Task scheduling method and device, equipment and storage medium

Abstract

The invention discloses a task scheduling method and device, equipment and a storage medium, and belongs to the technical field of flight maintenance task scheduling. The method comprises the following steps: collecting flight dynamic data, personnel qualification data, personnel state data and maintenance task information, and generating an input data set; based on a mixed integer nonlinear programming model, multi-objective optimization is carried out on the input data set, an initial task scheduling scheme is generated, and multiple objectives comprise task delay minimization, personnel work equalization, personnel moving time minimization and personnel fatigue minimization; when the flight dynamic data, the personnel state data or the maintenance task information are detected to change, generating a task rescheduling scheme based on a dynamic planning algorithm and a simulated annealing algorithm; pushing the task rescheduling scheme to maintenance personnel, and receiving task execution feedback data; and based on the task execution feedback data, a task scheduling strategy is updated through a deep Q network algorithm. According to the embodiment of the invention, the task scheduling efficiency and accuracy can be improved.

Description

Technical Field

[0001] This application relates to the technical field of route maintenance task scheduling, and particularly to a task scheduling method, device, equipment and storage medium. Background Art

[0002] Task scheduling refers to the allocation of tasks to available resources (such as personnel, equipment, etc.) under specific constraints to optimize certain objectives (such as efficiency, cost, time, etc.). Currently, the scheduling of route maintenance tasks mainly adopts the method of manual scheduling. In this mode, the workshop shift schedulers need to closely monitor the arrival and departure information of flights on the same day, and according to the professional qualifications of maintenance personnel such as release and signing cards, arrange the maintenance personnel who meet the work qualification requirements to the designated apron in the form of real-time message instructions in the work group, and require them to perform route maintenance tasks within the specified time. However, this traditional manual scheduling method has low efficiency and accuracy. Summary of the Invention

[0003] The purpose of the embodiments of this application is to provide a task scheduling method, device, equipment and storage medium, which can improve the efficiency and accuracy of task scheduling.

[0004] To achieve the above purpose, the first aspect of the embodiments of this application provides a task scheduling method, including:

[0005] Collect flight dynamic data, personnel qualification data, personnel status data and maintenance task information, and generate an input data set;

[0006] Based on a mixed integer non-linear programming model, perform multi-objective optimization on the input data set to generate an initial task scheduling plan, where the multi-objectives include minimizing task delay, equalizing personnel work, minimizing personnel movement time, and minimizing personnel fatigue;

[0007] When it is detected that the flight dynamic data, the personnel status data or the maintenance task information changes, generate a task rescheduling plan based on the dynamic programming algorithm and the simulated annealing algorithm;

[0008] Push the task rescheduling plan to the maintenance personnel and receive task execution feedback data;

[0009] Based on the task execution feedback data, update the task scheduling strategy corresponding to the task rescheduling plan through the deep Q-network algorithm.

[0010] Compared with the prior art, a task scheduling method provided by an embodiment of the present application has the following beneficial effects: By collecting flight dynamic data, personnel qualification data, personnel status data, and maintenance task information to generate an input data set, the present application solves the problems of low data collection efficiency and poor accuracy in traditional manual scheduling; based on a mixed-integer non-linear programming model, multi-objective optimization is performed on the input data set to generate an initial task scheduling plan, which solves the problem that traditional methods are difficult to balance task delays, personnel workload, travel time, and fatigue, and realizes efficient task completion and optimized resource allocation; a task re-scheduling plan is generated through a dynamic programming algorithm and a simulated annealing algorithm, which solves the problem that traditional methods cannot respond in real time to flight delays, apron changes, and personnel unavailability, and significantly improves the dynamic adjustment ability; the task re-scheduling plan is pushed to maintenance personnel and feedback data is received, which solves the problems of low efficiency and easy omission in traditional manual notifications, and ensures the timeliness and accuracy of information transmission; based on the feedback data, the task scheduling strategy is updated through a deep Q-network algorithm, which solves the problem that traditional methods lack self-learning ability, and realizes the continuous optimization of the scheduling strategy and the improvement of the system's adaptability. Finally, the present application realizes the efficiency, dynamics, and intelligence of task scheduling, significantly reduces task delays, and improves the efficiency and accuracy of task scheduling.

[0011] In some embodiments, the multi-objective optimization of the input data set based on the mixed-integer non-linear programming model to generate an initial task scheduling plan includes:

[0012] Construct a mixed-integer non-linear programming model, define the decision variables, objective function, and constraint conditions of the mixed-integer non-linear programming model, and use a meta-heuristic algorithm to solve the mixed-integer non-linear programming model to generate a Pareto optimal solution set;

[0013] Select a balanced solution from the Pareto optimal solution set to obtain an initial task scheduling plan.

[0014] In some embodiments, the construction process of the mixed-integer non-linear programming model includes the following steps:

[0015] (1) Determine the decision variables: y ik , t j , t1, t2, t3, and t4;

[0016] Among them, y ik is a binary variable, y ik = 1 means that the i-th maintenance personnel is responsible for the j-th task (i.e., task j), and y ij = 0 means that the i-th maintenance personnel is not responsible for the j-th task; t j$t_{j}$ is the start time of the $j$-th task; $t_1$ is the pick-up time; $t_2$ is the inspection time; $t_3$ is the start time of the guardianship task; $t_4$ is the drop-off time;

[0017] (2) Determine the objective function, which includes:

[0018] Minimization of task delay:

[0019] Balancing of personnel workload:

[0020] Minimization of personnel movement time: $\min\sum$ i,j,k $(y_{ik}$ ij $\cdot y_{jk}$ ik $\cdot d_{jk})$ jk )

[0021] And minimization of personnel fatigue: $\min\sum$ i $\text{fatigue}(T_i)$ i );

[0022] Where $t_j$ j is the start time of the $j$-th task, $\hat{t}_j$ represents the planned start time of task $j$; $T_i$ i represents the total working time of the $i$-th maintenance personnel, $\bar{T}$ is the average working time; $y_{ik}$ ik is a binary variable, $y_{ik}$ ik $ = 1$ means the $i$-th maintenance personnel is responsible for the $k$-th task (i.e., task $k$), $y_{ik}$ ik $ = 0$ means the $i$-th maintenance personnel is not responsible for the $k$-th task, $d_{jk}$ jk is the aircraft position distance between task $j$ and task $k$; $\text{fatigue}(T_i)$ i is the fatigue calculated according to the total working time $T_i$ i of the $i$-th maintenance personnel, $h$ is the initial fatigue, and $\lambda$ determines the growth rate of fatigue with working time;

[0023] (3) Determine the constraint conditions, which include:

[0024] Task timing constraint: $t_1 = t_j - 15$, j $t_2\geq t_1 + 5$,

[0025] $t_3\geq t_2+(20\delta_1 + 25\delta_2)$,

[0026] $t_4\geq t_j - 15$,

[0027] Personnel qualification constraint: d $t_4\geq t_j - 15$,

[0028] Personnel qualification constraint:

