Application system and method of meteorological information in flight scheduling

Abstract

The invention belongs to the technical field of flight scheduling, and discloses an application system of meteorological information in flight scheduling, which is characterized in that a meteorological data acquisition module provides standardized meteorological data for a feature extractor; a meteorological feature vector obtained by the feature extractor is used as an input basis of an influence evaluation module to obtain a flight drop preparation probability; the flight drop probability is input into a flight grouping unit for dynamic grouping; combining a grouping result with a multi-target optimizer to generate an initial scheduling scheme; the weight adjustment unit dynamically adjusts the optimization target based on the execution feedback; the scheduling strategy generator optimizes the scheduling scheme according to the adjusted weight coefficient; and the execution control module is responsible for implementing the scheme and feeding back an execution result to the weight adjustment unit.

Description

Technical Field

[0001] This application relates to the field of flight scheduling technology, and more specifically, to a system and method for applying meteorological information in flight scheduling. Background Technology

[0002] With the rapid development of the air transport industry, the complexity of flight scheduling management is increasing. Existing flight scheduling systems mainly suffer from the following technical problems:

[0003] First, existing flight scheduling systems process meteorological information in a relatively simple way. Most systems use only a single meteorological data source and rely on threshold judgments for impact assessment, which fails to comprehensively and accurately evaluate the actual impact of complex weather conditions on flight operations. For example, when dealing with complex weather along flight routes, the lack of multi-source meteorological data fusion and analysis capabilities can easily lead to errors in flight detour decisions.

[0004] Second, existing scheduling systems generally employ a fixed-weight single-objective optimization method. This method struggles to balance multiple objectives, such as flight punctuality, operating costs, and passenger service quality, in practical applications. Especially under complex weather conditions, it cannot dynamically adjust the priority of each objective based on actual circumstances, resulting in insufficient adaptability of the scheduling scheme.

[0005] Third, traditional flight scheduling methods often optimize flights as independent entities, neglecting the characteristics of flight groups. This approach ignores the correlation between flights, making it inefficient when handling high-density flight flows and prone to cascading delays, thus affecting overall operational efficiency.

[0006] Fourth, existing systems have significant shortcomings in execution monitoring and feedback optimization. Most systems employ static scheduling strategies, lacking real-time monitoring and dynamic adjustment mechanisms for the execution process. This makes them unable to effectively address uncertainties during execution, reducing the reliability of the scheduling scheme.

[0007] To overcome the aforementioned technical problems, there is an urgent need to provide a flight scheduling system that can fully utilize multi-source meteorological data, possess multi-objective dynamic optimization capabilities, support group-based scheduling strategies, and achieve closed-loop feedback optimization. Summary of the Invention

[0008] 1. The technical problems to be solved.

[0009] The purpose of this application is to provide a meteorological information application system for flight scheduling, which solves the technical problems mentioned in the background.

[0010] 2. Technical solution.

[0011] The technical solution of this application provides a system for the application of meteorological information in flight scheduling, including...

[0012] The meteorological data acquisition module is used to collect multi-source meteorological data, including: real-time data from airport meteorological stations, airway meteorological radar data, satellite cloud imagery, and upper-air meteorological sounding data; the real-time data from airport meteorological stations is used to obtain visibility parameters at the airport; the airway meteorological radar data is used to monitor precipitation intensity distribution along the airway; the satellite cloud imagery is used to analyze the real-time development trend of weather systems; and the upper-air meteorological sounding data is used to obtain the temperature field distribution at the cruising altitude.

[0013] The data preprocessing unit uses a cubic spline interpolation algorithm to unify the spatiotemporal resolution of the multi-source meteorological data; it uses wavelet transform to filter noisy data to achieve data quality control; and it uses an anomaly detection method based on Gaussian distribution to identify and correct outliers.

