risk of task delays, and reducing the overall operation efficiency. Summary of the Invention

[0003] The purpose of the present invention is to provide an intelligent scheduling method and system for aircraft skip hangars, which combines deep reinforcement learning and mixed integer programming to achieve dynamic adjustment of scheduling strategies and optimize task allocation, load balancing, and resource utilization.

[0004] The purpose of the present invention can be achieved by the following technical solutions:

[0005] The present application provides an intelligent scheduling method for aircraft skip hangars, including

[0006] An intelligent scheduling method for aircraft skip hangars includes the following steps:

[0007] The hangar obtains the flight operation data of local aircraft in real time;

[0008] The hangar obtains the parking request information of the skip aircraft in real time, communicates with the aircraft control center in real time according to the parking request information, obtains the flight operation data of the aircraft, and synchronizes the state of the aircraft with the hangar;

[0009] Obtain the real-time scene operation data of the hangar, and perform aircraft cluster scheduling according to the real-time scene operation data of the hangar; the scene operation data includes the working sites, supply operation sites, takeoff and landing operation sites, and idle operation sites;

[0010] Among them, the method for performing aircraft cluster scheduling according to the real-time scene operation data of the hangar includes: obtaining the real-time position information of the first aircraft from the first hangar; sending the real-time position information of the first aircraft to the second hangar; when the first hangar receives the real-time position information of the third aircraft, sending a task instruction to the third aircraft, where the third aircraft is an aircraft in the second hangar.

[0011] Further, the data synchronization of the aircraft state with the hangar is specifically used to obtain the position information and task status of the aircraft in real time, generate and display a task assignment table, the content of which includes:

[0012] The aircraft control center receives a first data packet from the first aircraft, the first data packet carrying the current position information and the current task information of the first aircraft; receives a second data packet from the third aircraft, the second data packet carrying the current position information and the current task information of the third aircraft; generates a task assignment table according to the first data packet and the second data packet, the task assignment table recording the position information and the current task information of each aircraft; displays the task assignment table.

[0013] Further, the flight operation data includes at least one of the following: the position information of the aircraft, the flight route, the takeoff and landing time, the flight load, the load information, the heading information, the airspeed information, the remaining fuel information, and the status information.

[0014] Further, before obtaining the real-time position information of the first aircraft from the first hangar, it further includes: receiving a task request message from the user terminal, the task request message carrying the identification information of the first aircraft.

[0015] Further, when generating the task assignment table, the task assignment is optimized through deep reinforcement learning, and the specific content includes:

[0016] Define the state vector s as the current state of the aircraft and the hangar; the state vector s is expressed as: ;

[0017] Among them, represents the real-time position coordinates of the 1st to the nth aircraft; represents the real-time speed of the 1st to the nth aircraft; represents the remaining fuel of the 1st to the nth aircraft; represents the real-time task status of the hangar;

[0018] Define the action a as the scheduling decision, including , where represents the scheduling action for the i-th aircraft;

[0019] Design a reward function R to evaluate the quality of each action;

[0020] Specifically where T(s,a) represents the task completion time, S(s,a) represents the safety index, U(s,a) represents the resource utilization rate, is the weight coefficient;

[0021] Define the state value function and the action value function to quantify the value of each state and action;

[0022] Use the loss function of the DQN algorithm for training, update the network weights θ, and select the optimal action value function;

[0023] The loss function is expressed as: , where y represents the target Q value; is the value estimate of the current network for taking action a in state s; θ is the current network parameter, represents the expected value, which is used to represent the statistical average of the samples;

[0024] Deploy the trained DRL model to the scheduling system and output the optimal scheduling decision in real time according to the current state s.

[0025] Furthermore, in the process of optimizing task allocation, the mixed-integer programming model is used to handle the constraint conditions and optimization problems. The specific content includes:

[0026] According to the action space of the DRL model, define the decision variables of the MIP model. The decision to assign tasks to each aircraft is represented as a binary variable , where i represents the index of the aircraft and j represents the index of the task;

[0027] Combine the reward function of the DRL model to construct the objective function of the MIP model;

[0028] The specific objective function is expressed as: ; where represents the accumulation of the costs for all aircraft i and all tasks j, is the weight coefficient of the task completion time, represents the safety index when aircraft i executes task j, is the weight coefficient of the resource utilization rate, represents the resource utilization rate when aircraft i executes task j, is the decision variable;

[0029] Add the constraint conditions of the MIP model according to the real-time states of the aircraft and the hangar;

[0030] Use a mathematical optimization solver to solve the MIP model and find the optimal solution that satisfies all the constraints;

[0031] Analyze the solution results, apply the optimization results to the actual aircraft scheduling, and adjust the model parameters according to the actual operation situation.

