摘要
针对四旋翼飞行器的强耦合、欠驱动、非线性且无法精确建模等问题,设计一种能够自主调节飞行器控制参数,且在工程实践中易于实现的径向基神经网络PID控制算法。该算法以高斯激活函数为主体,采用梯度下降法训练网络的中心矢量及权值参数,得出网络输入/输出之间的非线性关系,最后用于修正位置环PID的控制参数。搭建四旋翼飞行器实物平台,通过实验研究算法的控制性能。实验结果表明,神经网络PID控制算法控制效果优良,不依赖系统的精确建立且具有较强的鲁棒性及自适应能力。
Since the quadrotor aircraft has the problems such as strong coupling,underactuation,non-linearity and incapability of accurate modeling,a radial basis neural network PID control algorithm that can autonomously adjust the control parameters of the aircraft and is easy to be implemented in engineering practice is designed.In the algorithm,with the Gaussian activation function as the main body,the center vector and weight parameters of the network are trained by using the gradient descent method.The nonlinear relationship between the input and output of the network is obtained to modify the PID control parameters of the position loop.The physical platform of the quadrotor aircraft was built.The control performance of the algorithm was researched in virtue of experiments.The experimental results show that the neural network PID control algorithm has a good control effect,does not depend on the precise construction of the system,and has strong robustness and self-adaptability.
作者
余润芝
赵文龙
程若发
YU Runzhi;ZHAO Wenlong;CHENG Ruofa(College of Information Engineering,Nanchang Hangkong University,Nanchang 330063,China)
出处
《现代电子技术》
北大核心
2019年第10期108-112,共5页
Modern Electronics Technique
基金
国家自然科学基金(51567019)
江西省教育厅科技项目资助(GJJ150701)
江西省研究生创新专项资金项目资助(YC2017-S328)~~
关键词
四旋翼飞行器
神经网络
PID
径向基
高斯激活函数
梯度下降法
quadrotor aircraft
neural network
PID
radial basis
Gaussian activation function
gradient descent method