[0014] The data preprocessing unit includes a data synchronization subunit, a data standardization subunit, and a data fusion subunit. Specifically: the data synchronization subunit performs the following processing: timestamp alignment of real-time airport meteorological station data at 1-minute intervals; time-series resampling of route meteorological radar data at 5-minute intervals; time-series interpolation of satellite cloud imagery data at 15-minute intervals; and linear interpolation of upper-air meteorological detection data at 30-minute intervals. The data standardization subunit performs the following processing: min-max standardization of visibility data after logarithmic transformation; Z-score standardization of wind speed data; decimal standardization of temperature data; and mean normalization of air pressure data. The data fusion subunit performs the following processing: noise removal of multi-source data using the Kalman filtering algorithm; dimensionality reduction of the processed data using principal component analysis, retaining features with a cumulative contribution rate of 95%; repair of missing data using the iterative nearest neighbor method; outlier detection based on Mahalanobis distance, and correction through local weighted regression. The scheme clarifies the specific steps and methods for data preprocessing; adopts differentiated processing strategies for different types of meteorological data; ensures data quality and usability; and provides a standardized data foundation for subsequent feature extraction.

[0015] The feature extractor uses an improved ResNet deep network structure to extract local features from meteorological data; it achieves unified processing of features at different scales through bilinear interpolation; it calculates the weight relationship between different features using a self-attention mechanism; and it captures the temporal evolution features of meteorological elements based on an LSTM network.

[0016] The impact assessment module constructs a flight operation safety margin assessment model that takes into account the impact of meteorology; designs a flight delay risk assessment method based on fuzzy comprehensive evaluation; establishes a delay propagation prediction model that considers flight connectivity; and uses grey prediction theory to assess the probability of flight diversion.

[0017] The flight grouping unit calculates the initialization scheme for flight grouping based on the similarity of waypoint sequences; implements dynamic grouping of flights using the K-means clustering algorithm; determines flight priority weights using the analytic hierarchy process; and analyzes resource competition relationships using graph theory methods.

[0018] A multi-objective optimizer is used to establish an optimization function with minimizing total delay time as the primary objective; cost constraints considering fuel consumption are constructed; a penalty factor based on flight punctuality rate is designed; and the NSGA-III algorithm is used to solve the multi-objective optimization problem.

[0019] The weight adjustment unit is used to train the initial weights of the optimization targets based on historical data, and dynamically adjust the weight coefficients of each optimization target according to real-time scheduling feedback through a reinforcement learning method. The reinforcement learning method adopts a deep Q-network, uses the scheduling execution effect as a reward signal, and optimizes the weight adjustment strategy through an experience replay mechanism. The historical data includes scheduling schemes and their execution effects under different meteorological conditions.

[0020] The scheduling strategy generator generates flight takeoff and landing sequences based on dynamic programming algorithms; implements route dynamic planning using the A* algorithm; optimizes ground support resource allocation through the Hungarian algorithm; and employs heuristic algorithms to handle strategy adjustments in unforeseen circumstances.

[0021] The execution control module establishes an event-driven scheduling instruction issuance mechanism; it achieves real-time monitoring of scheduling execution through digital twin technology; it uses a sliding time window method for dynamic adjustment; and it employs a fuzzy control algorithm to implement feedback compensation.

[0022] The system is achieved through the collaborative work of the above modules: the meteorological data acquisition module provides standardized meteorological data to the feature extractor; the output of the feature extractor serves as the input to the impact assessment module; the impact assessment results drive the flight grouping unit to perform dynamic grouping; the grouping results are combined with the multi-objective optimizer to generate an initial scheduling scheme; the weight adjustment unit dynamically adjusts the optimization objectives based on execution feedback; the scheduling strategy generator optimizes the scheduling scheme according to the adjusted weight coefficients; and the execution control module is responsible for implementing the scheme and feeding back the execution results to the weight adjustment unit.

[0023] Furthermore, the specific structure of the ResNet deep network includes:

[0024] The input layer uses a 5×5 convolutional kernel to extract local features;

[0025] The intermediate layer contains 3 residual blocks, and each residual block consists of two 3×3 convolutional layers;

[0026] The feature fusion layer uses 1×1 convolution to achieve channel dimensionality reduction;

[0027] The output layer generates feature vectors through fully connected layers.

