Machine Learning Optimization of Model Following Control for Resilient Microburst Attenuation on Final Approach

Several commercial aviation incidents have been due to microbursts, resulting in the loss of hundreds of lives. Microbursts are a significant concern for pilots due to the phenomenon creating an abrupt headwind followed by a tailwind. Pilots may correct for the headwind only to find the tailwind puts them dangerously close to stalling. This, accompanied with the downdraft associated with microbursts, is a significant concern for aviation safety, especially when the phenomenon occurs during the landing portion of a flight. This project set out to evaluate the effectiveness of applying genetic algorithms to robust models and its ability to reject disturbances such as microbursts on a landing aircraft. Two types of control systems were analyzed: standard model following and robust model following. Model following is a control technique where an idealized trajectory is defined and followed. Similarly, robust model following is the introduction of a robust servomechanism into model following that allows the control system to address steady state error. Both control techniques use a Linear-Quadratic Regulator (LQR) optimal controller. Using a predefined cost function, genetic algorithms were used to weigh the importance of the states in both the model following and robust model following systems. Comparative analysis of the two approaches reveals that genetic algorithms hold promise in developing simplified yet effective control systems, potentially uncovering patterns that may elude human intuition. Approved for public release: distribution unlimited. PA# USAFA-DF-2025-30