[0029]

[0030] ∑iyi3 = 1,

[0031]

[0032] Moving time constraint: t move +t j ≤t k ,

[0033] And working hours constraint: T i ≤T max , T rest ≥30;

[0034] Wherein, t j is the task start time of the j-th task, t1 is the pick-up time, t2 is the inspection time, t3 is the start time of the guardianship task, t4 is the drop-off time, t d is the scheduled take-off time, δ1 represents that there are 2 maintenance personnel performing the inspection task, δ2 represents that there is 1 maintenance personnel performing the inspection task, δ1 and δ2 are binary variables, and δ1 + δ2 = 1, indicating that there is 1 or 2 maintenance personnel performing the inspection task; y i1 is the pick-up task, is the signing card qualification; y i2 is the inspection task, is the release qualification; y i3 is the guardianship task; y i4 is the drop-off task; t move is the moving time, t k is the task start time of the k-th task; T max is the maximum working limit; T rest is the rest time;

[0035] (4) Solve the mixed integer non-linear programming model by using a meta-heuristic algorithm to generate a Pareto optimal solution set.

[0036] In some embodiments, the collecting flight dynamic data, personnel qualification data, personnel status data, and maintenance task information to generate an input data set includes:

[0037] Collect maintenance task information, flight dynamic data of the flight scheduling system, personnel qualification data and personnel status data of the maintenance personnel management system in real time through the API interface; among them, the flight dynamic data includes the planned arrival time, actual arrival time, parking position number and task type of the flight; the personnel qualification data includes signing card qualification and release qualification, and the personnel status data includes the current position, workload and fatigue degree; the maintenance task information includes task type, task priority and task execution time requirement;

[0038] Structurally process the flight dynamic data, the personnel qualification data, the personnel status data and the maintenance task information to generate an input data set.

[0039] In some embodiments, when it is detected that the flight dynamic data, the personnel status data or the maintenance task information changes, generate a task rescheduling plan based on the dynamic programming algorithm and the simulated annealing algorithm, including:

[0040] Real-time detect the flight dynamic data, the personnel status data and the maintenance task information through the event-driven architecture. When flight delay, parking position change, personnel unavailability or task time change is detected, trigger task rescheduling;

[0041] Optimize the current task assignment status based on the dynamic programming algorithm to generate a preliminary rescheduling plan;

[0042] Use the simulated annealing algorithm to globally optimize the preliminary rescheduling plan to generate a task rescheduling plan.

[0043] In some embodiments, push the task rescheduling plan to the maintenance personnel and receive task execution feedback data, including:

[0044] Push the task rescheduling plan to the maintenance personnel through mobile APP notification, SMS, communication tool message notification or voice call notification, and receive task confirmation information;

[0045] Receive the task execution status data fed back by the maintenance personnel, and the task execution status data includes the task start time, task completion time and task execution result.

[0046] In some embodiments, based on the task execution feedback data, update the task scheduling strategy corresponding to the task rescheduling plan through the deep Q-network algorithm, including:

[0047] Construct a deep Q-network model. The state space of the model includes flight dynamics, maintenance personnel availability, workload, personnel qualifications and fatigue status, and the action space of the model includes task assignment, personnel adjustment and task priority adjustment;

[0048] Design the reward function of the deep Q-network model, where the reward function is designed based on task completion and personnel utilization rate. A positive reward is obtained when the task is completed on time, and a negative reward is obtained when the task is delayed.

[0049] Based on the task execution feedback data, update the Q-value table of the deep Q-network model through the experience replay mechanism.

[0050] Update the policy network parameters of the deep Q-network model through the target network to generate an updated task scheduling policy.

[0051] To achieve the above object, a second aspect of the embodiments of the present application provides a task scheduling device, which includes:

[0052] An acquisition module, configured to acquire flight dynamic data, personnel qualification data, personnel status data, and maintenance task information, and generate an input data set.

[0053] An optimization module, configured to perform multi-objective optimization on the input data set based on a mixed-integer nonlinear programming model to generate an initial task scheduling plan, where the multi-objectives include minimizing task delay, equalizing personnel work, minimizing personnel movement time, and minimizing personnel fatigue.

[0054] A detection module, configured to generate a task rescheduling plan based on the dynamic programming algorithm and the simulated annealing algorithm when it detects that the flight dynamic data, the personnel status data, or the maintenance task information changes.

[0055] A receiving module, configured to push the task rescheduling plan to maintenance personnel and receive task execution feedback data.

[0056] An update module, configured to update the task scheduling policy corresponding to the task rescheduling plan through the deep Q-network algorithm based on the task execution feedback data.

[0057] To achieve the above object, a third aspect of the embodiments of the present application proposes an electronic device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the method described in the first aspect above is implemented.

[0058] To achieve the above object, a fourth aspect of the embodiments of the present application proposes a computer-readable storage medium, which includes a stored computer program. When the computer program runs, it controls the device where the computer-readable storage medium is located to execute the method described in the first aspect above.

[0059] To achieve the above object, a fifth aspect of the embodiments of the present application provides a computer program product, which includes a computer program or computer instructions. When the computer program or the computer instructions are executed by a processor, the method described in the first aspect above is implemented. Description of the Drawings

[0060] Figure 1 is a flowchart of a task scheduling method provided by an embodiment of the present application;

[0061] Figure 2 is Figure 1 a flowchart of step S101 in

[0062] Figure 3 is Figure 1 a flowchart of step S102 in

[0063] Figure 4 is Figure 1 a flowchart of step S103 in

[0064] Figure 5 is Figure 1 a flowchart of step S104 in

[0065] Figure 6 is Figure 1 a flowchart of step S105 in

[0066] Figure 7 is a schematic structural diagram of a task scheduling device provided by an embodiment of the present application;

[0067] Figure 8 is a schematic hardware structure diagram of an electronic device provided by an embodiment of the present application. Detailed Embodiments

[0068] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0069] In the description of the present application, it should be understood that the orientation or positional relationship indicated by terms such as "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present application.

[0070] The terms "first" and "second" are only used for descriptive purposes and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present application, unless otherwise specified, the meaning of "a plurality" is two or more.

[0071] In the description of the present application, it should be noted that unless otherwise clearly specified and limited, the terms "install", "connect", and "couple" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances.

[0072] First, several terms involved in the present application are analyzed:

[0073] The scheduling optimization algorithm in operations research: It is a class of algorithms used to solve how to reasonably arrange the order and time of tasks or activities under limited resources to achieve the optimal goal (such as minimizing cost, maximizing efficiency, etc.).

