Flight scheduling method and device with autonomous guidance function and processing equipment

Abstract

The invention provides a flight scheduling method and device with an autonomous guidance function and processing equipment, and the method comprises the steps: carrying out the real-time scoring and adaptive learning of a flight removal-reinsertion operation through introducing a value network which can be updated online, thereby optimizing a landing sequence step by step on the premise of strictly meeting a constraint condition, and improving the efficiency of flight scheduling. In the search process, a neighborhood structure based on position offset limitation is adopted to effectively control the search space and improve the search efficiency, and meanwhile, an autonomous guidance learning device based on a neural network can also accurately predict and guide the algorithm convergence trend; through the hybrid interaction mechanism, the global exploration capability of the neural network and the local optimization advantage of heuristic search are organically combined, so that higher robustness, adaptability and calculation efficiency can be shown in a complex and dynamic aviation scheduling environment, and high-quality data support can be provided for flight scheduling work.

Description

Technical Field

[0001] This application relates to the field of flight scheduling, specifically to a flight scheduling method, apparatus, and processing equipment with autonomous guidance function. Background Technology

[0002] As a core component of aviation operations management, landing scheduling focuses on the scientific arrangement of flight takeoffs and landings, taxiing, parking positions, and gate allocation under the constraints of limited airspace resources, airport capacity, and ground support conditions, in order to achieve the operational goals of safety, efficiency, and economy. It encompasses a series of key issues such as flight slot allocation, terminal area traffic management, ground service resource allocation, and flight conflict resolution.

[0003] It is understandable that efficient and accurate air traffic control can not only improve flight punctuality, reduce delay costs, and optimize resource utilization, but is also the key to ensuring operational safety and supporting the efficient operation of airports and airlines.

[0004] As aviation networks become increasingly complex and flight volumes continue to grow, air traffic control exhibits highly complex combinations and strong real-time requirements, making it a typical nondeterministic polynomial-time (NP) problem.

[0005] Modern flight scheduling technology widely utilizes optimization algorithms, heuristic strategies, integer programming, simulation, and artificial intelligence (AI) methods. Among AI methods, machine learning is particularly prominent in dynamic scheduling environments.

[0006] However, the inventors of this application have discovered that existing flight scheduling technologies based on machine learning methods still have limitations in terms of robustness, adaptability, and computational efficiency, which are accompanied by corresponding flight delay risks. Summary of the Invention

[0007] This application provides a flight scheduling method, apparatus, and processing device with autonomous guidance. By introducing an online-updable value network, the flight removal-reinsertion operation is scored and adaptively learned in real time. Under strict constraints, the landing sequence is gradually optimized. During the search process, a neighborhood structure based on position offset constraints is used to effectively control the search space and improve search efficiency. At the same time, the autonomous guidance learner based on neural networks can accurately predict and guide the convergence trend of the algorithm. This hybrid interaction mechanism organically combines the global exploration capability of neural networks with the local optimization advantages of heuristic search, thus exhibiting stronger robustness, adaptability, and computational efficiency in complex and dynamic air traffic control environments. This provides high-quality data support for flight scheduling.

[0008] Firstly, this application provides a flight scheduling method with autonomous guidance function, the method comprising: After obtaining a set of flight data to be processed, sort them in ascending order based on the corresponding estimated arrival times to obtain a set of initial flight sequences. The initial total delay of the value network is updated based on the flight sequence, where the value network is used to predict the delay improvement brought about by flight swapping operations and output the optimal flight sequence; The flight sequence is used as the initial solution and is fed into the value network for iterative optimization. In the iterative optimization stage, candidate removal-re-insertion pairs are randomly generated to simulate the exchange, and the constraint detection function is used to ensure that the current sequence meets the flight position offset limit. Then, a lightweight feedforward neural network autonomously guides the learner to predict the delay improvement prediction value of each removal-re-insertion pair in order to determine the candidate flight sequence that is most likely to reduce the total delay and be output as the optimal flight sequence. Extract the processing results from the value network output.