[0032] Furthermore, the constraint conditions of the MIP model include: task assignment constraint, where each task is exactly assigned to one aircraft, which is expressed as: ;

[0033] The aircraft capacity constraint means that the load of the aircraft shall not exceed its maximum capacity, expressed as: ; where represents the resource requirement of mission j for aircraft i, represents the resource capacity of aircraft i;

[0034] The time window constraint means that the take-off and landing times of the aircraft must be within a specific time window, expressed as: ; where represents the time when aircraft i starts to execute mission j, and represent the earliest and latest take-off and landing times of aircraft i respectively;

[0035] The safety constraint is to ensure a safe distance between aircraft or other safety standards, expressed as: , where represents the safety threshold.

[0036] Furthermore, the method further includes real-time collision detection: when it is detected that there is a collision risk between the first aircraft and the third aircraft, an avoidance instruction is sent to the third aircraft so that the third aircraft bypasses the area where the first aircraft is located.

[0037] An intelligent scheduling system for aircraft skip hangar includes a data synchronization and communication module, a real-time scenario analysis module, and an intelligent scheduling decision module.

[0038] The data synchronization and communication module is used to obtain the local aircraft flight operation data in real time, obtain the parking request information of the skip flight aircraft, and perform data communication with the aircraft control center to synchronize the status of the aircraft and the hangar data.

[0039] The real-time scenario analysis module is used to obtain the real-time scenario operation data of the hangar, obtain the real-time position information of the first aircraft from the first hangar, and send the information to the second hangar, receive the real-time position information of the third aircraft, and send task instructions.

[0040] The intelligent scheduling decision module makes intelligent scheduling decisions through a deep reinforcement learning unit and a mixed integer programming unit, sends task instructions to the aircraft, and adjusts the flight route.

[0041] Furthermore, the deep reinforcement learning unit uses the DRL algorithm to optimize task allocation, trains the DRL model, updates the network weights to select the optimal action value function, deploys the trained DRL model to the scheduling system, and outputs the optimal scheduling decision in real time according to the current state.

[0042] The mixed-integer programming unit represents the decision of task allocation for each aircraft as a binary variable by defining the decision variables of the MIP model, constructs the objective function of the MIP model, adds the constraint conditions of the MIP model, and then uses a mathematical optimization solver to solve the MIP model to find the optimal solution that satisfies all constraints.

[0043] The beneficial effects of the present invention are as follows:

[0044] By obtaining and synchronizing the status data of the hangar and the aircraft in real time, combining deep reinforcement learning and mixed-integer programming, and dynamically adjusting the scheduling strategy, the present invention realizes load balancing and optimal utilization of resources, solves the problems of scheduling chaos and errors caused by communication failures in the existing aircraft jump hangar scheduling system. The DRL model optimizes task allocation by learning the real-time status of the aircraft and the hangar, ensures load balancing, reduces waiting time, and improves operation efficiency. At the same time, the MIP model processes constraint conditions such as aircraft capacity and time window to ensure scheduling under the premise of satisfying all constraints. In addition, the real-time collision detection function avoids the collision risk between aircraft by sending avoidance instructions, enhancing the safety of the system. This method not only improves the intelligence and efficiency of the scheduling system, but also enhances its flexibility and robustness, effectively coping with the dynamically changing operating environment and showing strong application potential in complex air traffic scheduling problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] For better understanding and implementation, the technical solutions of the present application will be described in detail below with reference to the drawings.

[0046] Figure 1 FIG. is a schematic flow chart of an intelligent scheduling method for an aircraft jump hangar provided in Embodiment 1 of the present application;

[0047] Figure 2 FIG. is a schematic flow chart of optimizing task allocation by deep reinforcement learning for an intelligent scheduling method for an aircraft jump hangar provided in Embodiment 1 of the present application;

[0048] Figure 3 FIG. is a schematic flow chart of a mixed-integer programming model for an intelligent scheduling method for an aircraft jump hangar provided in Embodiment 1 of the present application for processing constraint conditions and optimization problems. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] To further elaborate on the technical means and effects adopted by the present invention to achieve the intended invention purpose, exemplary embodiments will be described in detail herein, and the examples are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of methods and systems consistent with some aspects of the present application as detailed in the appended claims.