[0028] Furthermore, the flight operation safety margin assessment model includes:

[0029] Safety margin calculation formula: S=w1×V+w2×C+w3×W+w4×T where V is the visibility impact factor, C is the cloud height impact factor, W is the wind speed impact factor, T is the temperature impact factor, and w1-w4 are the corresponding weight coefficients.

[0030] The delay risk level is divided into five levels, and the risk is quantified using fuzzy membership functions.

[0031] The probability of flight diversion is predicted using the grey prediction model GM(1,1).

[0032] Furthermore, the flight grouping unit includes:

[0033] Similarity of waypoint sequences: Sim(i,j) = 2 × N com / (N i +N j Wherein, N com N represents the number of common waypoints for flights i and j. i and N j These represent the total number of waypoints for the two flights;

[0034] Priority score: P = α × L + β × N + γ × C, where L is the flight class coefficient, N is the passenger number coefficient, C is the connecting flight coefficient, and α, β, and γ are weight coefficients.

[0035] Furthermore, the NSGA-III algorithm is: Minimum F = ω1 × ∑D i +ω2×∑F i +ω3×(1-R p )+ω4×∑C i Where: Di is the delay time of the i-th flight; F i R represents the fuel consumption of the i-th flight. p For flight punctuality rate; C i ω1-ω4 are the operating costs; ω1-ω4 are the dynamically adjusted weighting coefficients.

[0036] Furthermore, the structure of the deep reinforcement learning network includes:

[0037] State space S: contains 9 key features of the current scheduling state; Action space A: contains 5 optional strategies for weight adjustment; Reward function R: R = λ1×ΔD + λ2×ΔF + λ3×ΔR + λ4×ΔC, where ΔD, ΔF, ΔR, and ΔC represent the degree of improvement of each optimization objective.

[0038] Furthermore, the scheduling policy generation method includes:

[0039] The takeoff and landing timing sequence is generated using an improved dynamic programming algorithm: the time is discretized into 5-minute intervals; the state transition equation is designed as: f(t+1)=min{f(t)+c(t,t+1)}; and a penalty term for meteorological constraints is introduced.

[0040] Furthermore, the execution control method includes:

[0041] Set the sliding time window length to 120 minutes;

[0042] Define the deviation threshold δ for triggering adjustments: trigger local adjustments when the actual delay exceeds 20% of the expected delay; trigger global re-optimization when weather conditions change abruptly.

[0043] Furthermore, the data preprocessing method includes:

[0044] Time synchronization is achieved using least squares fitting;

[0045] Spatial interpolation uses Kriging interpolation.

[0046] Outlier identification criteria: |x-μ|>3σ, where μ is the mean and σ is the standard deviation.

[0047] Furthermore, the route dynamic planning method includes:

[0048] Cost function design: g(n)=α×D(n)+β×W(n) where D(n) is the distance cost and W(n) is the meteorological cost;

[0049] Heuristic function design: h(n) = γ × H(n) where H(n) is the straight-line distance from the current point to the target point.

[0050] 3. Beneficial effects.

[0051] One or more technical solutions provided in this application have at least the following technical effects or advantages.

[0052] The technical effects of this invention are as follows: by constructing a multi-source meteorological data fusion network based on deep learning, and combining dynamic grouping and multi-objective collaborative optimization methods, the system's scheduling decision accuracy under complex meteorological conditions is increased from 75% to 92%, the response time to emergencies is shortened to less than 4 minutes, the computational efficiency of the scheduling scheme is improved by 85%, and the overall reliability and stability of the system are significantly enhanced.

[0053] The economic benefits of this invention are as follows: based on accurate meteorological impact assessment and intelligent scheduling strategies, it achieves an average reduction in flight fuel consumption, an improvement in flight punctuality, and an increase in the efficiency of ground support resource utilization. Actual verification has shown that it can save operating costs for a single hub airport every year, demonstrating significant economic value.