[0009] Secondly, this application provides a flight scheduling device with autonomous guidance function, the device comprising: The acquisition unit is used to acquire a set of flight data to be processed, and then sort it in ascending order based on the corresponding estimated arrival times to obtain a set of initial flight sequences. The update unit is used to update the initial total delay of the value network based on the flight sequence, wherein the value network is used to predict the delay improvement brought about by the flight exchange operation and output the optimal flight sequence; The iterative unit is used to import the flight sequence as the initial solution into the value network for iterative optimization. In the iterative optimization stage, candidate removal-re-insertion pairs are randomly generated to perform exchange simulations, and the constraint detection function is used to ensure that the current sequence meets the flight position offset limit. Then, a lightweight feedforward neural network autonomously guides the learner to predict the delay improvement prediction value of each removal-re-insertion pair in order to determine the candidate flight sequence that is most likely to reduce the total delay and be output as the optimal flight sequence. The extraction unit is used to extract the processing results output by the value network.

[0010] Thirdly, this application provides a processing device, including a processor and a memory, wherein a computer program is stored in the memory, and the processor executes the method provided in the first aspect of this application when it invokes the computer program in the memory.

[0011] Fourthly, this application provides a computer-readable storage medium storing a plurality of instructions adapted for loading by a processor to execute the method provided in the first aspect of this application.

[0012] From the above, it can be concluded that this application has the following beneficial effects: To address flight scheduling objectives, this application introduces an online-updable value network to perform real-time scoring and adaptive learning of flight removal-reinsertion operations. This gradually optimizes landing sequences while strictly meeting constraints. Furthermore, a neighborhood structure based on position offset constraints is employed during the search process to effectively control the search space and improve search efficiency. Simultaneously, a neural network-based autonomous learning engine accurately predicts and guides the algorithm's convergence trend. This hybrid interactive mechanism organically combines the global exploration capabilities of neural networks with the local optimization advantages of heuristic search, resulting in stronger robustness, adaptability, and computational efficiency in complex and dynamic air traffic control environments. This provides high-quality data support for flight scheduling operations. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart illustrating a flight scheduling method with autonomous guidance function according to this application. Figure 2 This is a logical diagram illustrating the autonomous flight sequence optimization logic of this application. Figure 3This is a schematic diagram illustrating an example of the removal-re-insertion action in this application; Figure 4 This is a schematic diagram of a flight scheduling device with autonomous guidance function according to this application. Figure 5 This is a schematic diagram of one type of processing equipment used in this application. Detailed Implementation

[0015] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0016] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules is not necessarily limited to those explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices. The naming or numbering of steps appearing in this application does not imply that the steps in the method flow must be performed in the chronological / logical order indicated by the naming or numbering. The execution order of named or numbered process steps can be changed according to the desired technical purpose, as long as the same or similar technical effect is achieved.

[0017] The module division described in this application is a logical division. In practical applications, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual coupling, direct coupling, or communication connections may be through interfaces, and the indirect coupling or communication connections between modules may be electrical or other similar forms, none of which are limited in this application. Moreover, the modules or sub-modules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed across multiple circuit modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution in this application.

[0018] Before introducing the flight scheduling method with autonomous guidance function provided in this application, we will first introduce the background content involved in this application.

[0019] The flight scheduling method, apparatus, and computer-readable storage medium with autonomous guidance provided in this application can be applied to processing equipment. By introducing an online-updable value network, the flight removal-reinsertion operation is scored and adaptively learned in real time, thereby gradually optimizing the landing sequence under strict constraints. During the search process, a neighborhood structure based on position offset constraints is used to effectively control the search space and improve search efficiency. At the same time, the autonomous guidance learner based on neural networks can accurately predict and guide the convergence trend of the algorithm. This hybrid interaction mechanism organically combines the global exploration capability of neural networks with the local optimization advantages of heuristic search, thus exhibiting stronger robustness, adaptability, and computational efficiency in complex and dynamic air traffic control environments. This can provide high-quality data support for flight scheduling.

[0020] The flight scheduling method with autonomous guidance function mentioned in this application can be implemented by a flight scheduling device with autonomous guidance function, or by different types of processing devices such as servers, physical hosts, or user equipment (UE) that integrate the flight scheduling device with autonomous guidance function. The flight scheduling device with autonomous guidance function can be implemented in hardware or software. The UE can be a terminal device such as a smartphone, tablet, laptop, desktop computer, or personal digital assistant (PDA). The processing devices can be configured in a device cluster.

