Intelligent airport scene operation scheduling system based on virtual simulation

Abstract

The invention discloses an intelligent airport scene operation scheduling system based on virtual simulation, and belongs to the field of airport resource scheduling. Comprising an acquisition preprocessing module, a modeling simulation module, a demand analysis module, an environment monitoring module, a traffic flow optimization module, a detection early warning module, a decision support module, a monitoring feedback module, a dynamic scheduling module, a command execution module and a performance evaluation module. The method can prevent problems in advance instead of passively dealing with the problems, effectively reduces the operation pressure and resource conflicts in peak periods, reduces the occurrence of congestion conditions, can make more economical and efficient decisions in resource use, reduces the extra cost caused by resource waste or improper allocation, remarkably improves the flexibility of the system in dealing with emergencies, and improves the safety of the system. The problem of falling into local optimum is avoided, the robustness and adaptability are greatly improved, the continuity and smoothness of airport operation are ensured, and large-scale delay is avoided.

Description

Technical Field

[0001] The invention relates to the field of airport resource scheduling, and in particular to an intelligent airport field operation scheduling system based on virtual simulation. Background Art

[0002] In the operation of modern airports, the dispatching and management of airport scenes is a vital link. The airport scene includes the taxiing of aircraft, the operation of ground handling vehicles, the allocation of parking stands and other ground service activities. With the rapid development of the aviation transportation industry, the complexity and pressure faced by the airport scene operation dispatching system are increasing. As an important node in air transportation, the operation efficiency of the airport directly affects the flight punctuality rate, passenger experience and the operating costs of airlines. In particular, the airport scene, including aircraft taxiing, parking stand allocation and ground handling vehicle dispatching, involves the efficient configuration and dynamic adjustment of multiple resources. Due to the dynamic, complex and uncertain nature of the airport scene, traditional dispatching methods have obvious limitations in responding to emergencies, optimizing resource utilization and improving operational efficiency.

[0003] The existing intelligent airport surface operation and dispatching system cannot prevent problems in advance, which increases the operating pressure and resource conflicts during peak hours, and the additional costs caused by resource waste or improper allocation are high. In addition, the existing intelligent airport surface operation and dispatching system has low flexibility in dealing with emergencies and is prone to falling into local optimal problems, and cannot guarantee the continuity and smoothness of airport operations. For this reason, we propose an intelligent airport surface operation and dispatching system based on virtual simulation. Summary of the invention

[0004] The purpose of the present invention is to solve the defects in the prior art and to propose an intelligent airport scene operation dispatching system based on virtual simulation.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] An intelligent airport operation and dispatching system based on virtual simulation includes a collection and preprocessing module, a modeling and simulation module, a demand analysis module, an environmental monitoring module, a traffic flow optimization module, a detection and early warning module, a decision support module, a monitoring and feedback module, a dynamic dispatching module, a command and execution module, and a performance evaluation module;

[0007] The collection and preprocessing module is used to collect real-time operation data inside and outside the airport, and preprocess the collected operation data;

[0008] The modeling and simulation module is used to create a three-dimensional virtual airport scene model based on the processed operation data, and simulate various airport scene activities according to the input of different scheduling schemes;

[0009] The demand analysis module is used to collect and analyze flight plans, predict flight take-off and landing times and their demand for surface resources;

[0010] The environmental monitoring module is used to monitor weather changes in real time, predict short-term weather changes, and issue early warnings for special weather conditions;

[0011] The traffic flow optimization module is used to simulate and optimize the flow path and scheduling scheme of ground traffic;

[0012] The detection and early warning module is used to analyze the scene operation situation in real time and detect potential safety hazards;

[0013] The decision support module is used to generate and evaluate multiple scheduling plans and adjust the scheduling plans in real time;

[0014] The monitoring feedback module is used to continuously monitor the field operation data and dynamically modify the operation strategy according to the real-time data feedback;

[0015] The dynamic scheduling module is used to dynamically schedule airport resources according to real-time flight demand and ground traffic conditions;

[0016] The command execution module is used to transmit dispatch instructions to relevant personnel and automated equipment, and to coordinate intelligent and automated airport scene dispatch through various modules;

[0017] The performance evaluation module is used to evaluate the overall performance of airport surface operations and record the scheduling performance of the system in different scenarios.

