This Master’s thesis presents an Artificial Intelligence (AI) approach for the actual Take-Off Weight (TOW) estimation of commercial aircraft, a critical parameter for accurate trajectory prediction, fuel consumption optimization, and environmental sustainability in Air Traffic Management (ATM). Given the proprietary nature and limited availability of actual TOW data, this research develops predictive models by integrating diverse data sources: operational flight data from CRIDA, public aircraft performance specifications, and benchmark TOW estimations from the EUROCONTROL Performance Review Commission (PRC) Data Challenge. A comprehensive methodology was employed, involving extensive data preprocessing, feature engineering (incorporating temporal, spatial, and dynamic flight characteristics, with a focus on climb phase data), and an in-depth Exploratory Data Analysis (EDA). The EDA revealed significant data imbalances, particularly a predominance of Medium (M) wake category aircraft. To address this, the study was structured into three distinct case studies: (1) M-wake flights with climb phase trajectory data, (2) M-wake flights without climb trajectory data, and (3) Heavy (H) wake category flights. State-of-the-art Gradient Boosting Decision Tree (GBDT) algorithms (XGBoost, LightGBM, CatBoost) were utilized, alongside the H2O AutoML framework, with hyperparameter optimization facilitated by Optuna and experiment tracking managed by MLFlow.
The models demonstrated high predictive accuracy, particularly for M-wake flights with climb phase data, achieving an R2 of approximately 0.967 and a Root Mean Squared Error (RMSE) of 1,333. This highlights the strong predictive power of early f light dynamics. For M-wake flights without climb trajectory data, performance remained robust (RMSE around 1,755, R2 > 0.95), demonstrating adaptability to varying data granularity. H-wake category flights presented greater challenges due to data scarcity and higher variability, resulting in larger absolute errors (RMSE around 5,233 ) despite a respectable R2 of 0.956. Feature importance analysis consistently identified aircraft-specific characteristics (MTOW, MLW, OEW) as primary predictors across all scenarios.
In conclusion, this research successfully demonstrates the potential of AI to provide robust, data-driven TOW estimations, directly supporting enhanced ATM efficiency and environmental objectives. While further research is needed to address data limitations for heavy aircraft and explore advanced hybrid modelling approaches, these models offer a valuable solution for operational planning and strategic assessments in aviation.
This Master’s thesis presents an Artificial Intelligence (AI) approach for the actual Take-Off Weight (TOW) estimation of commercial aircraft, a critical parameter for accurate trajectory prediction, fuel consumption optimization, and environmental sustainability in Air Traffic Management (ATM). Given the proprietary nature and limited availability of actual TOW data, this research develops predictive models by integrating diverse data sources: operational flight data from CRIDA, public aircraft performance specifications, and benchmark TOW estimations from the EUROCONTROL Performance Review Commission (PRC) Data Challenge. A comprehensive methodology was employed, involving extensive data preprocessing, feature engineering (incorporating temporal, spatial, and dynamic flight characteristics, with a focus on climb phase data), and an in-depth Exploratory Data Analysis (EDA). The EDA revealed significant data imbalances, particularly a predominance of Medium (M) wake category aircraft. To address this, the study was structured into three distinct case studies: (1) M-wake flights with climb phase trajectory data, (2) M-wake flights without climb trajectory data, and (3) Heavy (H) wake category flights. State-of-the-art Gradient Boosting Decision Tree (GBDT) algorithms (XGBoost, LightGBM, CatBoost) were utilized, alongside the H2O AutoML framework, with hyperparameter optimization facilitated by Optuna and experiment tracking managed by MLFlow.
The models demonstrated high predictive accuracy, particularly for M-wake flights with climb phase data, achieving an R2 of approximately 0.967 and a Root Mean Squared Error (RMSE) of 1,333. This highlights the strong predictive power of early f light dynamics. For M-wake flights without climb trajectory data, performance remained robust (RMSE around 1,755, R2 > 0.95), demonstrating adaptability to varying data granularity. H-wake category flights presented greater challenges due to data scarcity and higher variability, resulting in larger absolute errors (RMSE around 5,233 ) despite a respectable R2 of 0.956. Feature importance analysis consistently identified aircraft-specific characteristics (MTOW, MLW, OEW) as primary predictors across all scenarios.
In conclusion, this research successfully demonstrates the potential of AI to provide robust, data-driven TOW estimations, directly supporting enhanced ATM efficiency and environmental objectives. While further research is needed to address data limitations for heavy aircraft and explore advanced hybrid modelling approaches, these models offer a valuable solution for operational planning and strategic assessments in aviation. Read More
Related posts:
A machine learning approach to air traffic interdependency modelling and its application to trajectory prediction
Design of an ATC Tool for Conflict Detection Based on Machine Learning Techniques
Conflict-Free Trajectory Planning Based on a Data-Driven Conflict-Resolution Model
Comparative assessment of glucose prediction models for patients with type 1 diabetes mellitus applying sensors for glucose and physical activity monitoring
Formal intent-based trajectory description languages for quadrotor aircraft
Comparative analysis of aspen plus simulation strategies for woody biomass air gasification processes