[0074] Genetic Algorithm (GA): This is a stochastic search algorithm that simulates the biological evolution process in nature. It encodes the solutions of the scheduling problem as chromosomes and gradually searches for the optimal solution through operations such as selection, crossover, and mutation. The genetic algorithm has strong global search capabilities and is suitable for solving complex scheduling problems. For example, in large-scale production scheduling, in the face of numerous orders, machines, and processes, the genetic algorithm can search for the optimal scheduling plan in a huge solution space.

[0075] The multi-objective optimization algorithm based on the genetic algorithm: It is a class of algorithms that improve and expand the traditional genetic algorithm for multi-objective optimization problems, and can simultaneously optimize multiple conflicting objectives to find a set of Pareto optimal solutions (a set of solutions where there is no solution that can be better than other solutions in all objectives among multiple objectives).

[0076] Reinforcement Learning: A type of machine learning algorithm where an agent interacts with the environment and learns the optimal behavior strategy based on the reward signals feedback from the environment.

[0077] Heuristic Algorithms: A class of algorithms based on experience, intuition, or rules, used to find approximate or satisfactory solutions to problems within a reasonable time, without necessarily pursuing the optimal solution. Different from exact algorithms (such as the simplex method in linear programming, which can guarantee finding the optimal solution but may have excessive computational complexity for complex problems), heuristic algorithms sacrifice a certain degree of optimality of the solution in exchange for more feasible computational time and resource consumption in practical applications.

[0078] Dynamic Programming Algorithm (DP): A phased decision-making optimization algorithm that decomposes complex problems into sub-problems and solves them step by step, suitable for task reallocation problems with time dependence.

[0079] Simulated Annealing Algorithm (SA): The simulated annealing algorithm is derived from the simulation of the solid annealing process. It controls the temperature parameter and accepts worse solutions with a certain probability during the search process to avoid getting trapped in local optimal solutions. As the temperature gradually decreases, the algorithm finally converges to the global optimal solution. In vehicle scheduling problems, the simulated annealing algorithm can be used to find the optimal solutions for vehicle driving routes and task allocations.

[0080] Deep Q-Network Algorithm (DQN) is an algorithm that combines deep learning and reinforcement learning, mainly used to solve complex decision-making problems. DQN is based on the classic reinforcement learning algorithm Q-Learning. The core of Q-Learning is to learn a Q function that evaluates the expected long-term cumulative reward obtained after taking a certain action in a specific state. In Q-Learning, the agent selects actions according to the current state at each time step, observes the new state and the obtained reward after executing the action, and uses specific update rules to update the Q value. DQN introduces a deep neural network (usually a multi-layer perceptron or a convolutional neural network) to approximate the Q function. By taking the state as the input of the neural network, it outputs the Q value estimates for each possible action. In this way, even in the face of a high-dimensional state space (such as complex data like images and videos), DQN can effectively learn the Q function.

[0081] Mixed-Integer Nonlinear Programming (MINLP) model: It is a mathematical optimization model used to handle optimization problems that simultaneously contain discrete variables (such as integer variables, usually represented as decision variables of 0 and 1, used to represent whether to select a certain option, perform a certain task, etc.) and continuous variables (variables that can take any real value within a certain range), and the objective function and / or constraint conditions are non-linear functions.

[0082] Currently, the scheduling of line maintenance personnel in the aircraft maintenance system mainly adopts the manual scheduling method. Under this mode, the workshop schedulers need to closely monitor the arrival and departure information of flights on the same day. According to the professional qualifications of maintenance personnel such as release and signing cards, in the form of real-time message instructions in the work group, the maintenance personnel meeting the work qualification requirements are arranged to the designated apron and required to perform line maintenance tasks within the specified time.

[0083] However, there are many drawbacks to this traditional manual scheduling method. On the one hand, schedulers need to spend a lot of time and energy, and the efficiency and accuracy of scheduling are both low. On the other hand, in the face of the dynamic changes of flight information, it is difficult to make effective responses quickly. Once the task scheduling is not timely, it is very easy to cause delays in maintenance work, thus affecting the normal progress of the overall aircraft maintenance work.

[0084] Therefore, how to improve the efficiency and accuracy of task scheduling has become a technical problem to be solved urgently.

[0085] Please refer to Figure 1 , Figure 1 which is an optional flowchart of the task scheduling method provided by the embodiments of this application. Figure 1 The method in

[0086] Step S101: Collect flight dynamic data, personnel qualification data, personnel status data, and maintenance task information to generate an input data set.

[0087] Step S102: Based on the mixed-integer nonlinear programming model, perform multi-objective optimization on the input data set to generate an initial task scheduling plan. The multi-objectives include minimizing task delays, equalizing personnel work, minimizing personnel movement time, and minimizing personnel fatigue.

[0088] Step S103: When it is detected that the flight dynamic data, personnel status data, or maintenance task information has changed, generate a task rescheduling plan based on the dynamic programming algorithm and the simulated annealing algorithm.

[0089] Step S104: Push the task rescheduling plan to the maintenance staff and receive task execution feedback data;

[0090] Step S105: Based on the task execution feedback data, update the task scheduling strategy corresponding to the task rescheduling plan through the deep Q-network algorithm.

[0091] Steps S101 to S105 shown in the embodiments of the present application generate an input data set by collecting flight dynamic data, personnel qualification data, personnel status data, and maintenance task information, solving the problems of low data collection efficiency and poor accuracy in traditional manual scheduling; perform multi-objective optimization on the input data set based on the mixed integer non-linear programming model to generate an initial task scheduling plan, solving the problem that it is difficult for traditional methods to balance task delays, personnel workload, travel time, and fatigue, and realizing the efficient completion of tasks and the optimal allocation of resources; generate a task rescheduling plan through the dynamic programming algorithm and the simulated annealing algorithm, solving the problem that traditional methods cannot respond in real time to flight delays, apron changes, and personnel unavailability, and significantly improving the dynamic adjustment ability; push the task rescheduling plan to the maintenance staff and receive feedback data, solving the problems of low efficiency and easy omission in traditional manual notifications, and ensuring the timeliness and accuracy of information transmission; update the task scheduling strategy through the deep Q-network algorithm based on the feedback data, solving the problem that traditional methods lack self-learning ability, and realizing the continuous optimization of the scheduling strategy and the improvement of the system's adaptive ability. Finally, the present application realizes the efficiency, dynamics, and intelligence of task scheduling, significantly reducing task delays and improving the efficiency and accuracy of task scheduling.

[0092] In step S101 of some embodiments, the flight dynamic data may include the planned arrival time of the flight, the actual arrival time, the apron number, and the task type (such as pick-up, inspection, guardianship, send-off, etc.). The personnel qualification data may include the qualification types of the maintenance staff (such as signature card qualification, release qualification, etc.). The personnel status data may include the current location of the maintenance staff, the workload (such as the number of assigned tasks), the fatigue degree (such as the continuous working time), etc. The maintenance task information may include the task type (such as pick-up, inspection, guardianship, send-off), the task priority, and the task execution time requirement. The input data set may be a data set for scheduling optimization generated after structuring the above data.