[0018] As a further solution of the present invention, the specific steps of preprocessing the collected operation data by the collection preprocessing module are as follows:

[0019] S1.1: The acquisition preprocessing module obtains real-time operation data inside and outside the airport from multiple different sensors and cameras, and collects different sets of multi-source data at different times and aligns them in time through interpolation. Each frame of image data collected by the camera is input into the pre-trained Faster R-CNN model.

[0020] S1.2: The Faster R-CNN model first extracts the feature maps of each group of input images through a convolutional neural network, then generates multiple candidate regions containing objects and the bounding box of each object from the feature maps through RPN, and evaluates whether the region contains the object. Based on the evaluation results, it eliminates redundant candidate boxes through non-maximum suppression.

[0021] S1.3: The target object area is cropped from the original image according to the generated bounding box, and the depth features corresponding to the cropped image area are extracted through a deep convolutional network. Based on the extracted depth features of the target object, the cosine similarity between the targets under different cameras is calculated. When the similarity is higher than the preset threshold, the two groups of targets are judged to be the same object, otherwise, they are different objects. After the matching is completed, the current state of the target is predicted based on the previous state of the target, and after the prediction, the predicted value is corrected using the data from the multi-source sensors.

[0022] As a further solution of the present invention, the demand analysis module predicts the flight take-off and landing times and their demand for surface resources in the following specific steps:

[0023] S2.1: Collect historical flight flow data, align data from different sources at a uniform time interval, normalize the data, split the historical flight flow data into a training set and a validation set, select a set of sliding windows of length T, and create input-output pairs;

[0024] S2.2: The input data of the training set is fed into the LSTM model. The LSTM model then processes the input at each moment in turn, updates the hidden state at the current moment according to the input and the hidden state at the previous moment, and finally obtains the predicted flight flow value based on the hidden state output at the current moment. The loss value between the predicted result and the true label is calculated by the MSE function.

[0025] S2.3: Calculate the gradient of the loss value for each group of LSTM parameters through the chain rule, and update the model weights through the gradient descent algorithm based on the calculated gradient. After each training cycle, use the validation set to evaluate the model and verify the change in loss. If the model is overfitting, adjust the model's hyperparameters.

[0026] S2.4: Repeatedly train and validate the LSTM model until the preset iteration time is reached or the validation set loss no longer decreases, then input the latest flight flow data into the trained LSTM model, and predict the future flight flow curve through model forward propagation;

[0027] S2.5: When the predicted flight traffic exceeds the preset threshold, the time period is determined to be a peak period. If the predicted peak traffic lasts for a long time, an early warning is issued, indicating that congestion may occur.

[0028] As a further solution of the present invention, the traffic flow optimization module optimizes the flow path and scheduling scheme of ground traffic in the following specific steps:

[0029] S3.1: Collect airport basic operation data and extract basic information of aircraft, parking stands and ground handling vehicles. Based on the obtained information, generate feasible taxi paths for each aircraft from its current position to each group of parking stands, a list of candidate parking stands and a service path for ground handling vehicles through a cascade algorithm;

[0030] S3.2: According to the taxiing path of each aircraft, the list of candidate parking spaces and the service path of ground handling vehicles, a parking space that meets the conditions is randomly assigned to each aircraft, and different task execution sequences and paths of ground handling vehicles are generated based on the random assignment results. The randomly generated data is used as the preliminary assignment plan;

