基于在线学习误差反传算法的仿真伺服系统设计  

Design of On-line Learning Based Error Back Propagation Algorithm in Simulation Servo System

作　　者：段海滨[1] 王道波[1] 于秀芬[2] 朱家强[1] 

机构地区：[1]南京航空航天大学自动化学院 [2]中国科学院空间科学与应用研究中心,北京100080

出　　处：《中南大学学报（自然科学版）》2005年第2期267-271,共5页Journal of Central South University:Science and Technology

基　　金：国家航空基础科学基金资助项目(01C52015);江苏省"333工程"基金资助项目(2002年)

摘　　要：针对无人机(UAV)仿真伺服系统的驱动模型,提出了一种将误差反传算法用于UAV仿真伺服系统在线学习设计的新方案。在该算法中采用了BP神经网络的基本思想,设计了两输入、单隐层、两输出在线学习策略,输入层分别为给定指令信号和反馈数字解算后的位置信号;隐含层单元数为12 个;输出层设为2 个输出单元,即经在线学习误差反传算法学习训练后的数字位置和速度,其中位置控制器采用自调节比例积分微分(PID)控制,速度通过数字/模拟(D/A)转换后传送到速度控制器,设定精度误差指标为0.05,训练样本数为30。用研制的UAV仿真伺服系统对UAV光纤陀螺传感器进行含实物半物理实时仿真实验,结果表明,该在线学习误差反传算法控制方案的UAV仿真伺服系统具有收敛性好、动态响应快、鲁棒性强的特点。Based on the driving model of unmanned aerial vehicle (UAV) simulation servo system, a novel scheme based on on-line learning error back propagation algorithm (EBPA) was proposed in designing UAV simulation servo system. The idea of back propagation (BP) neural network was adopted in the proposed algorithm. The on-line learning strategy of EBPA with two inputs, single hidden layer and two outputs was applied in this scheme. The input layer included given signal and feedback digital position; the hidden layer had 12 nerve cells; the output layer had two output nerve cells, which were trained digital position and velocity, and self-tuning proportional-integral-differential (PID) control scheme was adopted in the position controller; the digital/analog (D/A) transformed velocity signal was transmitted to velocity controller. The resolution error was 0.05, and the number of training samples was 30. Finally, Hardware-in-loop real-time simulation experiments for a type of fiber optic gyro were conducted in the newly designed UAV simulation servo system. Simulation results illustrate that the UAV simulation servo system u-sing on-line learning based EBPA has good astringency, quick response and strong robust.

关 键 词：误差反传算法 BP神经网络 仿真伺服系统 在线学习 无人机 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置] TP391[自动化与计算机技术—控制科学与工程]