基于高斯过程的飞行转弯区一致性监视技术研究

In order to solve the problem of conformance monitoring during turns in which the disturbances, inherent uncertainties and variabilities, such as navigation errors,monitoring system errors, and differences of turning point and turning radius, make it difficult to determine whether an aircraft conforms to the assigned flight plan or the issued command, an approach based on the comparison between the expected state values and the measured state values was proposed. The interval of expected state values was calculated through modeling the conforming flights using Gaussian Processes. The deviation term was introduced to enable the model to describe the reasonable change of state values at different time points during turns. The calculation was completed under the framework of Bayesian inference in Gaussian Processes, and stochastic sampling was used to compute the posterior distribution which is analytically intractable after introducing deviation term. The proposed model for conforming flights was trained by the historical flight data. It can avoid common problems when the expected state values were inferred directly from flight plan or control command, mainly including the failure to account for the influences caused by various factors in practical situations. The experiments using flight data from realistic surveillance system showed that the calculated interval of expected state values in the proposed model reflects the reasonable range of state values and therefore is capable of monitoring conforming flights. Furthermore, compared with the conformance monitoring based on threshold value of trajectory deviation, this proposed approach achieved better false alarm rate and detection time and avoid the performance issues caused by improper settings of threshold value. The results also showed that this proposed approach has extensibility and can improve the performance of conformance monitoring by using more state values in addition to position information.