[0093] Please refer to Figure 2 In some embodiments, step S101 may include but is not limited to steps S201 to S202:

[0094] Step S201: Real-time collect maintenance task information, flight dynamic data of the flight scheduling system, personnel qualification data, and personnel status data of the maintenance staff management system through the API interface;

[0095] Step S202: Structurally process the flight dynamic data, personnel qualification data, personnel status data, and maintenance task information to generate an input data set.

[0096] In step S201 of some embodiments, the API interface can be an application programming interface for data interaction between different systems. The flight dynamic data includes the planned arrival time of the flight, the actual arrival time, the parking bay number, and the task type (such as picking up the plane, inspection, guardianship, sending off the plane, etc.); the personnel qualification data includes the signing card qualification (permission to sign maintenance documents) and the release qualification (permission to release the plane), and the personnel status data includes the current location, the workload (current task volume), and the fatigue level (the degree of fatigue caused by the work intensity); the maintenance task information includes the task type (such as picking up the plane, inspection, guardianship, sending off the plane), the task priority (such as urgent, routine), and the task execution time requirement (such as must be completed 15 minutes before takeoff). Real-time data is obtained from the flight scheduling system and the maintenance personnel management system through the API interface.

[0097] In step S202 of some embodiments, the structural processing can be to organize and standardize the collected raw data according to a predefined format to generate a normalized data set that can be used for model input. The input data set can be a data set generated after the structural processing of the flight dynamic data, personnel qualification data, personnel status data, and maintenance task information, and is used for the input of the task scheduling model.

[0098] In the embodiments of the present application, flight dynamic data, personnel qualification data, personnel status data, and maintenance task information are collected in real time through the API interface, and the data is structurally processed to generate a high-quality input data set. This process solves the problems of low efficiency and poor accuracy in traditional manual data collection, provides a reliable data basis for the optimization of the subsequent task scheduling model, and significantly improves the timeliness and accuracy of information.

[0099] In step S102 of some embodiments, minimize task delays: Ensure that tasks are completed within the specified time to reduce delays, especially for the timeliness of key tasks such as picking up the plane and sending off the plane. Equalize personnel workload: Balance the workload of maintenance personnel to avoid overwork of some personnel. Minimize personnel movement time: Reduce the movement time of maintenance personnel between different parking bays to improve efficiency. Minimize personnel fatigue: Through reasonable task allocation, reduce the fatigue level of maintenance personnel to ensure maintenance quality.

[0100] Please refer to Figure 3 , in some embodiments, step S102 may include but is not limited to steps S301 to S302:

[0101] Step S301, construct a mixed-integer non-linear programming model, define the decision variables, objective function, and constraint conditions of the mixed-integer non-linear programming model, and use a meta-heuristic algorithm to solve the mixed-integer non-linear programming model to generate a Pareto optimal solution set;

[0102] Step S302, select a balanced solution from the Pareto optimal solution set to obtain an initial task scheduling plan.

[0103] In step S301 of some embodiments, the construction process of the mixed-integer non-linear programming model includes the following steps:

[0104] (1) Determine the decision variables: y ij , t j , t1, t2, t3, and t4;

[0105] Among them, y ij is a binary variable, y ij = 1 indicates that the i-th maintenance personnel is responsible for the j-th task (i.e., task j), and y ij = 0 indicates that the i-th maintenance personnel is not responsible for the j-th task; the task types include aircraft reception, inspection, guardianship, and aircraft departure. The j-th task can be any one of the aircraft reception task, inspection task, guardianship task, and aircraft departure task, and no specific limitation is made here; t j is the start time of the j-th task; t1 is the aircraft reception time, and t1 should start 15 minutes before the arrival of the aircraft; t2 is the inspection time, and the inspection task must be completed after the aircraft reception task ends and before the aircraft departure task starts; t3 is the start time of the guardianship task; t4 is the aircraft departure time, and t4 starts 15 minutes before the takeoff of the aircraft.

[0106] (2) Determine the objective function. The objective function includes minimizing task delay, equalizing personnel work, minimizing personnel movement time, and minimizing personnel fatigue, specifically including:

[0107] Minimizing task delay:

[0108] Equalizing personnel work:

[0109] Minimizing personnel movement time: min∑ i,j,k (y ij ·y ik ·d jk ),

[0110] Minimizing personnel fatigue: min∑ i fatigue(T i );

[0111] Among them, t jis the start time of the j-th task, represents the planned start time of task j; T i represents the total working time of the i-th maintenance personnel, is the average working time; y ik is a binary variable, y ik = 1 indicates that the i-th maintenance personnel is responsible for the k-th task (i.e., task k), y ik = 0 indicates that the i-th maintenance personnel is not responsible for the k-th task, d jk is the aircraft position distance between task j and task k; fatigue(T i ) is the fatigue degree calculated according to the total working time T i of the i-th maintenance personnel, h is the initial fatigue, and λ determines the growth rate of fatigue with working time. The fatigue degree increases exponentially with working time.

[0112] (3) Determine the constraint conditions, which include task time sequence constraints, personnel qualification constraints, movement time constraints, and working duration constraints. Specifically, they include:

[0113] Task time sequence constraints: t1 = t k - 15. The aircraft receiving time t1 must be 15 minutes before the aircraft arrives, and the aircraft receiving task takes 5 minutes.

[0114] t2 ≥ t1 + 5. The inspection time t2 can be carried out after the aircraft receiving task, but must be completed before the aircraft sending task.

[0115] t3 ≥ t2 + (20δ1 + 25δ2). The guardianship time t3 starts after the inspection task ends until the start time of the aircraft sending task. The inspection time is 20 minutes (for 2 people) or 25 minutes (for 1 person) according to the number of people. Here, δ1 and δ2 are binary variables, and δ1 + δ2 = 1.

[0116] t4 ≥ t d - 15. The aircraft sending task t4 starts 15 minutes before takeoff, and all tasks must be completed before that.

[0117] Among them, t j is the start time of the j-th task, t1 is the aircraft receiving time, t2 is the inspection time, t3 is the start time of the guardianship task, t4 is the aircraft sending time, t d is the planned takeoff time, δ1 indicates that there are 2 maintenance personnel performing the inspection task, δ2 indicates that there is 1 maintenance personnel performing the inspection task. δ1 and δ2 are binary variables, and δ1 + δ2 = 1, indicating that the number of maintenance personnel performing the inspection task can only be 1 or 2;

[0118] Personnel qualification constraints:

[0119]

[0120] ∑ i y i3 = 1,

[0121]

[0122] where y i1 is the pick-up task, is the sign-in qualification; y i2 is the inspection task, is the release qualification; y i3 is the guardianship task; y i4 is the drop-off task;

[0123] Moving time constraint: t move + t j ≤ t k where t move is the moving time, t k is the start time of the k-th task;

[0124] Working hours constraint: T i ≤ T max T rest ≥ 30; where T max is the maximum working limit; T rest is the rest time.

