Wheels-Off Time Prediction Using Surface Traffic Metrics

Abstract

This paper is motivated by the need for wheels-off time prediction required for improving departure scheduling. It will be possible to estimate wheels-off time with high precision at major U. S. airports where airport surface automation, such as the Surface Management System, is deployed. At other airports, controllers are expected to estimate wheels-off time for departure scheduling. This paper analyzes Dallas-Fort Worth airport state data metrics derived from the Aviation System Performance Metrics database and Surface Management System logs for use as inputs to a neural network for predicting wheels-off time at airports where the Surface Management System will not be available. At airports without Surface Management System, these metrics will be either directly available to controllers or can be computed using combinations of flight-plan data, taxi clearance data, surface surveillance data, airline provided Out-Off-On-In (OOOI) data and airport surface geometry data. Analysis of metrics derived from Aviation System Performance Metrics database and Surface Management System logs show that there is a high degree of correlation between these metrics and gate to wheels-off time. Correlations of these metrics with gate departure delay were found to be quite low, so no attempt was made to predict gate departure delay using these metrics. This study assumed the gate departure time to be known for estimating wheels-off time. Gate to wheels-off time is predicted using the neural network with chosen metrics as inputs. The neural network is trained with six days of data and tested on one day of data. Results show that the trained neural network performance on the test data is as good as on the training data. The main finding of the study is that a neural network trained on several days of airport state data was able to generate gate to wheels-off time predictions within the Call for Release departure compliance window of two-minutes early to one-minute late 59% of the time. This performance can be improved further by removing outliers in the training and test sets. Next, a linear model was created with the same set of metrics as independent variables and gate to wheels-off time as the dependent variable. Six days of data were used with the least-squares method to compute the coefficients. These coefficients were then used with one day of test data to estimate the gate to wheels-off time. Results were found to be comparable to that generated by the neural network based on the Mahalanobis distance metric. Comparison of predictions generated using the neural network and the linear model to the test data provide evidence that the chosen metrics are suitable for predicting gate to wheels-off time.

Wheels-Off Time Estimation at Non-ASDE-X Equipped Airports

“…The study in Ref. 3 found the correlation between gate departure delay and metrics derived from airport state data such as, number of aircraft on the surface, airport departure rate, wind and visibility to be quite low. Like Ref.…”

“…The earlier study in Ref. 3 examined wheels-off time estimation using a neural network with metrics derived from historical surface surveillance data and ASPM database. That study analyzed data from Dallas-Fort Worth and showed that gate to runway distance is the most significant factor for wheels-off prediction.…”

A Data Driven Approach for Characterization of Ramp Area Push Back and Ramp-Taxi Processes

“…Accuracy of the wheels-off time estimate affects the quality of traffic demand estimates. Inaccurate traffic demand estimates can result in either inadequate or unneeded flow restrictions [1]. A large source of error in predicting the demand results from the prediction errors of the taxi times [2], which also plays a role in the prediction of the wheels-off time.…”

Taxi-Out Time Prediction for Departures at Charlotte Airport Using Machine Learning Techniques

“…11 In addition, reinforcement learning algorithms were developed and investigated to check the accuracy of taxi-out time prediction at several major airports in the United States. [12][13][14] Various regression methods, including multiple linear regression, least median squared linear regression, support vector regression, model trees, and fuzzy rule-based systems, were also applied to several airports in Europe for taxi time prediction problems.…”