Unstable Approach Detection and Analysis Based on Energy Management and a Deep Neural Network

The study of managing risk in aviation is the key to improving flight safety. Compared other operation phases, approach and landing phases are more critical dangerous. This aims detect analyze unstable approaches Taiwan through historical data. In addition weather factors such as low visibility crosswinds, human also account for a large part risk. From accidents studied stochastic report Flight Safety Foundation, nearly 70% occurred during which were caused by improper control aircraft energy. Since information data recorder (FDR) regarded airline’s confidential information, this calculates aircraft’s energy-related metrics investigates influence non-weather-related on publicly available source, automatic dependent surveillance-broadcast (ADS-B) To evaluate weather- factors, outliers each group classified labels detected eliminated from analysis applying hierarchical density-based spatial clustering applications with noise (HDBSCAN), utilized detecting abnormal flights that anomalies. deep learning method was adopted predict arrival at Taipei Songshan Airport. accuracy prediction normalized total energy trajectory deviation all 85.15% 82.11%, respectively. results show different kinds conditions, or not considering weather, models have similar good performance. input features analyzed after model obtained, discussed.