[0125] (4) Use a metaheuristic algorithm to solve the mixed-integer nonlinear programming model and generate a Pareto optimal solution set. Solving the above scheduling model belongs to the NP-Hard problem and it is difficult to find an exact optimal solution within polynomial time. To effectively solve this problem, a metaheuristic algorithm can be used to obtain a balanced solution. Among them, NSGA-III (the third generation of non-dominated sorting genetic algorithm) is a multi-objective optimization algorithm based on genetic algorithms, which has significant advantages in dealing with complex high-dimensional objective problems and can generate the Pareto optimal front.

[0126] The NSGA-III algorithm introduces a reference point-based archiving and selection strategy, which makes it perform well in solving optimization problems containing complex, nonlinear, and multiple conflicting objectives. In the flight line maintenance scheduling problem, there are multiple conflicting optimization objectives such as minimizing task delays, equalizing personnel workloads, minimizing personnel moving times, and minimizing personnel fatigue. Taking these objectives as the input of the NSGA-III algorithm, the algorithm can output a Pareto optimal solution set after running. Each solution in this solution set is non-dominated in multiple objectives, that is, there is no solution that is better than other solutions in all objectives.

[0127] The mixed-integer non-linear programming model of the embodiments of the present application not only considers the task time and personnel qualifications, but also introduces key factors such as personnel movement time and workload balance to ensure that tasks are completed on time, personnel do not work overloaded, and personnel movement and fatigue are minimized.

[0128] In step S302 of some embodiments, a balanced solution refers to finding a reasonable compromise among multiple conflicting goals so that each goal can be satisfied to a certain extent, rather than only pursuing the optimal value of a certain goal. Based on the Pareto optimal solution set, the user can flexibly select the balanced solution between different goals from the Pareto optimal solution set according to actual needs to generate an initial task scheduling plan to meet specific scheduling requirements and achieve more reasonable and efficient task scheduling.

[0129] In step S103 of some embodiments, the change of flight dynamic data may include the change of flight arrival and departure times, flight delays or cancellations, and apron changes; the change of personnel status data may include the change of personnel status information, such as task progress, fatigue, location information, etc.; the change of maintenance task information may include task delays, task priority changes, task time changes, and so on.

[0130] Please refer to Figure 4 , in some embodiments, step S103 may include but is not limited to steps S401 to S403:

[0131] Step S401, real-time detect flight dynamic data, personnel status data and maintenance task information through an event-driven architecture. When flight delays, apron changes, personnel unavailability or task time changes are detected, trigger task rescheduling;

[0132] Step S402, optimize the current task assignment status based on the dynamic programming algorithm to generate a preliminary rescheduling plan;

[0133] Step S403, globally optimize the preliminary rescheduling plan using the simulated annealing algorithm to generate a task rescheduling plan.

[0134] In step S401 of some embodiments, the event-driven architecture (EDA) is a software architecture pattern that triggers system behavior by real-time detecting and responding to events (such as flight delays, apron changes). Flight delay can mean that the actual arrival time of the flight is later than the planned arrival time. Apron change can mean that the originally scheduled apron for the flight has changed due to reasons. Personnel unavailability can mean that maintenance personnel cannot perform tasks due to unexpected situations (such as illness, task conflicts).

[0135] In the embodiments of the present application, data docking is performed with the flight scheduling system and the maintenance staff management system through the API interface to obtain information such as flight arrival and departure times, flight delays or cancellations, and apron position changes in real time. At the same time, staff status information such as task progress, fatigue level, and location information is collected in real time. Meanwhile, task information such as task delays, task priority changes, and task time changes is obtained in real time to ensure that corresponding task adjustments are triggered at the first moment of data changes. When flight delays, staff status changes, or task information changes occur, a framework combining EDA and Kafka distributed stream processing is adopted, enabling the system to perform low-latency data processing in a high-concurrency environment and ensuring that changes in flight and staff status can be promptly fed back to the scheduling engine and trigger real-time task scheduling.

[0136] In step S402 of some embodiments, the preliminary rescheduling plan may be a preliminary task reassignment plan generated by a dynamic programming algorithm based on the current task assignment status.

[0137] In some embodiments, the state space of dynamic programming is defined, including the current task assignment status, availability, and task dependencies of maintenance staff. The task assignment is gradually optimized through the state transition equation to generate a preliminary rescheduling plan, ensuring that tasks are completed on time and minimizing interference with the original schedule.

[0138] When flight delays, staff status changes, or task information changes occur, the flight maintenance sequence is optimized according to the time dependence of the flight through the DP algorithm to ensure that time-sensitive flight maintenance tasks can be completed on time. In the dynamic scheduling problem, the state S(i,j) represents the state when the i-th maintenance staff is responsible for the j-th task, including the current task start time, availability, and previously assigned task situation of the staff. The state transition equation describes the way to transfer from the current state S(i,j) to the next state S(i + 1,j + 1). In each step of decision-making in dynamic programming, the maintenance staff i is selected to execute the current task j, and the state transition equation is as follows:

[0139]

[0140] In each state transition, dynamic programming will attempt to assign the optimal maintenance staff i to execute the task j and minimize the actual execution time t j and the planned time the difference between them.

[0141] The design of the objective function is mainly to minimize the actual start time t of the maintenance staff (from the i-th to the m-th) executing the tasks (from the 1st task to the n-th task) j and the planned time the deviation between them, and try to ensure the efficient utilization of maintenance staff resources. y ij is a binary variable, yij = 1 indicates that the i-th maintenance staff is responsible for the j-th task (i.e., task j), and y ij = 0 indicates that the i-th maintenance staff is not responsible for the j-th task; the formula of the objective function is as follows:

[0142]

[0143] In step S403 of some embodiments, with the preliminary rescheduling plan as the initial solution, the simulated annealing algorithm is used to search in the global solution space. The search range is controlled by the temperature parameter, and it gradually converges to the optimal solution to generate the final task rescheduling plan.

[0144] Simultaneously with step S402, when tasks need to be reallocated, the simulated annealing algorithm generates a new task allocation plan through global search and local optimization to balance resource utilization and task delays. The objective function of task rescheduling is as follows:

[0145]

[0146] Among them, c ij represents the cost of assigning maintenance staff i to task j. This algorithm ensures that tasks are quickly assigned during real-time flight changes and minimizes the impact on other tasks. In addition, the embodiments of the present application dynamically maintain the current status of each maintenance staff through the resource pool, including whether they are performing tasks, resting, available, and their qualification information. The resource pool will automatically select appropriate personnel to participate in tasks according to the priority changes of tasks. This mechanism effectively avoids task delays and ensures the reasonable allocation of personnel.