[0031] S3.3: Take the preliminary allocation plan as the root node, and generate multiple groups of child nodes by randomly adjusting the aircraft taxi path, parking space allocation, and ground handling vehicle service path. At the same time, build an exploration tree based on the root node and each group of child nodes, then calculate the UCB value of each group of child nodes, and select the child node with the highest UCB value layer by layer according to the upper confidence limit selection strategy;

[0032] S3.4: When the selected child node is not expanded, stop selecting, and generate multiple groups of new allocation schemes for the selected child nodes through random perturbations. At the same time, add each group of generated allocation schemes as a new child node to the exploration tree, perform random simulation on the new child node, and trace the simulation results back to the root node, and update the statistical information of each child node on the same path;

[0033] S3.5: Repeat the selection, expansion, simulation and backtracking until the preset update time is reached, and use the allocation plan corresponding to the child node with the highest reward value in the simulation in the final exploration tree as the optimal scheduling plan, and apply it to the flow path control of actual ground transportation.

[0034] As a further solution of the present invention, the airport basic operation data described in S3.1 specifically includes the locations of each group of facilities such as airport runways, taxiways, parking stands and boarding gates, the current location information of each aircraft, the planned take-off time and the destination parking stand, as well as the number, type, current mission and route of vehicles.

[0035] As a further solution of the present invention, the specific steps of the decision support module adjusting the scheduling plan in real time are as follows:

[0036] S4.1: Collect airport surface operation information based on the optimal scheduling scheme in real time, and create corresponding state space and action space according to the airport operation status and executable scheduling actions, then calculate the probability of transitioning from state s to the next state s' after executing action a, and evaluate the reward obtained by executing action a in state s;

[0037] S4.2: Based on the probability of selecting action a in state s, generate multiple sets of different scheduling schemes by sampling or simulating different strategies, and calculate the expected total reward obtained by executing actions according to different strategies through the value function to evaluate the expected return of each state;

[0038] S4.3: Based on the Bellman optimal equation, the value function is continuously updated through recursive iteration until the latest expected total reward and the previous expected total reward change value converge to within the preset threshold. Then the iteration is stopped, and according to the current state of the airport scene, the strategy with the largest value function is selected to adjust the current scheduling plan and recommend it to the dispatcher for execution.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] 1. The present invention collects and processes historical flight flow data, divides the historical flight flow data into a training set and a validation set, creates input-output pairs, and feeds the input data of the training set into the LSTM model. The LSTM model then processes the input at each moment in turn, updates the hidden state at the current moment according to the input and the hidden state at the previous moment, and outputs the predicted flight flow value based on the hidden state at the current moment. The loss value between the predicted result and the true label is calculated by the MSE function. The gradient of the loss value for each group of LSTM parameters is calculated by the chain rule, and the weight of the model is updated by the gradient descent algorithm based on the calculated gradient. After each training cycle, the model is evaluated using the validation set to verify the change in the loss. If the model has overfitting, adjust the model's hyperparameters, and repeatedly train and verify the LSTM model until the preset iteration time is reached or the verification set loss no longer decreases. Then, input the latest flight traffic data into the trained LSTM model, and predict the future flight traffic curve through the model forward propagation. When the predicted flight traffic exceeds the preset threshold, the time period is determined to be a peak period. If the predicted peak traffic duration is long, an early warning is issued, indicating that congestion may occur. This can prevent problems in advance rather than responding passively, effectively reduce peak operating pressure and resource conflicts, reduce congestion, make more economical and efficient decisions on resource use, and reduce additional costs caused by resource waste or improper allocation.