[0021] In practical applications, flight scheduling schemes are usually used for decision-making by airport flight scheduling systems. Therefore, the processing equipment that executes the flight scheduling method with autonomous guidance function of this application or that carries the corresponding application service of the flight scheduling method with autonomous guidance function of this application is usually related equipment of the airport flight scheduling system. It only needs to meet the required data processing capabilities, and its specific equipment type and equipment deployment form are quite flexible.

[0022] In addition, if there is a need to display the processing progress (including the processing results), the processing device itself can be configured with the required display screen (including touch screen) to display the specific content. Of course, the processing device can also display the specific content through an external display device or other devices with a display screen.

[0023] The following section introduces the flight scheduling method with autonomous guidance function provided in this application.

[0024] First, refer to Figure 1 , Figure 1 This paper illustrates a flowchart of a flight scheduling method with autonomous guidance function according to this application. The flight scheduling method with autonomous guidance function provided by this application may specifically include the following steps S101 to S104: Step S101: After obtaining a set of flight data to be processed, sort them in ascending order based on the corresponding estimated arrival times to obtain a set of corresponding initial flight sequences; In practical applications, the proposed solution can be triggered manually to schedule a specific group of flights, or the system can trigger the scheduling of a specific group of flights under conditions such as dynamic updates of flights, timed or random triggering.

[0025] Correspondingly, the specific solution can be implemented by manually entering data, reading data locally, reading data online, or receiving relevant flight data forwarded from other places.

[0026] It is easy to understand that in the scheme of this application, the optimal flight sequence to be output is obtained by iterative optimization / search. In this regard, this application can, in the initial stage, use the flight sequence obtained from the flight data to be processed, which is sorted in ascending order based on the corresponding estimated time of arrival (ETA), as the initial flight sequence.

[0027] This process may involve encoding flights. Furthermore, an initial flight sequence can be generated using the First-Come First-Served (FCFS) method, which can be denoted as... , , This represents the total number of flights.

[0028] Step S102: Update the initial total delay of the value network according to the flight sequence, wherein the value network is used to predict the delay improvement brought about by the flight exchange operation and output the optimal flight sequence; Understandably, the value network carries the specific processing logic designed for iterative optimization of flight sequences in this application. Its goal is to predict the delay improvement / enhancement brought about by flight exchange operations and determine the flight sequence most likely to reduce the total delay as the optimal flight sequence output. It is itself a machine learning model.

[0029] The flights involved are fixed; what needs to be adjusted is the order in which different flights land. This corresponds to the initial flight sequence being obtained in ascending order of expected arrival times. The solution process needs to provide data references on the optimal flight sequence before these flights land.

[0030] In this case, the total delay of the previously determined flight sequence can be calculated during the network initialization phase as the initial total delay of the value network, and then iteratively optimized along with the flight sequence in each subsequent round of iterative optimization.

[0031] The value network can be denoted as: The initial total delay can be denoted as .

[0032] Step S103: The flight sequence is imported into the value network as the initial solution for iterative optimization. In the iterative optimization stage, candidate removal-re-insertion pairs are randomly generated for exchange simulation, and the constraint detection function is used to ensure that the current sequence meets the flight position offset limit. Then, a lightweight feedforward neural network autonomously guides the learner to predict the delay improvement prediction value of each removal-re-insertion pair in order to determine the candidate flight sequence that is most likely to reduce the total delay and be output as the optimal flight sequence.

[0033] Specifically, this application includes the following optimizations for flight scheduling: 1) An online-updable value network is introduced to perform real-time scoring and adaptive learning on flight removal-reinsertion operations, thereby progressively optimizing the landing sequence under strict constraints. 2) During the search process, a neighborhood structure based on Constrained Position Shifting (CPS) is adopted to effectively control the search space and improve search efficiency. It uses a removal-re-insertion neighborhood structure to evaluate newly generated working sequences, has a flexible search space, can effectively improve the rationality of the solution, and can find flight sequences that meet the requirements within a reasonable time. 3) The autonomous guided learner based on neural networks can also accurately predict and guide the convergence trend of the algorithm. Through this hybrid interaction mechanism, the global exploration capability of neural networks and the local optimization advantage of heuristic search are organically combined, thus exhibiting stronger robustness, adaptability and computational efficiency in complex and dynamic aviation scheduling environments.