[0050] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms "a", "the", and "said" used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0051] The following, in conjunction with the drawings and preferred embodiments, details the specific implementation manners, features, and effects of the present invention.

[0052] Embodiment 1

[0053] Please refer to Figures 1-3 , this embodiment provides an intelligent scheduling method and system for aircraft jump hangars, which combines deep reinforcement learning and mixed integer programming to dynamically adjust the scheduling strategy, optimize task allocation, load balancing, and resource utilization.

[0054] The present invention provides an intelligent scheduling method for aircraft jump hangars, including the following steps:

[0055] S1. The hangar obtains the local aircraft flight operation data in real time;

[0056] S2. The hangar obtains the parking request information of the jump aircraft in real time, communicates with the aircraft control center in real time according to the parking request information, obtains the flight operation data of the aircraft, and synchronizes the state of the aircraft with the hangar.

[0057] S3. Obtain the real-time scene operation data of the hangar, and perform aircraft cluster scheduling according to the real-time scene operation data of the hangar.

[0058] Among them, the method of performing aircraft cluster scheduling according to the real-time scene operation data of the hangar includes: obtaining the real-time position information of the first aircraft from the first hangar; sending the real-time position information of the first aircraft to the second hangar; when the first hangar receives the real-time position information of the third aircraft, sending a task instruction to the third aircraft, where the third aircraft is an aircraft in the second hangar.

[0059] The described scenario operation data includes the working sites, supply operation sites, take-off and landing operation sites, idle operation sites, etc.

[0060] The invention can globally control the operation state of the target hangar scenario and form a reasonable, efficient and orderly intelligent scheduling method for the aircraft cluster.

[0061] Furthermore, the synchronization of the aircraft state with the hangar is specifically used to obtain the position information and task status of the aircraft in real time, generate and display a task allocation table, and the content includes:

[0062] The aircraft control center receives a first data packet from the first aircraft, and the first data packet carries the current position information and the current task information of the first aircraft; receives a second data packet from the third aircraft, and the second data packet carries the current position information and the current task information of the third aircraft; generates a task allocation table according to the first data packet and the second data packet, and the task allocation table records the position information of each aircraft and the current task information; displays the task allocation table.

[0063] Furthermore, the flight operation data includes at least one of the following: the position information of the aircraft, the flight route, the take-off and landing time, the flight load, the load information, the heading information, the speed information, the remaining fuel information, and the status information.

[0064] Furthermore, before obtaining the real-time position information of the first aircraft from the first hangar, it further includes: receiving a task request message from the user terminal, and the task request message carries the identification information of the first aircraft.

[0065] Furthermore, when generating the task allocation table, deep reinforcement learning can be used to optimize the task allocation, ensure the load balance of the aircraft, reduce the waiting time, and improve the overall operation efficiency.

[0066] The specific content includes:

[0067] S11. Define the state vector s as the current state of the aircraft and the hangar;

[0068] Including: ;

[0069] Among them, represents the real-time position coordinates of the 1st to the nth aircraft; represents the real-time speed of the 1st to the nth aircraft; represents the remaining fuel of the 1st to the nth aircraft; represents the real-time task status of the hangar, including information such as task type, task priority, task completion degree, etc.

[0070] Specifically, the state vector s represents the current state information of the aircraft, reflecting the real-time data of the aircraft, which specifically includes: The position information is the position of the aircraft and can be part of the state vector, used to represent the physical position of the current aircraft; The velocity vector is the velocity of the aircraft and is also an important part of the state vector, describing the dynamic characteristics of the aircraft, and the real-time velocity is a key parameter necessary for the aircraft during navigation and mission execution; The mission information is that the state vector should also include the mission state of the aircraft (such as whether it is in the process of executing a mission, mission priority, etc.); The state vector s is actually a multi-dimensional information vector integrating the real-time data of the aircraft.