[0041] 2. The present invention collects airport basic operation data, and randomly allocates a parking space that meets the conditions to each aircraft according to each group of information obtained, and generates different task execution orders and paths for ground handling vehicles according to the random allocation results. At the same time, the randomly generated data is used as a preliminary allocation plan, and the preliminary allocation plan is used as the root node. Multiple groups of child nodes are generated through random adjustment, and an exploration tree is constructed based on the root node and each group of child nodes. The UCB value of each group of child nodes is then calculated, and the child node with the highest UCB value is selected layer by layer according to the upper confidence limit selection strategy. When the selected child node is not expanded, the selection is stopped, and multiple groups of new allocation plans are generated for the selected child nodes through random perturbations. Each group of allocation schemes is added to the exploration tree as a new child node, and the new child nodes are randomly simulated. The simulation results are traced back to the root node, and the statistical information of each child node on the same path is updated. The selection, expansion, simulation and backtracking are repeated until the preset update time is reached and stopped. The allocation scheme corresponding to the child node with the highest reward value in the simulation in the final exploration tree is used as the optimal scheduling scheme. At the same time, it is applied to the flow path control of actual ground transportation, which significantly improves the flexibility of the system in dealing with emergencies and avoids falling into the problem of local optimality. Its robustness and adaptability have been greatly improved, ensuring the continuity and smoothness of airport operations and avoiding large-scale delays. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0043] Figure 1 This is a system block diagram of the intelligent airport field operation dispatching system based on virtual simulation proposed by the present invention. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0045] Example 1

[0046] Reference Figure 1 The intelligent airport scene operation and dispatching system based on virtual simulation includes acquisition and preprocessing module, modeling and simulation module, demand analysis module, environmental monitoring module, traffic flow optimization module, detection and early warning module, decision support module, monitoring and feedback module, dynamic dispatching module, command execution module and performance evaluation module.

[0047] The collection and preprocessing module is used to collect real-time operation data inside and outside the airport and preprocess the collected operation data.

[0048] Specifically, the acquisition preprocessing module obtains real-time operation data inside and outside the airport scene from multiple different sensors and cameras, and collects groups of multi-source data at different times and aligns them in time through interpolation. Each frame of image data collected by the camera is input into the pre-trained Faster R-CNN model. The Faster R-CNN model first extracts the feature map of each group of input images through a convolutional neural network, and then generates multiple candidate regions containing targets and the bounding box of each target from the feature map through RPN, and evaluates whether the region contains the target. Based on the evaluation results, the redundant candidate boxes are eliminated through non-maximum suppression, and the target object area is cropped from the original image according to the generated bounding box. The depth features corresponding to the cropped image area are extracted through a deep convolutional network. Based on the extracted depth features of the target object, the cosine similarity between the targets under different cameras is calculated. When the similarity is higher than the preset threshold, the two groups of targets are judged to be the same object, otherwise, they are different objects. After the matching is completed, the current state of the target is predicted according to the previous state of the target, and after the prediction, the predicted value is corrected using the data of the multi-source sensors.

[0049] The modeling and simulation module is used to create a three-dimensional virtual airport scene model based on the processed operation data, and simulate various airport scene activities according to the input of different scheduling schemes; the demand analysis module is used to collect and analyze flight plans, predict flight take-off and landing times and their demand for surface resources; the environmental monitoring module is used to monitor weather changes in real time, predict short-term weather changes, and issue early warnings for special weather conditions.

[0050] Specifically, historical flight traffic data is collected, and data from different sources are aligned at a unified time interval. After the data is normalized, the historical flight traffic data is divided into a training set and a validation set. A set of sliding windows with a length of T is selected to create input-output pairs. The input data of the training set is sent to the LSTM model. The LSTM model then processes the input at each moment in turn, updates the hidden state at the current moment according to the input and the hidden state at the previous moment, and outputs the predicted flight traffic value based on the hidden state at the current moment. The loss value between the predicted result and the true label is calculated using the MSE function. The loss value pair is calculated for each LSTM instance using the chain rule. The gradient of the group parameters is calculated, and the weight of the model is updated through the gradient descent algorithm based on the calculated gradient. After each training cycle, the model is evaluated using the validation set to verify the change in loss. If the model is overfitting, the hyperparameters of the model are adjusted. The LSTM model is repeatedly trained and verified until the preset iteration time is reached or the validation set loss no longer decreases. The latest flight traffic data is then input into the trained LSTM model, and the future flight traffic curve is predicted through model forward propagation. When the predicted flight traffic exceeds the preset threshold, the time period is determined to be a peak period. If the predicted peak traffic duration is long, an early warning is issued, indicating that congestion may occur.