[0034] The following is a combination Figure 2 The diagram shown illustrates a logic diagram of the autonomous flight sequence optimization logic of this application, and provides a detailed explanation of the multiple optimization settings involved in the autonomous flight sequence optimization process of this application.

[0035] Candidate generation and filtering: In the current flight sequence In the middle, several candidate positions are randomly selected. To perform the corresponding remove-reinsert operation, the corresponding flight sequence is obtained. ; If flight sequence If the position offset constraint is satisfied, the corresponding feature vector can be calculated. and through This prediction function is used to predict the delay improvement value; At this point, all candidate values, or predicted scores, are improved according to the corresponding delay. Sort the candidates in descending order and retain the top k high-potential candidates to proceed to the next stage.

[0036] Learning and acceptance mechanisms: For each candidate removal-reinsertion action, calculate the corresponding real latency improvement. The corresponding representation is: , and with As a monitoring signal, the value network is updated online: , like This indicates that the delay is reduced, so the current remove-reinsert action can be accepted, and the current solution can be updated. .

[0037] In layman's terms, during the iterative optimization process, if the position offset constraint is still satisfied after the flight exchange, the new total delay can be calculated, and the value network can be updated based on the actual improvement value. If there is improvement, the exchange is accepted and the current flight sequence is updated.

[0038] From the above general iterative optimization logic, it can also be seen that in the network design of this application, the input to the delay-improved prediction processing is specifically the feature vector obtained by quantizing the current removal-reinsertion exchange simulation operation. .

[0039] And this feature vector Specifically, this can include relative position characteristics, aircraft type characteristics, and initial position difference characteristics.

[0040] And for candidate position pairs For the remove-re-insert action, see [reference]. Figure 3 The illustrated diagram illustrates one example of the removal-re-insertion operation of this application, to provide a more visual understanding of the following set of examples: In the first phase, for a flight sequence with n flights, select the position. and , ,but ; Phase Two, Deletion Position ; The third stage will Insert at position .

[0041] Furthermore, as an exemplary embodiment, the feature vector Specifically, it can be configured as a 12-dimensional vector, represented as... ,Right now , , …and .

[0042] Specifically, for relative position features, there can be the following 5-dimensional feature vectors: , , , , ; For aircraft type features, there can be the following 6 dimensions of feature vectors: , ; For the initial positional difference features, there can be a feature vector with the following one dimension: ; Meanwhile, for the above eigenvector quantization formula, the eigenvalue normalization process involved is as follows: ; Furthermore, the implementation process involved in the above feature vector quantization formula is as follows: Get the flight number to be exchanged: , , Indicates the initial flight sequence, position pair For candidate position pairs , The estimated arrival time for flight number a The estimated arrival time for flight number b. initialization , Calculate time and location difference features , For the aircraft type of flight number a and flight number b Perform one-hot encoding and fill in the blanks. , Calculate and normalize the original position difference. .

[0043] Understandably, this section specifically reveals the detailed input processing settings involved in the aforementioned 12-dimensional feature vector quantization formula. Thus, starting from the level of the refined quantization formula, a new feature vector structure is formed, which can objectively and clearly describe the features of the removal-re-insertion neighborhood structure. This lays a good foundation for the high-performance delay improvement prediction work carried out by the lightweight feedforward neural network autonomous guided learner specially designed in this application.

[0044] In this regard, the value network can also be configured with: Delayed Improved Predictive Value Prediction Processing by Predictive Score Sort candidate position pairs in descending order Prioritize positions with the greatest expected improvement, calculate the true delay of the corresponding new sequence under the scheduler, and use the true observations as the target to update the learner (including candidate position pairs). eigenvectors Predicted scores To avoid extreme values ​​affecting learning, some cropping can also be done.

[0045] Understandably, iterative optimization targeting areas where significant improvement is expected helps accelerate algorithm convergence.

