An Availability Prediction Method of Ground-Based Augmentation System Based on Support Vector Machine Algorithm
Zhipeng Wang, Jingtian Du, Beihang University, China; Wei Zhi, CETHIK Group Corporation Research Institute, China; Yanbo Zhu, Yuan Liu, Beihang University, China; Qian Sun, China Waterborne Transport Research Institute, China
Location: Grand Ballroom F
Date/Time: Wednesday, Jan. 31, 2:35 p.m.
The Ground-Based Augmentation System (GBAS) is a navigation system for aircraft designed to be used for precision approaches and landings. GBAS improves the accuracy of the satellite navigation system significantly. However, some flight accidents still occur occasionally due to lack of availability and it makes more stringent requirements on availability that CAT II/III precise approach is trying to introduce GBAS. Therefore, it is necessary to find an effective prediction algorithm for GBAS. The protection level is a key parameter in the process of evaluating the availability. There are three relevant parameters: quantile K represents the fault-free missed detection multiplier; projection matrix S is the vertical projection factor in the along-track coordinate system; variance of measurement noise is the standard deviation of the uncertainties of the residual differential error, which consists of the root sum square of uncertainties introduced through ionospheric and tropospheric decorrelation as well as the contribution of the ground and airborne multipath and noise. Some researchers have found that the model for the standard deviation of GBAS pseudo-range correction error is conservative, which causes decline of prediction accuracy. However, it is hard to explicitly model this parameter. We made a reevaluation based on support vector machine and improved the prediction algorithm for availability. We developed the simulation software and make practical measurement. Considering the influence of different distributions of based stations of GBAS, the availabilities of EWR, GIG and IAH are evaluated, and the results were compared with the actual values. The result shows that the prediction errors are 3.17%, 1.96% and 0.98% respectively.