[0051] Example 2

[0052] Reference Figure 1 The intelligent airport scene operation and dispatching system based on virtual simulation includes acquisition and preprocessing module, modeling and simulation module, demand analysis module, environmental monitoring module, traffic flow optimization module, detection and early warning module, decision support module, monitoring and feedback module, dynamic dispatching module, command execution module and performance evaluation module.

[0053] The traffic flow optimization module is used to simulate and optimize the flow paths and scheduling plans of ground traffic.

[0054] Specifically, the basic operation data of the airport is collected, and the basic information of the aircraft, parking stands and ground handling vehicles is extracted. According to the obtained groups of information, the feasible taxiing path, the candidate list of parking stands and the service path of the ground handling vehicles for each aircraft from the current position to each group of parking stands are generated through the cascade algorithm. According to the taxiing path, the candidate list of parking stands and the service path of the ground handling vehicles of each aircraft, a parking stand that meets the conditions is randomly assigned to each aircraft, and different task execution orders and paths of the ground handling vehicles are generated according to the random assignment results. At the same time, the randomly generated data is used as the preliminary allocation plan, and the preliminary allocation plan is used as the root node. Multiple groups of child nodes are generated by randomly adjusting the aircraft taxiing path, parking stand allocation and ground handling vehicle service path. At the same time, the root node and each group of child nodes are constructed. Build an exploration tree, then calculate the UCB value of each group of child nodes, and select the child node with the highest UCB value layer by layer according to the upper confidence limit selection strategy. When the selected child node is not expanded, stop selecting, and generate multiple groups of new allocation schemes for the selected child nodes through random perturbations. At the same time, add the generated groups of allocation schemes as new child nodes to the exploration tree, and perform random simulation on the new child nodes. The simulation results are traced back to the root node, and the statistical information of each child node on the same path is updated. Repeat the selection, expansion, simulation and backtracking until the preset update time is reached. The allocation scheme corresponding to the child node with the highest reward value during the simulation in the final exploration tree is used as the optimal scheduling scheme, and it is applied to the flow path control of actual ground transportation.

[0055] In this embodiment, the airport basic operation data specifically includes the location of each group of facilities such as the airport runway, taxiway, parking stand and boarding gate, the current location information of each aircraft, the planned take-off time and the destination parking stand, as well as the number, type, current mission and route of vehicles.

[0056] The detection and early warning module is used to analyze the scene operation status in real time and detect potential safety hazards; the decision support module is used to generate and evaluate multiple scheduling plans and adjust the scheduling plans in real time.

[0057] Specifically, the airport surface operation information based on the optimal scheduling scheme is collected in real time, and the corresponding state space and action space are created according to the operating status of the airport and the executable scheduling actions. Then, the probability of transferring from state s to the next state s' after executing action a is calculated, and the reward obtained by executing action a in state s is evaluated. According to the probability of selecting action a in state s, multiple groups of different scheduling schemes are generated by sampling or simulating different strategies, and the expected total reward obtained by executing actions according to different strategies is calculated through the value function to evaluate the expected return of each state. Based on the Bellman optimal equation, the value function is continuously updated through recursive iteration until the latest expected total reward and the previous expected total reward change value converge to within a preset threshold, then the iteration is stopped, and according to the current state of the airport surface, the strategy with the largest value function is selected to adjust the current scheduling scheme, and it is recommended to the dispatcher for execution.