[0071] S12. Define the action a as the scheduling decision, including , where represents the scheduling action for the i-th aircraft, such as task allocation, flight route adjustment, avoidance instruction issuance, etc.;

[0072] Specifically, the action a is the decision made by the system on the state of the aircraft, and the action can include the following: The position adjustment action is based on the current position and velocity of the aircraft (provided by the state vector), and the action can determine the next moving direction or velocity adjustment of the aircraft. For example, based on the distance between the real-time position of the aircraft and the target mission point, the action will guide its moving path; The task allocation action involves the decision-making of specific task allocation, especially based on the real-time priority of the task and the current task state of the aircraft, and can dynamically change the task execution order to ensure the optimal utilization of resources.

[0073] S13. Design the reward function R to evaluate the quality of each action, including indicators such as task completion time, safety, resource utilization rate, etc.,

[0074] Specifically where T(s, a) represents the task completion time, S(s, a) represents the safety index, U(s, a) represents the resource utilization rate, is the weight coefficient.

[0075] S14. Define the state value function and the action value function , quantifying the value of each state and action, and helping the intelligent scheduling system to make optimal decisions.

[0076] Specifically ; , where represents the cumulative reward after taking the action a in the state s.

[0077] S15. Use the loss function of the DQN algorithm for training, update the network weight θ, and select the optimal action value function;

[0078] Specifically calculate the target Q-value y: , where r is the immediate reward received after taking action a at time step t; γ is the discount factor, which is used to measure the current value of future rewards; is the environmental state at time step t+1; is at state the optimal action among all possible actions; is the set of network parameters other than the current action a, which is used to calculate the expected reward for the next state under.

[0079] The loss function is expressed as: , where is the value estimate of the current network for taking action a in state s; θ is the current network parameter, represents the expected value, which is used here to represent the statistical average of the samples.

[0080] Specifically, by minimizing the loss function L(θ), the network parameter θ can be adjusted to make the network's estimate of the action value closer to the target Q-value y, thereby improving the accuracy of model prediction.

[0081] S16. Deploy the trained DRL model to the scheduling system and output the optimal scheduling decision in real time according to the current state s , .

[0082] The DRL model can learn to make optimal scheduling decisions in complex dynamic environments, thereby achieving load balancing of the aircraft, reducing waiting time, and improving overall operation efficiency. Through deep reinforcement learning (DRL) to optimize task allocation, real-time analysis and decision-making of the aircraft and hangar status are realized, significantly improving the intelligence and efficiency of the operation process. By defining a detailed state vector and a flexible action space, the system can make accurate scheduling decisions based on the real-time position, speed, fuel level of the aircraft and the task status of the hangar. The reasonable design of the reward function further ensures the efficiency and safety of task allocation. At the same time, the introduction of the state value function and the action value function provides a mathematical basis for the DRL model to learn and make decisions. Through the training of the DQN algorithm, the model has learned how to select the optimal action in various states to minimize the task completion time, maximize safety, and optimize resource utilization. Finally, the deployed DRL model can continuously learn and adapt in a dynamically changing environment, providing a load-balanced scheduling scheme for the aircraft, effectively reducing waiting time, improving overall operation efficiency, and showing strong practical application potential.

[0083] Furthermore, in the process of optimizing task allocation, a mixed-integer programming model is used to handle constraints and optimization problems. The specific content includes:

[0084] S21. Define the decision variables of the MIP model according to the action space of the DRL model. The decision of task allocation for each aircraft is represented as a binary variable , where i represents the index of the aircraft and j represents the index of the task;

[0085] S22. Combine the reward function of the DRL model to construct the objective function of the MIP model. The objective function may aim to minimize the task completion time, improve safety and resource utilization;

[0086] The specific objective function is expressed as: ; where represents the accumulation of costs for all aircraft i and all tasks j, is the weight coefficient of the task completion time, indicating the importance of the task completion time in the total cost, represents the safety index when aircraft i executes task j, is the weight coefficient of the resource utilization rate, indicating the importance of the resource utilization rate in the total cost, represents the resource utilization rate when aircraft i executes task j, is a decision variable, a binary variable, indicating whether aircraft i executes task j (1 means execute, 0 means not execute).

[0087] Specifically, the objective function can be interpreted as: for each pair of aircraft and tasks, calculate a cost weighted combination of task completion time, safety and resource utilization, and then accumulate these costs through the decision variables to find the aircraft-to-task allocation plan that minimizes the total cost Z.

