A prediction method for deck-motion of air-carrier based on PSO-KELM

Abstract

Prediction for deck-motion is a practical measure to improve the landing/taking off safety of carrier-based aircraft when those deck-motions in six-degree freedoms cannot be effectively controlled/restrained. Deck-motions excited by waves and winds own characteristics of randomness and nonlinearity. It is generally believed those classical feed-forward neural networks, such as back propagation networks have excellent nonlinear fitting ability but suffers from slow training speed and local optimum falling which cannot satisfy those real-time and high accuracy requirements for deck-motion. In this paper, a prediction method based on extreme learning machine, support vector machine and particle swarm optimization (PSO-KELM) is introduced to fulfill deck-motion. In this method, the fundamental structure of extreme learning machine is used but the hidden function is substituted the kernel function from support vector machine. Further, aiming to select optimal parameters including penalty coefficient and kernel parameter, auto-adaptive particle swarm optimization is adopted. Simulation results indicate that the prediction method based on PSO-KELM owns advantages of simple structure, fast training speed and good generalization ability, and can obtain high accuracy prediction results when used for deck-motion prediction of air-carrier.