[0054] The social benefits of this invention are: by improving the safety and punctuality of flight operations, passenger satisfaction is enhanced, and carbon emissions are reduced. It provides a replicable and scalable technical solution for the intelligent and green development of the air transport industry, and has significant demonstrative significance and promotional value. Attached Figure Description

[0055] Figure 1 This is a schematic diagram of the application system for meteorological information in flight scheduling, as disclosed in a preferred embodiment of this application. Detailed Implementation

[0056] The present application will be further described in detail below with reference to the accompanying drawings.

[0057] Reference Figure 1 This application provides a system for applying meteorological information in flight scheduling, including...

[0058] The meteorological data acquisition module is used to collect multi-source meteorological data, including: real-time data from airport meteorological stations, airway meteorological radar data, satellite cloud imagery, and upper-air meteorological sounding data; the real-time data from airport meteorological stations is used to obtain visibility parameters at the airport; the airway meteorological radar data is used to monitor precipitation intensity distribution along the airway; the satellite cloud imagery is used to analyze the real-time development trend of weather systems; and the upper-air meteorological sounding data is used to obtain the temperature field distribution at the cruising altitude.

[0059] The data preprocessing unit uses a cubic spline interpolation algorithm to unify the spatiotemporal resolution of the multi-source meteorological data; it uses wavelet transform to filter noisy data to achieve data quality control; and it uses an anomaly detection method based on Gaussian distribution to identify and correct outliers.

[0060] The feature extractor uses an improved ResNet deep network structure to extract local features from meteorological data; it achieves unified processing of features at different scales through bilinear interpolation; it calculates the weight relationship between different features using a self-attention mechanism; and it captures the temporal evolution features of meteorological elements based on an LSTM network.

[0061] The impact assessment module constructs a flight operation safety margin assessment model that takes into account the impact of meteorology; designs a flight delay risk assessment method based on fuzzy comprehensive evaluation; establishes a delay propagation prediction model that considers flight connectivity; and uses grey prediction theory to assess the probability of flight diversion.

[0062] The flight grouping unit calculates the initialization scheme for flight grouping based on the similarity of waypoint sequences; implements dynamic grouping of flights using the K-means clustering algorithm; determines flight priority weights using the analytic hierarchy process; and analyzes resource competition relationships using graph theory methods.

[0063] A multi-objective optimizer is used to establish an optimization function with minimizing total delay time as the primary objective; cost constraints considering fuel consumption are constructed; a penalty factor based on flight punctuality rate is designed; and the NSGA-III algorithm is used to solve the multi-objective optimization problem.

[0064] The weight adjustment unit is used to train the initial weights of the optimization targets based on historical data, and dynamically adjust the weight coefficients of each optimization target according to real-time scheduling feedback through a reinforcement learning method. The reinforcement learning method adopts a deep Q-network, uses the scheduling execution effect as a reward signal, and optimizes the weight adjustment strategy through an experience replay mechanism. The historical data includes scheduling schemes and their execution effects under different meteorological conditions.

[0065] The scheduling strategy generator generates flight takeoff and landing sequences based on dynamic programming algorithms; implements route dynamic planning using the A* algorithm; optimizes ground support resource allocation through the Hungarian algorithm; and employs heuristic algorithms to handle strategy adjustments in unforeseen circumstances.

[0066] The execution control module establishes an event-driven scheduling instruction issuance mechanism; it achieves real-time monitoring of scheduling execution through digital twin technology; it uses a sliding time window method for dynamic adjustment; and it employs a fuzzy control algorithm to implement feedback compensation.

[0067] The system is achieved through the collaborative work of the above modules: the meteorological data acquisition module provides standardized meteorological data to the feature extractor; the output of the feature extractor serves as the input to the impact assessment module; the impact assessment results drive the flight grouping unit to perform dynamic grouping; the grouping results are combined with the multi-objective optimizer to generate an initial scheduling scheme; the weight adjustment unit dynamically adjusts the optimization objectives based on execution feedback; the scheduling strategy generator optimizes the scheduling scheme according to the adjusted weight coefficients; and the execution control module is responsible for implementing the scheme and feeding back the execution results to the weight adjustment unit.