[0088] In practical applications, the objective function will be used in combination with a series of constraints, such as ensuring that each task is exactly assigned to one aircraft, or the flight route and payload constraints of the aircraft, etc., to form a complete MIP model and use a mathematical optimization solver for solution.

[0089] S23. Add the constraint conditions of the MIP model according to the real-time states of the aircraft and the hangar;

[0090] Specifically include: task allocation constraint, each task is exactly assigned to one aircraft, expressed as: ;

[0091] Aircraft capacity constraint, the load of the aircraft shall not exceed its maximum capacity, expressed as: ; where Represents the resource requirements of task j for vehicle i. Represents the resource capacity of vehicle i.

[0092] Time window constraint. The takeoff and landing times of the vehicle must be within a specific time window, which is expressed as: ; where Represents the time when vehicle i starts to execute task j. and Represent the earliest and latest takeoff and landing times of vehicle i respectively.

[0093] Safety constraint. Ensure the safety distance between vehicles or other safety standards, which is expressed as: , where Represents the safety threshold.

[0094] S24. Use a mathematical optimization solver to solve the MIP model and find the optimal solution that satisfies all constraints.

[0095] S25. Analyze the solution results to ensure the feasibility and effectiveness of the solution. Apply the optimization results to the actual vehicle scheduling, and adjust the model parameters according to the actual operation situation to improve future scheduling decisions.

[0096] Specifically, by combining deep reinforcement learning (DRL) with mixed integer programming (MIP), in the process of optimizing task allocation, make full use of the decision-making ability of DRL and the advantages of MIP in dealing with complex constraints. The action space of the DRL model guides the definition of the decision variables of MIP, while the reward function of DRL inspires the construction of the MIP objective function to minimize the task completion time and improve safety and resource utilization. The MIP model ensures the feasibility of the solution through a series of constraint conditions, such as task allocation, vehicle capacity, time window, and safety standards. Use a mathematical optimization solver to solve the MIP model, find the optimal solution that satisfies all constraints, and apply these solutions to the actual vehicle scheduling, so as to achieve load balancing, reduce waiting time, and improve the overall operation efficiency. In addition, by analyzing the solution results and adjusting the model parameters according to the actual operation situation, the scheduling decision can be continuously improved, making the system more adaptable to the dynamically changing operation environment. This method not only improves the intelligence level of task allocation, but also enhances the flexibility and robustness of the system, showing strong application potential in complex air traffic scheduling problems.

[0097] It should be noted that in the application of combining deep reinforcement learning (DRL) with mixed integer programming (MIP), DRL is responsible for learning and exploration in the continuous decision space, while MIP deals with the optimization problem of discrete decisions. Specifically, DRL can be used for preliminary exploration and evaluation of various possible scheduling decisions, which can be continuous, such as flight path selection or task assignment priorities. Then, these continuous decisions can be quantified or discretized to meet the needs of the MIP model. For example, the DRL model can assign a priority score to each task, and these scores are subsequently converted into binary decision variables (such as whether a task is assigned to a specific aircraft). In this way, the soft decisions provided by DRL can be transformed into hard decisions in the MIP model, enabling MIP to optimize task assignment and scheduling strategies while satisfying all discrete constraints. In this way, the flexibility of DRL and the constraint satisfaction ability of MIP are combined to jointly achieve an efficient and reliable intelligent scheduling system.

[0098] Furthermore, the method further includes real-time collision detection: when it is detected that there is a collision risk between the first aircraft and the third aircraft, an avoidance instruction is sent to the third aircraft to enable the third aircraft to bypass the area where the first aircraft is located.

[0099] Embodiment 2

[0100] This embodiment provides an intelligent scheduling system for aircraft jump hangars, which obtains and synchronizes the status information of aircraft and hangars in real time through a data synchronization and communication module; uses a real-time scene analysis module to monitor the hangar scene and transmit aircraft position information; the intelligent scheduling decision module combines deep reinforcement learning (DRL) and mixed integer programming (MIP) to optimize task assignment and flight routes, achieve load balancing, reduce waiting time, improve operation efficiency and resource utilization rate. DRL optimizes the decision-making strategy, and MIP finds the optimal solution under multiple constraint conditions through a mathematical solver to ensure the safe and efficient completion of tasks.

