Evaluating Effectiveness of Fixed-Flight Path Angle Descent to Kansai International Airport Using A320 Flight Data via Machine Learning Approaches

The full flight simulator experiments in this paper show that Fixed-flight Path Angle (FPA) descent consumes less fuel than flight with current control, and that fuel consumption can be reduced by properly selecting that angle of descent. It was also suggested that the optimal angle of descent varies with weather conditions. Therefore, this study applied machine learning to develop a highly accurate fuel consumption estimation model to determine the optimal angle of descent for FPA descent. As a result, a model with higher accuracy than existing models based on physical laws, such as BADA and ICAO, was successfully constructed. However, the accuracy of fuel flow estimation during FPA descent was lower than other flight phases including climb, cruise, and non-FPA descent. This result shows that applicability of the machine learning model is limited by the actual flight envelope of the underlying training data. These results suggest our future works to add extra loss terms including physical knowledge to evaluation function of the machine-learning model. We discuss on the selecting features and evaluation functions of the model to further improve the accuracy.