Wavelet neural network aerodynamic modeling from flight data based on pso algorithm with information sharing and velocity disturbance
来源期刊:中南大学学报(英文版)2013年第6期
论文作者:GAN Xu-sheng(甘旭升) DUANMU Jing-shun(端木京顺) MENG Yue-bo(孟月波) CONG Wei(丛伟)
文章页码:1592 - 1601
Key words:aerodynamic modeling; flight data; wavelet; neural network; particle swarm optimization
Abstract: For the accurate description of aerodynamic characteristics for aircraft, a wavelet neural network (WNN) aerodynamic modeling method from flight data, based on improved particle swarm optimization (PSO) algorithm with information sharing strategy and velocity disturbance operator, is proposed. In improved PSO algorithm, an information sharing strategy is used to avoid the premature convergence as much as possible; the velocity disturbance operator is adopted to jump out of this position once falling into the premature convergence. Simulations on lateral and longitudinal aerodynamic modeling for ATTAS (advanced technologies testing aircraft system) indicate that the proposed method can achieve the accuracy improvement of an order of magnitude compared with SPSO-WNN, and can converge to a satisfactory precision by only 60-120 iterations in contrast to SPSO-WNN with 6 times precocities in 200 times repetitive experiments using Morlet and Mexican hat wavelet functions. Furthermore, it is proved that the proposed method is feasible and effective for aerodynamic modeling from flight data.
GAN Xu-sheng(甘旭升)1, DUANMU Jing-shun(端木京顺)2, MENG Yue-bo(孟月波)3, CONG Wei(丛伟)2
(1. Department of Basic Courses, Xijing College, Xi’an 710123, China;2. Engineering College, Air Force Engineering University, Xi’an 710038, China;3. Systems Engineering Institute, Xi’an Jiaotong University, Xi’an 710049, China)
Abstract:For the accurate description of aerodynamic characteristics for aircraft, a wavelet neural network (WNN) aerodynamic modeling method from flight data, based on improved particle swarm optimization (PSO) algorithm with information sharing strategy and velocity disturbance operator, is proposed. In improved PSO algorithm, an information sharing strategy is used to avoid the premature convergence as much as possible; the velocity disturbance operator is adopted to jump out of this position once falling into the premature convergence. Simulations on lateral and longitudinal aerodynamic modeling for ATTAS (advanced technologies testing aircraft system) indicate that the proposed method can achieve the accuracy improvement of an order of magnitude compared with SPSO-WNN, and can converge to a satisfactory precision by only 60-120 iterations in contrast to SPSO-WNN with 6 times precocities in 200 times repetitive experiments using Morlet and Mexican hat wavelet functions. Furthermore, it is proved that the proposed method is feasible and effective for aerodynamic modeling from flight data.
Key words:aerodynamic modeling; flight data; wavelet; neural network; particle swarm optimization