[0046] Furthermore, the lightweight feedforward neural network autonomous guided learner designed in this application is a novel autonomous guided learner, and also a lightweight feedforward neural network used to learn feature vectors. The nonlinear mapping relationship between the predicted value and the delayed improvement is specifically implemented using an online learning mechanism during operation. This mechanism achieves dynamic adaptive capability through continuous parameter updates. Further details include: 1) Regarding forward propagation, the calculation process can be expressed as: , in, To improve the predicted value with a delay, For the input vector, , For the input layer dimension, and For different hidden layer parameters, , , For the hidden layer dimension, and For different output layer parameters, , , The hyperbolic tangent activation function is used. The weight parameters employ the Xavier initialization strategy, a parameter initialization method specifically designed for deep neural networks. This strategy aims to address the vanishing or exploding gradient problems that occur as network layers deepen. Correspondingly: , in, This indicates that the mean is 0 and the variance is 0. The normal distribution; 2) Regarding the loss function, the Mean Squared Error (MSE) can be used, which can be expressed as: , in, This is the single-sample mean squared error. Improve observations to achieve true time delay; 3) In terms of backpropagation and gradient calculation, in order to minimize Backpropagation can be performed according to the chain rule, calculating the gradient layer by layer, corresponding to the error terms involved. It can be represented as: , The process of calculating the parameter gradient can be represented as: , , , , , , in, , This represents element-wise multiplication. This represents the outer product operation. This is an intermediate quantity.

[0047] 4) Regarding the parameter update mechanism, after obtaining the gradient, the parameters are updated using the stochastic gradient descent method, which can be expressed as: , , , , in, This is the learning rate.

[0048] Furthermore, in combination Figure 2 It can also be seen that the value network of this application can be configured with perturbation and escape from local optima strategies, or low-probability perturbation mechanisms, corresponding to: When an iteration does not accept any remove-reinsert actions, a random perturbation is performed, accepting a slightly worse solution with a small probability, to increase search diversity.

[0049] On the other hand, once the maximum number of iterations is reached or the delay convergence condition is met, the value network can complete the iterative optimization work and output the corresponding optimal flight scheduling sequence. There is also a corresponding minimum total delay. (Or it can be written as total_delay).

[0050] Step S104: Extract the processing results output by the value network.

[0051] After the optimal flight scheduling sequence is processed through the value network, it is understandable that the processing result output by the value network can be extracted, namely the optimal flight sequence, or the optimal flight sequence and its corresponding feature indicators.

[0052] In this case, the output of the processing result can be further processed according to the specific application scenario of the solution.

[0053] For example, processing results or processes can be stored locally, stored off-site, displayed, forwarded, output as a completion message, or further analyzed.

[0054] The specific output processing content can be flexibly configured according to the pre-configured and implemented output strategies.

[0055] Taking results presentation as an example, solution implementers such as airport flight scheduling systems can use the display screens configured on the workbench to present specific content, providing intuitive visual data support to relevant on-site staff for further work decision-making.

[0056] In conclusion, regarding the above solutions, this application introduces an online-updable value network to perform real-time scoring and adaptive learning of flight removal-reinsertion operations, thereby gradually optimizing landing sequences while strictly meeting constraints. Furthermore, a neighborhood structure based on position offset constraints is employed during the search process to effectively control the search space and improve search efficiency. Simultaneously, a neural network-based autonomous learning engine can accurately predict and guide the algorithm's convergence trend. This hybrid interactive mechanism organically combines the global exploration capabilities of neural networks with the local optimization advantages of heuristic search, resulting in stronger robustness, adaptability, and computational efficiency in complex and dynamic air traffic control environments. This provides high-quality data support for flight scheduling.

[0057] The above is an introduction to the flight scheduling method with autonomous guidance function provided in this application. In order to facilitate better implementation of the flight scheduling method with autonomous guidance function provided in this application, this application also provides a flight scheduling device with autonomous guidance function from the perspective of functional modules.

