Identifying transient and persistent errors in aircraft cruise trajectory prediction using Bayesian state estimation

Abstract

Recorded aircraft trajectory data may vary significantly from predictions based on physics-informed models. This discrepancy may be attributed to inadequacies in the trajectory prediction models, including errors in modeling aircraft dynamics, and omission of inputs such as weather data, equipment malfunctions, and pilot errors. In this work, we represent model errors as additional external inputs to the trajectory prediction model, evaluate them using Bayesian state estimation, and update the trajectory prediction model to address these errors. In doing so, we distinguish between two types of model errors – transient and persistent errors – of which only persistent errors are considered for correction. In this study, we estimate the model error for the completed portion of ongoing flights, separate out the persistent error and account for it in the system model, which allows for an improved prediction of aircraft trajectory for the remaining flight duration. We perform verification and validation of the proposed methodology using both synthetically generated data as well as historically recorded data corresponding to flights affected by adverse weather conditions and engine failure.