In this research, we tackle the problem of effectively modeling the behavior of trajectories in a complex and structured terminal airspace with RNAV. In previous research, trajectories are clustered to produce groups of trajectories (or patterns) with similar behavior that are distinct from other patterns and study each cluster separately. However, well-known clustering algorithms (DBSCAN, K-means, etc) are not suitable for RNAV terminal airspace, as it produces clusters that have wide inter-cluster variance and is not informative to data-driven algorithms. To tackle this problem, we developed a trajectory pattern identification framework using hierarchical agglomerative clustering and Dynamic Time Warping (DTW) that can produce patterns with small variance. With the identified cluster as labels, we trained a Recurrent Neural Network-based model that can classify trajectory patterns online and compute the probabilities of the partial trajectory belonging to the identified patterns. By combining the trajectory pattern identification and real-time classification frameworks, historical air surveillance data recorded in the RNAV terminal airspace can be better studied and therefore data-driven air traffic assistant tools can be better developed and implemented. 

To incorporate the historical behaviors of aircraft with the current flight mode or dynamics, we developed a framework for more accurate trajectory prediction in terminal airspace by combining machine learning and estimation techniques. This framework consists of three steps:

• Data preparation is first performed by using data cleaning and trajectory pattern identification.

• A machine-learning-based trajectory prediction model is trained with historical data to generate the probability distribution of an incoming flight’s future states.

• The aircraft’s dynamics is modeled as a stochastic linear hybrid system, i.e., Residual-Mean Interacting Multiple Model (RM-IMM), to estimate the future states of the aircraft using the data-driven prediction as the pseudo-measurement.

The proposed framework improves the prediction accuracy by using RM-IMM to correct the data-driven prediction to follow the aircraft’s dynamics.

Accurate Estimated Time of Arrival (ETA) predictions have become increasingly important in modern air traffic management systems, which rely on scheduling and metering aircraft based on their remaining flight time in order to ensure safety and efficiency. We proposed a multi-agent ETA prediction method that accounts for the current air traffic situation and captures the decision made by air traffic controllers. This approach relies on an agent-aware attention mechanism, which enables a more comprehensive understanding of traffic situations, resulting in improved prediction accuracy, compared to existing ETA prediction methods have focused on a single aircraft which is assumed to be independent. The experimental results demonstrated with real air traffic surveillance data show the accuracy of the proposed method is superior compared to existing algorithms. In addition, the test results of the aircraft landing sequence present the potential applicability to the aircraft sequencing problem.

Publication:

H. Choi, H. Park, C. Deng, and I. Hwang, “Multi-agent Aircraft Estimated Time of Arrival Prediction in Terminal Airspace,” In the Proceedings of the 42nd AIAA/IEEE Digital Avionics Systems Conference (DASC), Barcelona, Spain, October 1-5, 2023

H. Choi, C. Deng, H. Park, J. Ryu, H. Lee, and I. Hwang, “Multi-agent Estimated Time of Arrival Prediction and Dynamic Arrival Sequencing by Emulating Air Traffic Controllers,” Journal of Air Transport Management (under review)

Identifying anomalous flight trajectories is critical in airspace operations due to their potential safety risks. One of the challenges in distinguishing abnormal aircraft paths within a vectored RNAV terminal airspace is the difficulty in differentiating them from structured procedures. Existing trajectory pattern identification algorithms applied to vectored paths often yield patterns with significant variability within a pattern, which hinders effective anomaly detection. Moreover, conventional anomaly detection methods designed for historical data are not readily adaptable for real-time implementation. To address these challenges, this study proposes an online anomaly detection algorithm for vectored flights in RNAV terminal airspace based on the Gaussian mixture model (GMM). This approach accommodates the complexity of airspace operations by integrating GMM with dynamic trajectory pattern classification and hybrid trajectory prediction, which can detect dynamic changes in the trajectory pattern online and switch the anomaly detection model accordingly.

Publication:

H. Choi, C. Deng, H. Park, and I. Hwang, “Gaussian Mixture Model-based Online Anomaly Detection for Vectored Area Navigation Arrivals,” AIAA Journal of Aerospace Information Systems, Vol. 20(1), pp. 37-52, January 2023, DOI: https://doi.org/10.2514/1.I011128 

Safety is of the utmost importance in the air traffic system. In recent years, data-driven algorithms have emerged to identify anomalous and potentially unsafe operations based on machine learning techniques. Although many algorithms have shown notable progress in anomaly detection, they overlook the fact that data can be corrupted by noise and uncertainty (e.g., navigation system error), leading to frequent misdetection and false alarms which could disturb air traffic controllers and result in system performance degradation. Therefore, ensuring accurate and reliable identification of emerging safety risks requires addressing and mitigating the uncertainty present in the data. To achieve this, we proposed a framework based on conformal prediction that explicitly considers the uncertainty in data to enhance the reliability of anomaly detection and continuously learns from new streaming data. Beyond supporting air traffic controllers in monitoring tasks, the proposed method can also offer strategies to resolve anomalies when the probability exceeds a predefined threshold during control tasks.