[0058] The monitoring and feedback module is used to continuously monitor the surface operation data, and dynamically correct the operation strategy according to the real-time data feedback; the dynamic scheduling module is used to dynamically schedule the airport resources according to the real-time flight demand and ground traffic conditions; the command execution module is used to transmit the scheduling instructions to the relevant personnel and automated equipment, and to coordinate the intelligent and automated airport surface scheduling through the modules; the performance evaluation module is used to evaluate the overall performance of the airport surface operation and record the scheduling performance of the system in different scenarios.

Claims

1. Intelligent airport operation dispatching system based on virtual simulation, characterized by: It includes collection preprocessing module, modeling and simulation module, demand analysis module, environmental monitoring module, traffic flow optimization module, detection and warning module, decision support module, monitoring feedback module, dynamic scheduling module, command execution module and performance evaluation module; The collection and preprocessing module is used to collect real-time operation data inside and outside the airport, and preprocess the collected operation data; The modeling and simulation module is used to create a three-dimensional virtual airport scene model based on the processed operation data, and simulate various airport scene activities according to the input of different scheduling schemes; The demand analysis module is used to collect and analyze flight plans, predict flight take-off and landing times and their demand for surface resources; The environmental monitoring module is used to monitor weather changes in real time, predict short-term weather changes, and issue early warnings for special weather conditions; The traffic flow optimization module is used to simulate and optimize the flow path and scheduling scheme of ground traffic; The detection and early warning module is used to analyze the scene operation situation in real time and detect potential safety hazards; The decision support module is used to generate and evaluate multiple scheduling plans and adjust the scheduling plans in real time; The monitoring feedback module is used to continuously monitor the field operation data and dynamically modify the operation strategy according to the real-time data feedback; The dynamic scheduling module is used to dynamically schedule airport resources according to real-time flight demand and ground traffic conditions; The command execution module is used to transmit dispatch instructions to relevant personnel and automated equipment, and to coordinate intelligent and automated airport field dispatch through various modules; The performance evaluation module is used to evaluate the overall performance of airport surface operations and record the scheduling performance of the system in different scenarios.

2. The intelligent airport field operation dispatching system based on virtual simulation according to claim 1 is characterized in that: The specific steps of preprocessing the collected operation data by the collection preprocessing module are as follows: S1.1: The acquisition preprocessing module obtains real-time operation data inside and outside the airport from multiple different sensors and cameras, and collects different sets of multi-source data at different times and aligns them in time through interpolation. Each frame of image data collected by the camera is input into the pre-trained Faster R-CNN model. S1.2: The Faster R-CNN model first extracts the feature maps of each group of input images through a convolutional neural network, then generates multiple candidate regions containing objects and the bounding box of each object from the feature maps through RPN, and evaluates whether the region contains the object. Based on the evaluation results, it eliminates redundant candidate boxes through non-maximum suppression. S1.3: The target object area is cropped from the original image according to the generated bounding box, and the depth features corresponding to the cropped image area are extracted through a deep convolutional network. Based on the extracted depth features of the target object, the cosine similarity between the targets under different cameras is calculated. When the similarity is higher than the preset threshold, the two groups of targets are judged to be the same object, otherwise, they are different objects. After the matching is completed, the current state of the target is predicted based on the previous state of the target, and after the prediction, the predicted value is corrected using the data from the multi-source sensors.

