Research On Estimated Time of Arrival Prediction Based Upon ADS-B and Spatiotemporal Analysis

Estimated Time of Arrival (ETA) is of great importance for airport department of operation and ground service. Precise ETA prediction is the core element of Airport Collaborative Decision Making (A-CDM) system, which is intended to improve the real-time airport gate assignment, reduce runway intrusion risk and cut down waiting time for ground maintenance vehicles. In this paper, we introduce a novel hybrid model to address ETA prediction problem by application of supervised learning methods based on multisource data. The proposed model consists of two parts: spatiotemporal trajectories storage and clustering-based prediction. Firstly, in the preprocessing part, after data cleaning, filtering, re-sampling and data storage, we increase the dimension of trajectory variable. Then, the discrete track points are clustered into several patterns by clustering algorithm. Using real-time data cross validation, the model is trained to find out the different prediction factor of ETA for each individual cluster cell under various conditions. Finally, the predicted ETA for each flight is generated. Lots of validation work has been approached with this model. Experimental results demonstrate that our proposed method based on spatiotemporal data analysis is the most effective and accurate. The system that implements this model has been developed and deployed into the production runtime environment in Xian Airport, Harbin Airport, Nanjing Airport, Shenzhen Airport, etc.