[0058] See Figure 4 , Figure 4 This is a schematic diagram of a flight scheduling device with autonomous guidance function according to this application. In this application, the flight scheduling device 400 with autonomous guidance function may specifically include the following structure: The acquisition unit 401 is used to acquire a set of flight data to be processed, and then sort them in ascending order based on the corresponding expected arrival times to obtain a set of corresponding initial flight sequences. The update unit 402 is used to update the initial total delay of the value network according to the flight sequence, wherein the value network is used to predict the delay improvement brought about by the flight exchange operation and output the optimal flight sequence; Iteration unit 403 is used to import the flight sequence as the initial solution into the value network for iterative optimization. In the iterative optimization stage, candidate removal-re-insertion pairs are randomly generated to perform exchange simulations, and the constraint detection function is used to ensure that the current sequence meets the flight position offset limit. Then, a lightweight feedforward neural network autonomously guides the learner to predict the delay improvement prediction value of each removal-re-insertion pair in order to determine the candidate flight sequence that is most likely to reduce the total delay and be output as the optimal flight sequence. Extraction unit 404 is used to extract the processing results output by the value network.

[0059] In one exemplary embodiment, the input to the delay-improved prediction processing is the feature vector obtained by quantizing the current remove-reinsertion swap simulation operation. ; Feature vector This includes relative position characteristics, aircraft type characteristics, and initial position difference characteristics.

[0060] In yet another exemplary embodiment, the feature vector For a 12-dimensional vector, it is represented as ; For relative position features, we have: , , , , ; Regarding the characteristics of the model type, we have: , ; For the initial positional differences, we have: ; The eigenvalue normalization processes involved include: ; The implementation process involved includes: Get the flight number to be exchanged: , , Indicates the initial flight sequence, position pair For candidate position pairs , The estimated arrival time for flight number a The estimated arrival time for flight number b. initialization , Calculate time and location difference features , For the aircraft type of flight number a and flight number b Perform one-hot encoding and fill in the blanks. , Calculate and normalize the original position difference. .

[0061] In yet another exemplary embodiment, the delay-improved prediction processing is performed according to the delay-improved prediction value. Sort candidate position pairs in descending order Prioritize trying positions where the expected improvement is large, calculate the actual delay of the corresponding new sequence under the scheduler, and use the actual observations as the target to update the learner.

[0062] In yet another exemplary embodiment, a lightweight feedforward neural network autonomously guided learner is used to learn feature vectors. The nonlinear mapping relationship between the prediction value and the delay improvement is achieved by adopting an online learning mechanism during operation, which continuously updates the parameters to realize dynamic adaptive capability. In terms of forward propagation, the calculation process is represented as follows: , in, To improve the predicted value with a delay, For the input vector, , For the input layer dimension, and For different hidden layer parameters, , , For the hidden layer dimension, and For different output layer parameters, , , The hyperbolic tangent activation function is used. The weight parameters use the Xavier initialization strategy, which includes: , in, This indicates that the mean is 0 and the variance is 0. The normal distribution; Regarding the loss function, the single-sample mean squared error is used, expressed as: , in, This is the single-sample mean squared error. Improve observations to achieve true time delay; In terms of backpropagation and gradient calculation, backpropagation is performed according to the chain rule, calculating the gradient layer by layer, corresponding to the error terms involved. Represented as: , The process of calculating the parameter gradient is expressed as follows: , , , , , , in, , This represents element-wise multiplication. This represents the outer product operation. This is an intermediate quantity; Regarding the parameter update mechanism, after obtaining the gradient, the parameters are updated using the stochastic gradient descent method, which is represented as follows: , , , , in, This is the learning rate.

[0063] In yet another exemplary embodiment, the value network is further configured with perturbation and escape strategies from local optima, corresponding to: When an iteration does not accept any remove-reinsert actions, a random perturbation is performed, accepting a slightly worse solution with a small probability, to increase search diversity.

[0064] In yet another exemplary embodiment, the value network outputs the optimal flight scheduling sequence after the maximum number of iterations is reached or the delay convergence condition is met. There is also a corresponding minimum total delay. .

[0065] This application also provides a processing device from a hardware architecture perspective. As mentioned earlier, in practice, a processing device may exist as a device cluster. In this case, each device in the device cluster can also be referred to as a processing device. See [reference needed]. Figure 5 , Figure 5 This diagram illustrates a structural schematic of the processing device of this application. Specifically, the processing device may include a processor 501, a memory 502, and an input / output device 503. The processor 501 executes the computer program stored in the memory 502 to implement, for example... Figure 1 The corresponding steps of the flight scheduling method with autonomous guidance function in the embodiment; or, when the processor 501 executes the computer program stored in the memory 502, it implements as follows: Figure 4 Corresponding to the functions of each unit in the embodiment, the memory 502 is used to store the functions executed by the processor 501 as described above. Figure 1 The computer program required for the flight scheduling method with autonomous guidance function in the corresponding embodiment.