[0068] Furthermore, the specific structure of the ResNet deep network includes:

[0069] The input layer uses a 5×5 convolutional kernel to extract local features;

[0070] The intermediate layer contains 3 residual blocks, and each residual block consists of two 3×3 convolutional layers;

[0071] The feature fusion layer uses 1×1 convolution to achieve channel dimensionality reduction;

[0072] The output layer generates feature vectors through fully connected layers.

[0073] Furthermore, the flight operation safety margin assessment model includes:

[0074] Safety margin calculation formula: S=w1×V+w2×C+w3×W+w4×T where V is the visibility impact factor, C is the cloud height impact factor, W is the wind speed impact factor, T is the temperature impact factor, and w1-w4 are the corresponding weight coefficients.

[0075] The delay risk level is divided into five levels, and the risk is quantified using fuzzy membership functions.

[0076] The probability of flight diversion is predicted using the grey prediction model GM(1,1).

[0077] Furthermore, the flight grouping unit includes:

[0078] Similarity of waypoint sequences: Sim(i,j) = 2 × N com / (N i +N j Wherein, N com N represents the number of common waypoints for flights i and j. i and N j These represent the total number of waypoints for the two flights;

[0079] Priority score: P = α × L + β × N + γ × C, where L is the flight class coefficient, N is the passenger number coefficient, C is the connecting flight coefficient, and α, β, and γ are weight coefficients.

[0080] Furthermore, the NSGA-III algorithm is: Minimum F = ω1 × ∑D i +ω2×∑F i +ω3×(1-R p )+ω4×∑C i Where: Di is the delay time of the i-th flight; F i R represents the fuel consumption of the i-th flight. p For flight punctuality rate; C i ω1-ω4 are the operating costs; ω1-ω4 are the dynamically adjusted weighting coefficients.

[0081] Furthermore, the structure of the deep reinforcement learning network includes:

[0082] State space S: contains 9 key features of the current scheduling state; Action space A: contains 5 optional strategies for weight adjustment; Reward function R: R = λ1×ΔD + λ2×ΔF + λ3×ΔR + λ4×ΔC, where ΔD, ΔF, ΔR, and ΔC represent the degree of improvement of each optimization objective.

[0083] Furthermore, the scheduling policy generation method includes:

[0084] The takeoff and landing timing sequence is generated using an improved dynamic programming algorithm: the time is discretized into 5-minute intervals; the state transition equation is designed as: f(t+1)=min{f(t)+c(t,t+1)}; and a penalty term for meteorological constraints is introduced.

[0085] Furthermore, the execution control method includes:

[0086] Set the sliding time window length to 120 minutes;

[0087] Define the deviation threshold δ for triggering adjustments: trigger local adjustments when the actual delay exceeds 20% of the expected delay; trigger global re-optimization when weather conditions change abruptly.

[0088] Furthermore, the data preprocessing method includes:

[0089] Time synchronization is achieved using least squares fitting;

[0090] Spatial interpolation uses Kriging interpolation.

[0091] Outlier identification criteria: |x-μ|>3σ, where μ is the mean and σ is the standard deviation.

[0092] Furthermore, the route dynamic planning method includes:

[0093] Cost function design: g(n)=α×D(n)+β×W(n) where D(n) is the distance cost and W(n) is the meteorological cost;

[0094] Heuristic function design: h(n) = γ × H(n) where H(n) is the straight-line distance from the current point to the target point.

