基于在线学习误差反传算法的仿真伺服系统设计

来源期刊：中南大学学报(自然科学版)2005年第2期

论文作者：段海滨 王道波 于秀芬 朱家强

文章页码：267 - 271

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

Key words：error back propagation algorithm; back propagation neural network; simulation servo system; on-line learning; unmanned aerial vehicle

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。

Abstract: 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 proportionalintegral-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-timesimulation 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 using on-line learning based EBPA has good astringency, quick response and strong robust.

基金信息：国家航空基础科学基金资助项目
江苏省“333工程”基金资助项目