[0066] For example, a computer program may be divided into one or more modules / units, one or more of which are stored in memory 502 and executed by processor 501 to complete this application. One or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in a computer device.

[0067] The processing device may include, but is not limited to, processor 501, memory 502, and input / output device 503. Those skilled in the art will understand that the illustrations are merely examples of the processing device and do not constitute a limitation on the processing device. It may include more or fewer components than illustrated, or combine certain components, or different components. For example, the processing device may also include network access devices, buses, etc., and processor 501, memory 502, input / output device 503, etc., are connected via a bus.

[0068] Processor 501 can be a Central Processing Unit (CPU), or 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. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the processing device, connecting various parts of the device through various interfaces and lines.

[0069] The memory 502 can be used to store computer programs and / or modules. The processor 501 implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory 502 and by calling data stored in the memory 502. The memory 502 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function, etc.; the data storage area may store data created according to the use of the processing device, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0070] When processor 501 executes a computer program stored in memory 502, it can specifically perform the following functions: After obtaining a set of flight data to be processed, sort them in ascending order based on the corresponding estimated arrival times to obtain a set of initial flight sequences. The initial total delay of the value network is updated based on the flight sequence, where the value network is used to predict the delay improvement brought about by flight swapping operations and output the optimal flight sequence; The flight sequence is used as the initial solution and is fed into the value network for iterative optimization. In the iterative optimization stage, candidate removal-re-insertion pairs are randomly generated to simulate the exchange, and the constraint detection function is used to ensure that the current sequence meets the flight position offset limit. Then, a lightweight feedforward neural network autonomously guides the learner to predict the delay improvement prediction value of each removal-re-insertion pair in order to determine the candidate flight sequence that is most likely to reduce the total delay and be output as the optimal flight sequence. Extract the processing results from the value network output.

[0071] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the flight scheduling device, processing equipment, and its corresponding units with autonomous guidance function described above can be referred to as follows: Figure 1 The description of the flight scheduling method with autonomous guidance function in the corresponding embodiment will not be repeated here.

[0072] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0073] Therefore, this application provides a computer-readable storage medium storing a plurality of instructions that can be loaded by a processor to execute the present application. Figure 1 For details on the steps of the flight scheduling method with autonomous guidance function in the corresponding embodiment, please refer to the following: Figure 1 The description of the flight scheduling method with autonomous guidance function in the corresponding embodiment will not be repeated here.

[0074] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0075] Because of the instructions stored in the computer-readable storage medium, the present application can be executed as described above. Figure 1 The steps of the flight scheduling method with autonomous guidance function in the corresponding embodiment can therefore achieve the results of this application. Figure 1 The beneficial effects that the flight scheduling method with autonomous guidance function can achieve in the corresponding embodiment are detailed in the preceding description and will not be repeated here.

[0076] The foregoing has provided a detailed description of the flight scheduling method, apparatus, processing equipment, and computer-readable storage medium with autonomous guidance function provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are only for the purpose of helping to understand the core ideas of this application; furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A flight scheduling method with autonomous guidance function, characterized in that, The method includes: After obtaining a set of flight data to be processed, sort them in ascending order based on the corresponding estimated arrival times to obtain a set of initial flight sequences. The initial total delay of the value network is updated based on the flight sequence, wherein the value network is used to predict the delay improvement brought about by flight swapping operations and output the optimal flight sequence; The flight sequence is used as the initial solution and imported into the value network for iterative optimization. In the iterative optimization stage, candidate removal-re-insertion pairs are randomly generated for exchange simulation, and a constraint detection function is used to ensure that the current sequence meets the flight position offset limit. Then, a lightweight feedforward neural network autonomously guides the learner to predict the delay improvement prediction value of each removal-re-insertion pair to determine the candidate flight sequence that is most likely to reduce the total delay and be output as the optimal flight sequence. Extract the processing results output by the value network.