[0101] The specific content includes:

[0102] A data synchronization and communication module, which is used to obtain local aircraft flight operation data in real time, obtain the parking request information of jump aircraft, and conduct data communication with the aircraft control center to synchronize the status of the aircraft and hangar data;

[0103] A real-time scene analysis module, which is used to obtain real-time scene operation data of the hangar, including operation sites, supply operation sites, takeoff and landing operation sites, and idle operation sites, etc., obtain the real-time position information of the first aircraft from the first hangar, and send the information to the second hangar, receive the real-time position information of the third aircraft, and send task instructions;

[0104] The intelligent scheduling decision-making module makes intelligent scheduling decisions through a deep reinforcement learning unit and a mixed integer programming unit, sends task instructions to the aircraft, adjusts the flight route, and ensures the implementation of the scheduling strategy.

[0105] Further, the deep reinforcement learning unit uses the DRL algorithm to optimize task allocation, ensure the load balance of the aircraft, reduce the waiting time, and improve the overall operation efficiency. It trains the DRL model, updates the network weights to select the optimal action value function, deploys the trained DRL model to the scheduling system, and outputs the optimal scheduling decision in real time according to the current state.

[0106] The mixed integer programming unit represents the decision of task allocation to each aircraft as binary variables by defining the decision variables of the MIP model, constructs the objective function of the MIP model, aiming to minimize the task completion time, improve safety and resource utilization, adds the constraint conditions of the MIP model, such as task allocation constraints, aircraft capacity constraints, time window constraints, safety constraints, etc., and then uses a mathematical optimization solver to solve the MIP model to find the optimal solution that satisfies all constraints.

[0107] The above are only the preferred embodiments of the present invention, and do not impose any form of limitation on the present invention. Although the present invention has been disclosed above with preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to equivalent embodiments by using the above-disclosed technical content within the scope of the technical solution of the present invention. However, as long as it does not depart from the content of the technical solution of the present invention, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. An intelligent scheduling method for an aircraft to jump into an aircraft hangar, characterized in that: It includes the following steps: The hangar obtains the flight operation data of local aircraft in real time; The hangar obtains the parking request information of the skip-flight aircraft in real time, conducts data communication with the aircraft control center in real time according to the parking request information, obtains the flight operation data of the aircraft, and synchronizes the state of the aircraft with the hangar; Obtain the real-time scene operation data of the hangar, and conduct aircraft cluster scheduling according to the real-time scene operation data of the hangar; The scene operation data includes the working sites, supply operation sites, takeoff and landing operation sites, and idle operation sites; Among them, the method of conducting aircraft cluster scheduling according to the real-time scene operation data of the hangar includes: obtaining the real-time position information of the first aircraft from the first hangar; sending the real-time position information of the first aircraft to the second hangar; when the first hangar receives the real-time position information of the third aircraft, sending a task instruction to the third aircraft, where the third aircraft is an aircraft in the second hangar; Optimize the task allocation through deep reinforcement learning, and the specific content includes: Define the state vector s as the current state of the aircraft and the hangar; the state vector s is expressed as: ； Among them, represents the real-time position coordinates of the 1st to the nth aircraft; represents the real-time speed of the 1st to the nth aircraft; represents the remaining fuel of the 1st to the nth aircraft; represents the real-time task status of the hangar; Define action a as a scheduling decision, including , where represents the scheduling action for the i-th aircraft; Design a reward function R to evaluate the quality of each action; Specific , where T(s,a) represents the task completion time, S(s,a) represents the safety index, and U(s,a) represents the resource utilization rate, is the weight coefficient; Define the state value function and the action value function to quantify the value of each state and action; Use the loss function of the DQN algorithm for training, update the network weight θ, and select the optimal action value function; The loss function is expressed as: , where y represents the target Q value; is the value estimation of the current network for taking action a in state s; θ is the current network parameter, represents the expected value and is used to represent the statistical average of the samples; Deploy the trained DRL model to the scheduling system, and output the optimal scheduling decision in real time according to the current state s; According to the action space of the DRL model, define the decision variables of the MIP model. The decision of assigning tasks to each aircraft is represented as a binary variable , where i represents the index of the aircraft and j represents the index of the task; Combine the reward function of the DRL model to construct the objective function of the MIP model; The specific objective function is expressed as: ; among which represents the accumulation of costs for all aircraft i and all missions j, is the weight coefficient of the mission completion time; represents the safety index when aircraft i executes mission j, is the weight coefficient of the resource utilization rate, represents the resource utilization rate when aircraft i executes mission j, is a decision variable; Add the constraint conditions of the MIP model according to the real-time states of the aircraft and the hangar.