[0147] In addition, the embodiments of the present application can also implement scenario analysis and adaptive optimization. By pre-deducting the possible impacts caused by flight changes or personnel status changes in advance, it ensures that the system can perform task scheduling predictively. At the same time, the scheduling decision is continuously optimized through the reinforcement learning mechanism to improve the system's adaptability to complex scenarios.

[0148] First, the embodiments of the present application predict possible emergencies such as flight delays and personnel task delays based on historical data and the current flight plan, and generate multiple response plans based on different scenarios. And task adjustment plans are generated based on different scenarios, and the optimal strategy is selected through the scenario decision tree. When the system encounters a similar scenario, it quickly selects and implements the corresponding scheduling plan to ensure flexible adjustment of the work schedule.

[0149] After each task execution, the DQN model is used to learn and accumulate task execution feedback, such as task delays and personnel fatigue. According to the completion of the task, the scheduling strategy is adjusted and optimized to gradually improve the system's adaptability and enhance the overall scheduling efficiency. The core formula of the Bellman equation for reinforcement learning is as follows:

[0150]

[0151] Among them, Q(s,a) represents the long-term reward obtained by executing action a in state s, α is the learning rate, γ is the discount factor, r is the immediate reward, and max a′ Q(s′,a′) represents the maximum Q value among all possible actions a′ in the next state s′, that is, the highest reward that can be obtained under the optimal policy. In the embodiments of the present application, through reinforcement learning and scenario analysis, task execution data and scenario processing experience are continuously accumulated. As time goes by, the system will become more and more mature in the task reallocation strategy for complex scenarios, be able to dynamically adapt to different emergencies, and finally achieve automated and intelligent scheduling.

[0152] The present application designs a dynamic adjustment mechanism, through three core functions of real-time data collection, rescheduling of the scheduling plan, scenario analysis and adaptive optimization, to ensure that it can quickly respond to changes in flight plans and the status of maintenance personnel, flexibly handle changes in flights and the status of maintenance personnel, ensure the efficient completion of tasks and the reasonable allocation of resources, and continuously improve the self-adaptability of the system through learning and optimization.

[0153] In step S104 of some embodiments, the task rescheduling plan is pushed to the maintenance personnel through mobile APPs, text messages, and voice notification channels. The task execution status data (such as task start time, completion time, and execution result) feedback by the maintenance personnel is received.

[0154] Please refer to Figure 5 , in some embodiments, step S104 may include but is not limited to steps S501 to S502:

[0155] Step S501, push the task rescheduling plan to the maintenance personnel through mobile APP notifications, text messages, communication tool message notifications, or voice calls, and receive task confirmation information;

[0156] Step S502, receive the task execution status data feedback by the maintenance personnel, and the task execution status data includes the task start time, task completion time, and task execution result.

[0157] In step S501 of some embodiments, Mobile APP Notification: Push the task rescheduling plan through the application installed on the mobile phone of the maintenance personnel. SMS Notification: Send the task rescheduling plan to the mobile phone of the maintenance personnel in the form of SMS. Communication Tool Message Notification: Send the task rescheduling plan to the maintenance personnel through the enterprise instant messaging tool. Voice Call Notification: Inform the maintenance personnel of the task rescheduling plan through the automatic voice call system. Task Confirmation Information: After receiving the task rescheduling plan, the maintenance personnel feedback the confirmation information through the Mobile APP, SMS or communication tool to ensure the accuracy and timeliness of task information transmission. After dynamic adjustment, the system needs to quickly push information such as the latest shift plan, task changes, and emergency notifications to ensure that the personnel performing the tasks can respond in a timely manner.

[0158] In step S502 of some embodiments, Task Execution Status Data: The data fed back by the maintenance personnel during the task execution process, including the task start time, task completion time, and task execution result (such as completed, delayed, failed, etc.).

[0159] The embodiments of the present application push the task rescheduling plan to the maintenance personnel through multiple channels, and receive task confirmation information and task execution status data, improving the coverage rate and timeliness of information transmission. This process solves the problems of low efficiency and easy omission in traditional manual notification, ensures the timeliness and accuracy of task information transmission, and at the same time provides data support for the optimization of task scheduling strategies through a real-time feedback mechanism, significantly improving the transparency and task execution efficiency of the system.

[0160] The embodiments of the present application can ensure that the task adjustment and scheduling results can convey the task update information to relevant personnel in a real-time, reliable, and multi-channel manner, and optimize the task execution process through a feedback loop. The present application can not only achieve efficient information transmission, but also continuously optimize the shift scheduling strategy based on real-time feedback.

[0161] In step S105 of some embodiments, the embodiments of the present application introduce the DQN algorithm to continuously optimize the task assignment and personnel scheduling strategies through real-time feedback. DQN uses a neural network to approximate the Q-value function to handle complex states and dynamic task environments. The present application constructs a state space by real-time detecting flight status, task progress, and personnel information, including factors such as flight dynamics, availability of maintenance personnel, workload, qualifications, and fatigue status. Based on these states, DQN makes decisions on task assignment, personnel adjustment, or task priority in the action space.

[0162] Design a reward function based on task completion and personnel utilization. Completing tasks on time will receive positive rewards, while delays or failures will result in negative rewards, thus guiding the system to make more optimal scheduling decisions. After each scheduling operation, the system obtains the task execution results through a feedback mechanism and updates the Q value according to the rewards. By continuously updating the Q value, DQN adjusts the optimal actions of the system in each state to ensure more efficient future task allocation. And through the experience replay mechanism, randomly sample past task execution experiences during training to break temporal correlations and improve model stability. In addition, a target network is used to improve the stability of the update process and ensure the gradual convergence of the policy. After each task is completed, the system continuously updates the model based on the task results and personnel status feedback, so as to continuously optimize the scheduling strategy.

[0163] Please refer to Figure 6 , in some embodiments, step S105 may include but is not limited to steps S601 to S604:

[0164] Step S601, construct a deep Q-network model. The state space of the model includes flight dynamics, maintenance personnel availability, workload, personnel qualifications, and fatigue status. The action space of the model includes task allocation, personnel adjustment, and task priority adjustment;

[0165] Step S602, design the reward function of the deep Q-network model. The reward function is designed based on task completion and personnel utilization. Positive rewards are obtained when tasks are completed on time, and negative rewards are obtained when tasks are delayed;

[0166] Step S603, update the Q-value table of the deep Q-network model based on the task execution feedback data through the experience replay mechanism;

[0167] Step S604, update the policy network parameters of the deep Q-network model through the target network to generate an updated task scheduling strategy.