2. The method according to claim 1, characterized in that, The input to the delayed improved prediction processing is the feature vector obtained by quantizing the current removal-reinsertion exchange simulation operation. ; The feature vector This includes relative position characteristics, aircraft type characteristics, and initial position difference characteristics.

3. The method according to claim 2, characterized in that, The feature vector For a 12-dimensional vector, it is represented as ; For the aforementioned relative position feature, we have: ， ， ， ， ； Regarding the aforementioned model type characteristics, we have: ， ； Regarding the initial position difference characteristics, we have: ； The eigenvalue normalization processes involved include: ； The implementation process involved includes: Get the flight number to be exchanged: , , This represents the initial flight sequence, position pair. For candidate position pairs , The estimated arrival time for flight number a The estimated arrival time for flight number b. initialization , Calculate time and location difference features , For the aircraft type of flight number a and flight number b Perform one-hot encoding and fill in the blanks. , Calculate and normalize the original position difference. .

4. The method according to claim 2, characterized in that, The delay-improved prediction value prediction processing is based on the delay-improved prediction value. Sort candidate position pairs in descending order Prioritize positions with the expected large improvement, calculate the actual delay of the corresponding new sequence under the scheduler, and use the actual observations as the target to update the learner.

5. The method according to claim 3, characterized in that, The lightweight feedforward neural network self-guided learner is used to learn the feature vector. The nonlinear mapping relationship between the prediction value and the delay improvement is achieved by adopting an online learning mechanism during operation, which continuously updates the parameters to realize dynamic adaptive capability. In terms of forward propagation, the calculation process is represented as follows: ， in, The predicted value for the delay improvement. For the input vector, , For the input layer dimension, and For different hidden layer parameters, , , For the hidden layer dimension, and For different output layer parameters, , , The hyperbolic tangent activation function is used. The weight parameters use the Xavier initialization strategy, which includes: ， in, This indicates that the mean is 0 and the variance is 0. The normal distribution; Regarding the loss function, the single-sample mean squared error is used, expressed as: ， in, This is the single-sample mean squared error. Improve observations to achieve true time delay; In terms of backpropagation and gradient calculation, backpropagation is performed according to the chain rule, calculating the gradient layer by layer, corresponding to the error terms involved. Represented as: ， The process of calculating the parameter gradient is expressed as follows: ， ， ， ， ， ， in, , This represents element-wise multiplication. This represents the outer product operation. This is an intermediate quantity; Regarding the parameter update mechanism, after obtaining the gradient, the parameters are updated using the stochastic gradient descent method, which is represented as follows: ， ， ， ， in, This is the learning rate.

6. The method according to claim 1, characterized in that, The value network is also configured with perturbation and escape strategies from local optima, corresponding to: When an iteration does not accept any remove-reinsert actions, a random perturbation is performed, accepting a slightly worse solution with a small probability, to increase search diversity.

7. The method according to claim 1, characterized in that, Once the maximum number of iterations is reached or the delayed convergence condition is met, the value network outputs the optimal flight scheduling sequence. There is also a corresponding minimum total delay. .

8. A flight scheduling device with autonomous guidance function, characterized in that, The device includes: The acquisition unit is used to acquire a set of flight data to be processed, and then sort it in ascending order based on the corresponding estimated arrival times to obtain a set of initial flight sequences. An update unit is used to update the initial total delay of the value network according to the flight sequence, wherein the value network is used to predict the delay improvement brought about by flight exchange operations and output the optimal flight sequence; An iterative unit is used to import the flight sequence as an initial solution into the value network for iterative optimization. In the iterative optimization stage, candidate removal-re-insertion pairs are randomly generated for exchange simulation, and a constraint detection function is used to ensure that the current sequence meets the flight position offset limit. Subsequently, a lightweight feedforward neural network autonomously guides the learner to predict the delay improvement prediction value of each removal-re-insertion pair to determine the candidate flight sequence that is most likely to reduce the total delay and be output as the optimal flight sequence. An extraction unit is used to extract the processing results output by the value network.

9. A processing device, characterized in that, The method includes a processor and a memory, wherein the memory stores a computer program, and the processor executes the method as described in any one of claims 1 to 7 when it invokes the computer program in the memory.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions adapted for loading by a processor to perform the method of any one of claims 1 to 7.

Images