Publication:

H. Choi and I. Hwang, “Toward Real-Time Stochastic Conformal Anomaly Detection in Terminal Airspace,” In the Proceedings of the International Conference on Research in Air Transportation (ICRAT) 2022, Tampa, FL, June 19-23, 2022

H. Choi, C. Deng, H. Park, and I. Hwang, “Stochastic Conformal Anomaly Detection and Resolution for Air Traffic Control,” Transportation Research Part C: Emerging Technologies, Vol.15, September 2023, doi: 10.1016/j.trc.2023.104259 

Research in trajectory anomaly detection in the air traffic domain has mainly considered each trajectory to be independent. However, an aircraft’s trajectory is dependent on the trajectories of nearby aircraft, especially in a terminal airspace, as air traffic controllers actively alter the aircraft’s trajectory based on traffic density, separation, and scheduling to maintain safety and efficiency. To capture the interaction between flights and find anomalies in relation to the situation, we proposed a situational anomaly detection framework based on multi-agent trajectory distribution prediction with an agent-aware attention mechanism. The situational anomalies are defined based on the predicted distribution of the trajectory. 

Publication:

H. Park and I. Hwang, “Situational Anomaly Detection Using Multi-agent Trajectory Prediction for Terminal Airspace Operations,” AIAA SciTech 2023: Intelligent Systems, National Harbor, MD, January 23-27, 2023. doi: 10.2514/6.2023-2538

Modern navigation systems such as Area Navigation (RNAV) yield new challenges for developing data-driven algorithms and new perspectives in defining the safety and complexity of the terminal airspace due to the complicated maneuvers of aircraft. In this paper, we propose a complexity estimation framework for RNAV terminal airspace. The framework integrates our previously developed algorithms for trajectory pattern identification, multi-agent trajectory prediction, and Gaussian mixture model-based anomaly detection. All algorithms are developed to be implemented in the complex situation of RNAV terminal airspace. The estimated complexity prompts researchers and air traffic controllers to investigate situations where the complexity is abnormally high for potential risks or operational errors. The proposed complexity estimation framework is tested with real air traffic surveillance data recorded in Incheon International Airport, South Korea.

A case study that is highlighted by the proposed framework is shown on the left. That traffic scene is between 6:23:00 and 6:33:00 on January 2, 2019 in the Incheon International Airport. This case is highlighted by the proposed framework because the horizontal separation becomes too small (flights in the red circles) at 06:25:00 and 06:29:00, respectively. It can be seen that a large number of flights enter from GUKDO (an entry fix at the southeast) in a short period time and air traffic controllers (ATCs) try to delay some of them so that a proper sequencing is maintained. However, it is not sure why the ATCs do not choose to vector some flight to KARBU (an entry fix at the northeast) or put some flights in the holding position. Possible reasons could be communication issues, pilot’s nonconformance, or weather. In addition, this case shows that too many flights from the same entry fix could result in safety risks, horizontal separation being violated in this case, and ATCs responsible for sectors outside of ICN should take note of this and avoid such situations.

Publication:

C. Deng, H. Choi, H. Park, and I. Hwang, “Area Navigation Terminal Airspace Complexity Estimation for Arrivals,” in 15th Air Traffic Management Research and Development Seminar, 2023.

The continuous growth of demand on commercial airlines has made it crucial to guarantee the safety of airspace operations. Although adverse events are rare, once they happen, they can cause unpredictable risky factors and degrade airspace efficiency. Thus, studying historical air traffic data to discover precursors, features, or events that contribute to the occurrence of the adverse event in the future is important and has gained interest in recent years. In this paper, a novel and real-time applicable temporal precursor discovery (TPD) framework based on the long short-term memory neural network and the feature attention mechanism is proposed. The feature attention mechanism enables the framework to pay attention to certain features at a certain time, and the attention score is defined as the temporal precursor. The temporal precursor reflects the rationale behind the neural network’s prediction at each time step, providing a data-driven explanation of how the adverse event occurs. The proposed TPD framework was tested with real air traffic data and weather data recorded at Incheon International Airport in South Korea in 2019. 

Publication:

C. Deng, H. Choi, H. Park, and I. Hwang, “Temporal Precursor Discovery Using LSTM with Feature Attention,” AIAA Journal of Aerospace Information Systems, October 18, 2023, doi: 10.2514/1.I011225