[0168] In step S601 of some embodiments, the state space can be the set of states input to the model, including flight dynamics (such as flight delays, gate changes), maintenance personnel availability (such as whether on duty), workload (such as the current task volume), personnel qualifications (such as signing card qualifications, release qualifications), and fatigue status (such as fatigue level). The action space can be the set of actions output by the model, including task allocation (such as allocating tasks to specific personnel), personnel adjustment (such as replacing unavailable personnel), and task priority adjustment (such as promoting the priority of urgent tasks). Constructing a deep Q-network model includes: defining the state space, defining the action space, and initializing the neural network structure of the deep Q-network model, including the input layer (state space), hidden layer (multi-layer neural network), and output layer (action space).

[0169] In step S602 of some embodiments, the reward function can be a function for evaluating the decision-making effect of the model. A positive reward is obtained when the task is completed on time, and a negative reward is obtained when the task is delayed. The reward function is designed based on the task completion situation and personnel utilization rate: a positive reward is obtained when the task is completed on time; a negative reward is obtained when the task is delayed; the personnel utilization rate (such as the balance between workload and fatigue) is used as an additional condition for the reward function.

[0170] In step S603 of some embodiments, the experience replay mechanism can be a training technique that updates the model by randomly sampling historical experience data, breaks the temporal correlation, and improves the model stability. The Q-value table can store the expected cumulative reward values for executing each action in each state, and is used to guide the model decision-making. The present application can update the Q-value table using the above Bellman equation.

[0171] In step S604 of some embodiments, the target network can be an auxiliary network for stabilizing the training process. By periodically updating the policy network parameters, it avoids oscillations during the training process. The policy network parameters can be the neural network parameters in the deep Q-network model for generating decision-making strategies. By periodically updating the policy network parameters through the target network, the stability of the training process is ensured. The gradient descent method is used to optimize the policy network parameters, gradually improving the accuracy of the model decision-making. A updated task scheduling policy is generated for guiding future task allocation and adjustment.

[0172] The present application realizes the continuous optimization of the task scheduling policy by constructing a deep Q-network model, designing a reward function, updating the Q-value table and the policy network parameters. This process solves the problem that traditional scheduling methods lack the ability of self-learning, significantly improves the efficiency, dynamic adaptability and resource utilization rate of task scheduling, and provides an intelligent solution for task scheduling in complex scenarios.

[0173] In some embodiments, the present application further includes a feedback mechanism. The present application can collect information during the task execution process to form a closed-loop feedback system for optimizing future scheduling and decision-making. The feedback not only helps the manager manage the task execution, but also provides data support for the adaptive learning of the system. The present application collects and records the execution status of each task in real time during the task execution process to ensure the effectiveness of the scheduling plan. Through a mobile APP or other tools, maintenance personnel can update the task status and feedback the actual progress and results of the task. The system aggregates and analyzes the feedback data and inputs it into the task execution database for identifying scheduling and execution problems, and feeds back the problem information to the scheduling module. Through the feedback closed-loop, the system can adaptively handle changes in different task and personnel states, optimize the scheduling efficiency, and ultimately achieve intelligent scheduling.

[0174] Please refer to Figure 7, the embodiments of the present application further provide a task scheduling device that can implement the above task scheduling method. The device includes:

[0175] An acquisition module 701, configured to acquire flight dynamic data, personnel qualification data, personnel status data, and maintenance task information, and generate an input data set;

[0176] An optimization module 702, configured to perform multi-objective optimization on the input data set based on a mixed integer non-linear programming model to generate an initial task scheduling plan. The multi-objectives include minimizing task delay, equalizing personnel work, minimizing personnel movement time, and minimizing personnel fatigue;

[0177] A detection module 703, configured to generate a task rescheduling plan based on a dynamic programming algorithm and a simulated annealing algorithm when it detects that the flight dynamic data, personnel status data, or maintenance task information has changed;

[0178] A receiving module 704, configured to push the task rescheduling plan to maintenance personnel and receive task execution feedback data;

[0179] An update module 705, configured to update the task scheduling strategy corresponding to the task rescheduling plan based on the task execution feedback data through a deep Q-network algorithm.

[0180] The specific implementation manner of this task scheduling device is basically the same as that of the specific embodiment of the above task scheduling method, and will not be elaborated here.

[0181] In a third aspect, the embodiments of the present application provide an electronic device. Refer to Figure 8 shown in the figure, which is a schematic structural diagram of an electronic device provided by the present application.

[0182] As Figure 8 shown in the figure, the device includes:

[0183] A memory 31, configured to store a computer program;

[0184] A processor 32, configured to execute the computer program;

[0185] Among them, when the processor 32 executes the computer program, it implements the task scheduling method in any of the above embodiments.

[0186] Exemplarily, the computer program can be divided into one or more modules / units. One or more modules / units are stored in the memory 31 and executed by the processor 32 to complete the present application. One or more modules / units can be a series of computer program instruction segments that can complete specific functions, and these instruction segments are used to describe the execution process of the computer program in the electronic device.

[0187] The so-called processor 32 may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0188] The memory 31 can be used to store computer programs and / or modules. The processor 32 realizes various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory 31, and by calling the data stored in the memory 31. The memory 31 mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created according to the use of the mobile phone (such as audio data, phone book, etc.). In addition, the memory 31 can include high-speed random access memory, and can also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0189] It should be noted that the above-mentioned electronic device includes a processor and a memory, but is not limited to the processor and the memory. Those skilled in the art can understand that Figure 8 The structural schematic diagram is only an example of the above-mentioned electronic device, and does not constitute a limitation on the electronic device. It may include more components than shown in the figure, or combine some components, or different components.

[0190] In a fourth aspect, the embodiments of the present application further provide a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed, the task scheduling method of any of the above embodiments is implemented.

[0191] It should be understood that all or part of the processes in the above task scheduling method of this application can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above task scheduling method can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in relevant jurisdictions. For example, in some relevant jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0192] In a fifth aspect, an embodiment of the present application further provides a computer program product. The computer program product is stored in a storage medium, and the program product is executed by at least one processor to implement the task scheduling method of any of the above embodiments.

[0193] Those of ordinary skill in the art can understand that all or part of the processes in the above method embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above method embodiments. Among them, the storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0194] The above are the preferred embodiments of the present application. It should be noted that for those of ordinary skill in the art of this technology, without departing from the principle of the present application, several improvements and refinements can still be made, and these improvements and refinements are also regarded as the protection scope of the present application.

Claims

1. A task scheduling method, characterized in that, including: Collecting flight dynamic data, personnel qualification data, personnel status data, and maintenance task information to generate an input data set; Performing multi-objective optimization on the input data set based on a mixed-integer non-linear programming model to generate an initial task scheduling plan, where the multi-objectives include minimizing task delays, equalizing personnel workloads, minimizing personnel movement time, and minimizing personnel fatigue; When it is detected that the flight dynamic data, the personnel status data, or the maintenance task information has changed, generating a task re-scheduling plan based on the dynamic programming algorithm and the simulated annealing algorithm; Pushing the task re-scheduling plan to maintenance personnel and receiving task execution feedback data; Based on the task execution feedback data, updating the task scheduling strategy corresponding to the task re-scheduling plan through the deep Q-network algorithm.

