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 0.009 8°, while the RMS by the improved algorithm is 0.004°, approximately 69.8 μrad, 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.