Claims

1. A system for applying meteorological information in flight scheduling, characterized in that, include: The system includes a meteorological data acquisition module, a data preprocessing unit, a feature extractor, an impact assessment module, a flight grouping unit, a multi-objective optimizer, a weight adjustment unit, a scheduling strategy generator, and an execution control module. The meteorological data acquisition module provides standardized meteorological data to the feature extractor; the meteorological feature vectors obtained by the feature extractor are used as input to the impact assessment module to obtain the flight diversion probability; the flight diversion probability is input to the flight grouping unit for dynamic grouping; the grouping results are combined with the multi-objective optimizer to generate an initial scheduling plan; the weight adjustment unit dynamically adjusts the optimization objective based on execution feedback; The scheduling strategy generator optimizes the scheduling scheme based on the adjusted weight coefficients; the execution control module is responsible for implementing the scheme and feeding back the execution results to the weight adjustment unit. The meteorological data acquisition module is used to collect multi-source meteorological data. The module also includes a data preprocessing unit, which uses a cubic spline interpolation algorithm to unify the spatiotemporal resolution of the multi-source meteorological data; filters noisy data using wavelet transform to achieve data quality control; identifies and corrects outliers using a Gaussian distribution-based anomaly detection method; and provides standardized meteorological data. The feature extractor uses an improved ResNet deep network structure to extract local features from meteorological data; it achieves unified processing of features at different scales through bilinear interpolation; it calculates the weight relationship between different features using a self-attention mechanism; and it captures the temporal evolution features of meteorological elements based on an LSTM network. Obtain meteorological feature vectors, The impact assessment module takes meteorological feature vectors as input and constructs a flight operation safety margin assessment model; designs a flight delay risk assessment method based on fuzzy comprehensive evaluation; establishes a delay propagation prediction model that considers flight connectivity; and uses grey prediction theory to assess the probability of flight diversion. Flight grouping unit: Calculate flight grouping initialization scheme based on the similarity of waypoint sequences; utilize K-means source competition relationships; A multi-objective optimizer is used to establish an optimization function with minimizing the total delay time as the primary objective; cost constraints that take fuel consumption into account are constructed. Design a penalty factor based on flight punctuality rate; The NSGA-III algorithm is used to solve the multi-objective optimization problem. The weight adjustment unit is used to train the initial weights of the optimization targets based on historical data, and dynamically adjust the weight coefficients of each optimization target according to real-time scheduling feedback through a reinforcement learning method. The reinforcement learning method adopts a deep Q-network, uses the scheduling execution effect as a reward signal, and optimizes the weight adjustment strategy through an experience replay mechanism. The historical data includes scheduling schemes and their execution effects under different meteorological conditions. The scheduling strategy generator generates flight takeoff and landing sequences based on dynamic programming algorithms; implements route dynamic planning using the A* algorithm; optimizes ground support resource allocation through the Hungarian algorithm; and employs heuristic algorithms to handle strategy adjustments in unforeseen circumstances.

2. The system according to claim 1, characterized in that, The specific structure of the ResNet deep network used in the feature extractor includes: The input layer uses a 5×5 convolutional kernel to extract local features; The intermediate layer contains 3 residual blocks, and each residual block consists of two 3×3 convolutional layers; The feature fusion layer uses 1×1 convolution to achieve channel dimensionality reduction; The output layer generates feature vectors through fully connected layers.

3. The system according to claim 1, characterized in that, The flight operation safety margin assessment model includes: The delay risk level is divided into five levels, and the risk is quantified using fuzzy membership functions. The probability prediction of flight diversion is based on the grey prediction model GM(1,1).

4. The system according to claim 1, characterized in that, The flight grouping unit includes: The execution control module establishes an event-driven scheduling instruction issuance mechanism; and uses digital twin technology to achieve real-time monitoring of scheduling execution.

5. The system according to claim 1, characterized in that, The NSGA-III algorithm consists of: an execution control module that uses a sliding time window method for dynamic adjustment; and a fuzzy control algorithm for feedback compensation.

6. The system according to claim 1, characterized in that, The structure of the deep reinforcement learning network includes: State space S: contains 9 key features of the current scheduling state; Action space A: contains 5 optional strategies for weight adjustment.

7. The system according to claim 1, characterized in that, The scheduling policy generation method includes: The takeoff and landing sequence is generated using an improved dynamic programming algorithm.

8. The system according to claim 1, characterized in that, The execution control method includes: Set the sliding time window length to 120 minutes; Define the deviation threshold δ for triggering adjustments: trigger local adjustments when the actual delay exceeds 20% of the expected delay; trigger global re-optimization when weather conditions change abruptly.

9. The system according to claim 1, characterized in that, The data preprocessing method includes: Time synchronization is achieved using least squares fitting; Spatial interpolation uses Kriging interpolation.

Images