3. The intelligent airport field operation dispatching system based on virtual simulation according to claim 1 is characterized in that: The specific steps of the demand analysis module for predicting flight take-off and landing times and their demand for surface resources are as follows: S2.1: Collect historical flight flow data, align data from different sources at a uniform time interval, normalize the data, split the historical flight flow data into a training set and a validation set, select a set of sliding windows of length T, and create input-output pairs; S2.2: The input data of the training set is fed into the LSTM model. The LSTM model then processes the input at each moment in turn, updates the hidden state at the current moment according to the input and the hidden state at the previous moment, and finally obtains the predicted flight flow value based on the hidden state output at the current moment. The loss value between the predicted result and the true label is calculated by the MSE function. S2.3: Calculate the gradient of the loss value for each group of LSTM parameters through the chain rule, and update the model weights through the gradient descent algorithm based on the calculated gradient. After each training cycle, use the validation set to evaluate the model and verify the change in loss. If the model is overfitting, adjust the model's hyperparameters. S2.4: Repeatedly train and validate the LSTM model until the preset iteration time is reached or the validation set loss no longer decreases, then input the latest flight flow data into the trained LSTM model, and predict the future flight flow curve through model forward propagation; S2.5: When the predicted flight traffic exceeds the preset threshold, the time period is determined to be a peak period. If the predicted peak traffic lasts for a long time, an early warning is issued, indicating that congestion may occur.

4. The intelligent airport field operation dispatching system based on virtual simulation according to claim 3 is characterized in that: The traffic flow optimization module optimizes the flow path and scheduling plan of ground traffic in the following specific steps: S3.1: Collect airport basic operation data and extract basic information of aircraft, parking stands and ground handling vehicles. Based on the obtained information, generate feasible taxi paths for each aircraft from its current position to each group of parking stands, a list of candidate parking stands and a service path for ground handling vehicles through a cascade algorithm; S3.2: According to the taxiing path of each aircraft, the list of candidate parking spaces and the service path of ground handling vehicles, a parking space that meets the conditions is randomly assigned to each aircraft, and different task execution sequences and paths of ground handling vehicles are generated based on the random assignment results. The randomly generated data is used as the preliminary assignment plan; S3.3: Take the preliminary allocation plan as the root node, and generate multiple groups of child nodes by randomly adjusting the aircraft taxi path, parking space allocation, and ground handling vehicle service path. At the same time, build an exploration tree based on the root node and each group of child nodes, then calculate the UCB value of each group of child nodes, and select the child node with the highest UCB value layer by layer according to the upper confidence limit selection strategy; S3.4: When the selected child node is not expanded, stop selecting, and generate multiple groups of new allocation schemes for the selected child nodes through random perturbations. At the same time, add each group of generated allocation schemes as a new child node to the exploration tree, perform random simulation on the new child node, and trace the simulation results back to the root node, and update the statistical information of each child node on the same path; S3.5: Repeat the selection, expansion, simulation and backtracking until the preset update time is reached, and use the allocation plan corresponding to the child node with the highest reward value in the simulation in the final exploration tree as the optimal scheduling plan, and apply it to the flow path control of actual ground transportation.

5. The intelligent airport field operation dispatching system based on virtual simulation according to claim 4 is characterized in that: The airport basic operation data described in S3.1 specifically includes the location of each group of facilities such as airport runways, taxiways, parking stands and boarding gates, the current location information of each aircraft, the planned take-off time and the destination parking stand, as well as the number, type, current mission and route of vehicles.

6. The intelligent airport field operation dispatching system based on virtual simulation according to claim 4 is characterized in that: The specific steps of the decision support module to adjust the scheduling plan in real time are as follows: S4.1: Collect airport surface operation information based on the optimal scheduling scheme in real time, and create corresponding state space and action space according to the airport operation status and executable scheduling actions, then calculate the probability of transitioning from state s to the next state s' after executing action a, and evaluate the reward obtained by executing action a in state s; S4.2: Based on the probability of selecting action a in state s, generate multiple sets of different scheduling schemes by sampling or simulating different strategies, and calculate the expected total reward obtained by executing actions according to different strategies through the value function to evaluate the expected return of each state; S4.3: Based on the Bellman optimal equation, the value function is continuously updated through recursive iteration until the latest expected total reward and the previous expected total reward change value converge to within the preset threshold. Then the iteration is stopped, and according to the current state of the airport scene, the strategy with the largest value function is selected to adjust the current scheduling plan and recommend it to the dispatcher for execution.

Images