In a recent article published in Aerospace, researchers explored using machine learning (ML) techniques to estimate hidden parameters related to flight and aircraft efficiency, specifically focusing on predicting key performance indicators (KPIs) such as fuel consumption, flight distance, and gate-to-gate time.
Study: Machine-Learning Methods Estimating Flights’ Hidden Parameters for the Prediction of KPIs. Image Credit: Gorodenkoff/Shutterstock.com
They introduced a novel methodology that integrates mechanistic models with advanced artificial intelligence (AI) and ML techniques to enhance air traffic management (ATM) performance and minimize the environmental impact associated with air travel.
Air Traffic Management and Performance Modeling
ATM aims to ensure the safe and efficient flow of air traffic but often faces environmental challenges, including increased carbon dioxide (CO2) emissions due to operational constraints.
To address these challenges, modernization initiatives such as the Single European Sky ATM Research (SESAR) in Europe and the Next Generation Air Transportation System (NextGen) in the United States have introduced innovative solutions aimed at reducing inefficiencies. Evaluating the performance of these initiatives is critical for identifying gaps between current outcomes and high-level targets, facilitating necessary improvements.
Organizations like the International Civil Aviation Organization (ICAO) have established performance-based frameworks and KPIs to monitor and manage ATM system performance effectively.
Within the ATM research community, considerable effort has been devoted to developing modeling methods that analyze the interdependencies between key performance areas (KPAs) and their influence on KPIs. These modeling approaches are generally divided into two categories: macroscopic models, which evaluate overall system behavior, and microscopic models, such as agent-based models, which offer detailed insights into individual actions and interactions.
Microscopic models, especially agent-based ones, have demonstrated significant potential for capturing the complex behaviors of ATM systems. However, their practical application is often limited by the challenge of estimating hidden parameters that are not directly observable or measurable. Advancing robust performance modeling methodologies remains a critical goal for enhancing ATM efficiency and promoting environmental sustainability.
About the Research
This paper presents a data-driven methodology for estimating hidden flight parameters, specifically payload mass (PL) and cost index (CI), by integrating mechanistic models with AI and ML techniques. The proposed approach leverages historical flight data and trajectories generated by the DYNAMO optimization engine, which simulates realistic flight conditions and serves as a foundation for effectively training ML algorithms.
The researchers utilized two primary ML methods: graph convolutional networks (GCN) and gradient boosting machines (GBM). The GCN method employed an ensemble of agents, each representing specific flight phases, to collaboratively predict hidden parameters based on local observations. In contrast, GBM followed an iterative decision tree process, refining predictions by correcting errors from previous estimations.
The dataset comprised simulated flight data from DYNAMO, historical weather records, and operational flight plans provided by EUROCONTROL. Training and testing datasets were constructed from this data, with feature extraction focusing on trajectory variables. Various AI and ML algorithms were then applied to estimate the hidden parameters. The study assessed the accuracy of these estimations in predicting KPIs, ultimately contributing to improved efficiency in ATM systems.
Key Findings and Insights
The outcomes demonstrated that GCN and GBM methods accurately estimated hidden flight parameters. Specifically, the GBM method achieved a mean absolute error (MAE) of under 4 % for CI and about 2 % for PL, showcasing its robustness and stability. Meanwhile, GCN delivered comparable or even better results, particularly in estimating PL under real-world conditions, highlighting its potential for practical applications.
The study also emphasized the critical role of accurate hidden parameter estimation in predicting fuel consumption. It revealed that significant discrepancies in PL lead to considerable impacts on fuel consumption predictions, underscoring the need for precise parameter estimation in order to optimize flight efficiency.
Furthermore, the proposed AI/ML methods effectively supported trajectory-related KPI predictions, achieving a mean absolute percentage error (MAPE) below 1 % for fuel consumption and 0 % for flown distance and gate-to-gate time. This indicates that even minor inaccuracies in hidden parameter estimations can significantly affect KPI prediction.
Practical Implications
The proposed methodology for estimating hidden flight parameters and its impact on KPI prediction has significant applications in ATMs. It can support performance assessments and decision-making by accurately evaluating the impact of ATM concepts on key performance metrics, even when certain operational parameters are not directly observable.
Additionally, predicting flight KPIs using these estimated parameters is valuable for trajectory optimization, air traffic flow management, and environmental impact assessments. The methodology’s robustness to the sensitivity of aircraft performance models further enhances its utility, allowing for the aggregation of KPIs to evaluate ATM performance effectively, even without detailed aircraft data.
Conclusion and Future Directions
In conclusion, the proposed data-driven methodology successfully estimated hidden flight parameters and evaluated their impact on KPI predictions. By integrating mechanistic models with advanced AI and ML techniques, the approach demonstrated substantial potential for improving performance assessment and decision-making within ATM systems.
Future research should focus on exploring a broader range of prediction and optimization models, refining GCN configurations, and extending the methodology to estimate other hidden KPIs. Additionally, integrating this framework with ATM applications such as trajectory optimization and air traffic flow management could further enhance its applicability and impact. As the aviation industry continues to emphasize sustainability and efficiency, this methodology offers a promising pathway for advancing the capabilities of ATM systems.
Journal Reference
Vouros, G.;& et al. Machine-Learning Methods Estimating Flights’ Hidden Parameters for the Prediction of KPIs. Aerospace 2024, 11, 937. DOI: 10.3390/aerospace11110937, https://www.mdpi.com/2226-4310/11/11/937
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.