A Novel Altitude Measurement Channel Reconstruction Method Based on Symbolic Regression and Information Fusion

Abstract

Accurate altitude data are imperative for precise aircraft flight control, navigation planning, and air traffic management, especially in global positioning system (GPS)-denied environments. While deep learning methods offer promising solutions for altitude prediction through complex predictive models, their inherent lack of interpretability raises safety concerns, particularly in safety-critical aviation contexts. This article introduces a novel symbolic regression (SR)-based approach to altitude prediction. Initially, raw data undergo random projection (RP) to a feature space, addressing challenges associated with feature extraction in SR. Subsequently, altitude-related information is discerned from the inertial navigation system (INS) and atmospheric system (AS), employing genetic programming (GP) to formulate fully interpretable altitude prediction equations. To enhance robustness, information fusion (IF) technology integrates the prediction equations with vertical velocity, establishing a resilient virtual altitude channel. In scenarios where the GPS is entirely unavailable, our proposed method undergoes effective validation across diverse aircraft types and under various flight conditions. Furthermore, the robustness of our fusion algorithm is verified across different noise levels, underscoring its reliability in challenging conditions.