2. The intelligent scheduling method for an aircraft jump hangar according to claim 1, wherein: The synchronization of the state of the aircraft with the hangar is specifically used to obtain the position information and task status of the aircraft in real time, generate and display the task allocation table, and the content includes: The aircraft control center receives the first data packet from the first aircraft, and the first data packet carries the current position information and the current executed task information of the first aircraft; receives the second data packet from the third aircraft, and the second data packet carries the current position information and the current executed task information of the third aircraft; generates a task allocation table according to the first data packet and the second data packet, and the task allocation table records the position information and the current executed task information of each aircraft; displays the task allocation table.

3. The intelligent scheduling method of an aircraft jump hangar according to claim 1, characterized in that: The flight operation data includes at least one of the following: the position information of the aircraft, the flight route, the takeoff and landing time, the flight payload, the load information, the heading information, the speed information, the remaining fuel information, and the status information.

4. The intelligent scheduling method for an aircraft jump hangar according to claim 1, characterized in that: Before obtaining the real-time position information of the first aircraft from the first hangar, it further includes: receiving a task request message from the user terminal, and the task request message carries the identification information of the first aircraft.

5. The intelligent scheduling method of an aircraft jump hangar according to claim 1, characterized in that: In the process of optimizing the task allocation, the mixed integer programming model is used to handle the constraint conditions and optimization problems, and it further includes: Use a mathematical optimization solver to solve the MIP model and find the optimal solution that satisfies all the constraints; Analyze the solution results, apply the optimization results to the actual aircraft scheduling, and adjust the model parameters according to the actual operation situation.

6. The intelligent scheduling method of an aircraft jump hangar according to claim 5, characterized in that: The constraint conditions of the MIP model include: Task assignment constraint, where each task is exactly assigned to one aircraft, expressed as: ； Aircraft capacity constraint, where the load of the aircraft shall not exceed its maximum capacity, expressed as: ; among which represents the resource requirement of task j for vehicle i, represents the resource capacity of vehicle i; Time window constraint, where the take-off and landing times of the aircraft must be within a specific time window, expressed as: ; where represents the time when vehicle i starts to execute mission j, and represent the earliest and latest takeoff and landing times of vehicle i, respectively; Safety constraint, ensuring the safety distance between aircraft or other safety standards, expressed as: , where represents the safety threshold.

7. The intelligent scheduling method for an aircraft jump hangar according to claim 1, characterized in that: The method further includes real-time collision detection: when a collision risk is detected between the first aircraft and the third aircraft, a avoidance instruction is sent to the third aircraft to enable the third aircraft to bypass the area where the first aircraft is located.

8. An intelligent scheduling system for an aircraft jump hangar, which is applied to an intelligent scheduling method for an aircraft jump hangar according to any one of claims 1-7, characterized in that: It includes a data synchronization and communication module, a real-time scenario analysis module, and an intelligent scheduling decision-making module. The data synchronization and communication module is used to obtain the local aircraft flight operation data in real time, obtain the shutdown request information of the skip aircraft, and perform data communication with the aircraft control center to synchronize the status of the aircraft and the hangar data. The real-time scenario analysis module is used to obtain the real-time scenario operation data of the hangar, obtain the real-time position information of the first aircraft from the first hangar, and send the information to the second hangar, receive the real-time position information of the third aircraft, and send task instructions. The intelligent scheduling decision-making module makes intelligent scheduling decisions through a deep reinforcement learning unit and a mixed integer programming unit, sends task instructions to the aircraft, and adjusts the flight route.

9. An intelligent scheduling system for aircraft skip hangar according to claim 8, characterized in that: The deep reinforcement learning unit uses the DRL algorithm to optimize task assignment, trains the DRL model, updates the network weights to select the optimal action value function, deploys the trained DRL model to the scheduling system, and outputs the optimal scheduling decision in real time according to the current state. The mixed integer programming unit defines the decision variables of the MIP model, represents the decision of task assignment to each aircraft as a binary variable, constructs the objective function of the MIP model, adds the constraint conditions of the MIP model, and then uses a mathematical optimization solver to solve the MIP model to find the optimal solution that satisfies all constraints.

Images