2. The task scheduling method according to claim 1, characterized in that, The performing multi-objective optimization on the input data set based on the mixed-integer non-linear programming model to generate an initial task scheduling plan includes: Constructing a mixed-integer non-linear programming model, defining the decision variables, objective function, and constraint conditions of the mixed-integer non-linear programming model, and using a meta-heuristic algorithm to solve the mixed-integer non-linear programming model to generate a Pareto optimal solution set; Selecting a balanced solution from the Pareto optimal solution set to obtain an initial task scheduling plan.

3. The task scheduling method according to claim 2, wherein The construction process of the mixed-integer non-linear programming model includes the following steps: (1) Determine the decision variables: y ij , t j , t1, t2, t3 and t4; Among them, y ij is a binary variable. y ij = 1 indicates that the ith maintenance personnel is responsible for the jth task (i.e., task j), and y ij = 0 indicates that the ith maintenance personnel is not responsible for the jth task; t j is the start time of the jth task; t1 is the pick-up time; t2 is the inspection time; t3 is the start time of the guardianship task; t4 is the drop-off time; (2) Determining the objective function, where the objective function includes: Minimize task delays: Balancing of personnel work: Minimization of personnel movement time: min∑ i,j,k (y ij ·y ik ·d jk ), And minimize personnel fatigue: min∑ i fatigue(T i ); Among them, t j is the task start time of the jth task, represents the planned start time of task j; T i represents the total working time of the i-th maintenance personnel, is the average working time; ik is a zero-one variable, y ik =1 means that the i-th maintenance personnel is responsible for the k-th task (i.e., task k), y ik =0 means that the i-th maintenance personnel is not responsible for the k-th task, d jk is the distance between mission j and mission k; fatigue(T i ) is based on the total working time T of the i-th maintenance personnel i The calculated fatigue level, h is the initial fatigue, and λ determines the rate at which fatigue increases with working time; (3) Determining the constraint conditions, where the constraint conditions include: Task timing constraint: t1 = t j -15, t2≥t1 + 5, t3≥t2+(20δ1 + 25δ2), t4≥t d -15, Personnel qualification constraint: ∑ i y i1 = 2, ∑ i y i2 ∈{1,2}， ∑ i y i3 ＝1， ∑ i y i4 ＝2, Moving time constraint: t move +t j ≤t k , and working hour constraint: T i ≤T max ,T rest ≥30; Among them, t j is the start time of the j-th task, t1 is the pick-up time, t2 is the inspection time, t3 is the start time of the guardianship task, t4 is the drop-off time, t d is the planned take-off time, δ1 represents that there are 2 maintenance personnel performing the inspection task, δ2 represents that there is 1 maintenance personnel performing the inspection task, δ1 and δ2 are binary variables, and δ1 + δ2 = 1, indicating that there is 1 or 2 maintenance personnel performing the inspection task; y i1 is the pick-up task, is the signing card qualification; y i2 is the inspection task, is the release qualification; y i3 is the guardianship task; y i4 is the drop-off task; t move is the moving time, t k is the start time of the k-th task; T max is the maximum working time limit; T rest is the rest time; (4) Using a meta-heuristic algorithm to solve the mixed-integer non-linear programming model to generate a Pareto optimal solution set.

4. The task scheduling method according to claim 1, wherein The collecting flight dynamic data, personnel qualification data, personnel status data, and maintenance task information to generate an input data set includes: Real-time collecting maintenance task information, flight dynamic data of the flight scheduling system, personnel qualification data and personnel status data of the maintenance personnel management system through an API interface; among them, the flight dynamic data includes the flight planned arrival time, actual arrival time, parking position number, and task type; the personnel qualification data includes signing card qualification and release qualification, and the personnel status data includes the current location, workload, and fatigue; the maintenance task information includes task type, task priority, and task execution time requirement; Structuring the flight dynamic data, the personnel qualification data, the personnel status data, and the maintenance task information to generate an input data set.

5. The task scheduling method according to claim 1, wherein The when it is detected that the flight dynamic data, the personnel status data, or the maintenance task information has changed, generating a task re-scheduling plan based on the dynamic programming algorithm and the simulated annealing algorithm includes: Real-time detecting the flight dynamic data, the personnel status data, and the maintenance task information through an event-driven architecture, and triggering task re-scheduling when flight delays, parking position changes, personnel unavailability, or task time changes are detected; Optimizing the current task allocation status based on the dynamic programming algorithm to generate a preliminary re-scheduling plan; The simulated annealing algorithm is used to globally optimize the preliminary rescheduling scheme to generate a task rescheduling scheme.

6. The task scheduling method according to claim 1, wherein Pushing the task rescheduling scheme to the maintenance personnel and receiving task execution feedback data, including: Pushing the task rescheduling scheme to the maintenance personnel through mobile APP notifications, SMS, communication tool messages, or voice calls, and receiving task confirmation information; Receiving the task execution status data feedback by the maintenance personnel, where the task execution status data includes the task start time, task completion time, and task execution result.

7. The task scheduling method according to claim 1, characterized in that Based on the task execution feedback data, updating the task scheduling strategy corresponding to the task rescheduling scheme through the deep Q-network algorithm, including: Constructing a deep Q-network model, where the state space of the model includes flight dynamics, maintenance personnel availability, workload, personnel qualifications, and fatigue status, and the action space of the model includes task allocation, personnel adjustment, and task priority adjustment; Designing the reward function of the deep Q-network model, where the reward function is designed based on task completion and personnel utilization rate, obtaining a positive reward when the task is completed on time, and obtaining a negative reward when the task is delayed; Based on the task execution feedback data, updating the Q-value table of the deep Q-network model through an experience replay mechanism; Updating the policy network parameters of the deep Q-network model through a target network to generate an updated task scheduling strategy.

8. A task scheduling device, characterized in that Including: A collection module for collecting flight dynamic data, personnel qualification data, personnel status data, and maintenance task information to generate an input data set; An optimization module for performing multi-objective optimization on the input data set based on a mixed-integer nonlinear programming model to generate an initial task scheduling scheme, where the multi-objectives include minimizing task delay, equalizing personnel workload, minimizing personnel movement time, and minimizing personnel fatigue; A detection module for generating a task rescheduling scheme based on dynamic programming algorithm and simulated annealing algorithm when it detects changes in the flight dynamic data, the personnel status data, or the maintenance task information; A receiving module for pushing the task rescheduling scheme to the maintenance personnel and receiving task execution feedback data; An updating module for updating the task scheduling strategy corresponding to the task rescheduling scheme through the deep Q-network algorithm based on the task execution feedback data.

9. An electronic device, characterized in that, Including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, where the processor implements the task scheduling method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, where when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the task scheduling method according to any one of claims 1 to 7.

Images