Flight Path Planning Surrogate Model Based on Stacking Ensemble Learning

Abstract

This paper proposes a method of establishing flight path planning surrogate model based on stacking ensemble learning, which can solve the real-time problem of complex flight mission’s on-line waypoints calculation. Airborne navigation system mainly utilizes several discrete waypoints to guide flight path or flight control. These waypoints are usually calculated by a set of equations based on flight dynamics, flight kinematics and flight mission constraints, and therefore path planning for complex missions cannot guarantee real-time performance. In this paper, flight samples are generated offline by taking flight mission characteristic parameters as input and flight waypoint coordinate series as output. Then two-layer coupling model is constructed based on stacking ensemble learning. A series of base-learners are constructed to learn the quantity of waypoints or each waypoint’s coordinate values respectively. At last, flight path planning surrogate model is built by combining all the base-learners, establishing the direct mapping relationship between input and output. The results show that this surrogate model can effectively calculate the aircraft flight waypoints, and meanwhile maintains ideal accuracy and real-time performance.