忆阻神经网络自适应滑模控制及其应用

Adaptive sliding mode control by memristor-based neural network and its application

光电吊舱系统中存在各种扰动和未建模动态,常规控制算法难以适应复杂情况。采用神经网络实现模型未知部分的自适应估计,结合滑模变结构控制,可以有效提升控制精度,但在控制初始阶段,神经网络估计没有收敛到实际模型时,滑模控制存在抖振现象。因此,提出了基于忆阻器神经网络自适应滑模控制算法,采用神经网络可以逼近未建模态,提升控制精度,而忆阻器神经网络来保存权值参数,可以减小神经网络收敛时间。在控制的初始段,改进自适应增益来减小由于神经网络估计误差带来的抖振现象,仿真与实验结果表明,采用忆阻神经网络及改进的自适应增益算法,初始抖振显现得到了很好的控制,算法收敛时间减小为常规滑模控制算法的1/2,稳态控制精度较常规滑模变结构控制算法提升了59.18%。

Objective In the optoelectronic pod system,there are various disturbances and unmodeled dynamics.Therefore,it is difficult for conventional control algorithms to adapt to complex situations.The neural network is adopted to realize the adaptive estimation of the unknown dynamics of the model,combined with sliding mode variable structure control,the control accuracy can be effectively improved.However,if the neural network estimation fails to converge to the parameters in the actual model at the initial control stage,chattering phenomenon will arise in the sliding mode control.In order to achieve fast convergence of neural network estimation,suppress the chattering at the initial stage of sliding mode control,and improve control accuracy and stability,the algorithm of adaptive sliding mode control based on memristor-based neural network is proposed herein.Methods An improved memristor-based neural network is adopted to store the weight parameters to approach the unmodeled dynamics,which can reduce network convergence time and improve control accuracy compared to the conventional neural network.In the initial stage of sliding mode variable structure control,a neural network based on memristors is adopted.The adaptive gain is improved to reduce the chattering caused by estimation error of neural network.The improved algorithm in overall significantly reduced the chattering and quickly and accurately estimated unmodeled dynamics,enhancing control accuracy and stability.Under analog simulation conditions,the improved algorithm is compared with conventional sliding mode variable structure method regarding to the sinusoidal position response,and the result shows that the convergence time by the improved algorithm is reduced to half of that of the conventional sliding mode control algorithm(Fig.9).When an actual unmanned aerial vehicle tracking detection is conducted in the outfield,the control accuracy under the improved algorithm is increased by 59.18%compared to the conventional sliding mode control algorithm(Fig.12).Results and Discussions Under analog simulation conditions,compared with conventional sliding mode variable structure method,the convergence accuracy for the sinusoidal position response by adopting the improved algorithm is within 0.0002°while the one by conventional algorithm is within 0.001°,which means the convergence time by the improved algorithm is reduced to half of that of the conventional sliding mode control algorithm(Fig.9).When an unmanned aerial vehicle targets detection is conducted in the outfield,with a maximum speed of maneuvering flight of 15 m/s and a distance of 1 km from the unmanned aerial vehicle to tracking turntable,the stably tracking miss distance(RMS)by the conventional sliding mode control algorithm is 0.0098°,while the RMS by the improved algorithm is 0.004°,approximately 69.8μrad,resulting in the increase of accuracy under the improved algorithm by 59.18%compared to the conventional sliding mode control algorithm(Fig.12).Conclusions By adopting the improved algorithm of adaptive sliding mode variable structure control based on the memristor-based neural network,the convergence time of estimation for unknown unmodeled dynamics is reduced,up to half of that of conventional sliding mode control algorithm.In an actual outfield detection experiment,the stably tracking control accuracy by the improved algorithm is increased by 59.18%compared to that by the conventional sliding mode control algorithm.The experimental results show that the use of the improved algorithm of adaptive sliding mode variable structure control based on the memristor-based neural network can not only help the system to realize fast convergence and suppress chattering,but also effectively improve the tracking accuracy and stability of the optoelectronic pod system,which has certain application value in engineering.