Runway Configuration Prediction Using 3D Tensor CNNs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current air traffic management systems lack the capability to accurately and efficiently predict runway configurations and airport acceptance rates (AARs) in real-time, especially for complex multi-airport systems, due to reliance on fragmented weather forecast data and lack of tools to translate this data into actionable configurations and AARs.
Innovation Solution
A data-driven Deep Learning framework that utilizes ensemble gridded weather forecasts to predict both runway configurations and AARs simultaneously, capturing operational interdependencies within the parameter learning process, and employing Convolutional Neural Networks (CNNs) to process three-dimensional tensor data from numerical weather prediction sources like Rapid Refresh (RAP) data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If current operational protocol with coordination of different entities is used to determine runway configurations and AARs, then human decision-making can handle complex interdependencies, but intensive verbal communications and lack of automated tools reduce efficiency and productivity
Solution Approach 1:
The patent replaces the manual mechanical coordination process with an automated Deep Learning system. The CNN framework automatically processes weather forecast data and generates runway configuration and AAR predictions, substituting human verbal communications and manual coordination with an automated computational system that handles the complex decision-making process.
Solution Approach 2:
The system enables self-service by allowing the Deep Learning model to autonomously predict runway configurations and AARs without requiring intensive human coordination. The automated framework independently processes weather data and generates operational recommendations, reducing the need for continuous human intervention and verbal communications between entities.
2Device complexity
If isolated station-based terminal weather forecast is used, then data processing is simpler, but spatial features of weather forecasts are ignored reducing prediction accuracy
Solution Approach 1:
The patent transitions from isolated station-based (0D/1D) weather forecasts to gridded ensemble weather forecasts with spatial dimensions (2D/3D). The CNN architecture processes this multi-dimensional data, capturing spatial variations in weather conditions across the terminal area, thereby improving prediction accuracy while managing the increased data complexity through automated feature extraction.
Solution Approach 2:
The system changes the parameters of weather forecast data from single-station measurements to gridded ensemble predictions with multiple spatial and temporal parameters. The Deep Learning model automatically processes these enhanced parameters, extracting relevant features that improve the accuracy of runway configuration and AAR predictions without requiring manual simplification.
3Measurement precision
If operational interdependency among airports in MAS is fully captured, then prediction accuracy improves, but the problem becomes mathematically difficult to formulate and complex to solve
Solution Approach 1:
The patent replaces complex mathematical formulation approaches with a Deep Learning-based computational system. The CNN framework automatically learns the complex interdependencies among airports in the multi-airport system through data-driven training, avoiding the need for explicit mathematical modeling while capturing the operational relationships and improving prediction accuracy.
Solution Approach 2:
The system transforms the complex modeling problem into a parameter optimization problem for the Deep Learning model. By changing from explicit mathematical formulation to implicit learning through gradient descent and backpropagation, the system handles the complexity of interdependent airport operations through automated parameter adjustment during training rather than manual model formulation.
Data Source
AI summary
Numerical weather prediction data during a time period of interest for a multi-airport system is used to generate a three-dimensional (3-D) tensor data for the time period of interest. The 3-D tensor data is then used to predict hourly runway configurations for each of a plurality of airports of the multi-airport system during the time period of interest using the 3-D tensor data as input to a plurality of runway configuration convolutional neural networks (CNN) branches and to predict an AAR for each of the plurality of airports of the multi-airport system during the time period of interest using the 3-D tensor data as input to an airport acceptance rate (AAR) CNN branch. The predicted hourly runway configuration for each of the plurality of airports and the predicted AAR for each of the plurality of airports are then used to manage flights within the multi-airport system.