A multiple connected recurrent neural network based super-twisting terminal sliding mode control for quad-rotor UAV

Considering the complex environment in which quad-rotor unmanned aerial vehicle (UAV) perform their tasks, it is difficult to establish an accurate UAV model and obtain information about external disturbances. Aiming at the above problems, a multiple connected recurrent neural network (MCRNN) based on super-twisting algorithm (STA) terminal sliding mode control (TSMC) strategy is proposed. Due to the unknown dynamics, the equivalent control law of sliding mode control cannot be directly applied to UAVs. Therefore, the MCRNN controller is used to approximate the equivalent control rather than to estimate the dynamics of the quad-rotor UAV. All hidden layer neurons in MCRNN receive self-feedback as well as signals from other hidden layer neurons, thereby augmenting their capacity to capture intricate dynamic features. In addition, the robustness of the sliding mode is used to suppress the mismatched disturbance instead of the traditional disturbance observer. This solution is more flexible, and reduces computing costs. Lyapunov stability theory is used to ensure the finite-time stability of the whole system, and the real-time update law of MCRNN weights is derived. Finally, the proposed method is applied to a path-following task, obtaining a maximum overshoot of 4.58e−02 and the settling time of 0.935s. By comparing the results obtained by different methods, it is concluded that the proposed controller is insensitive to model parameter variations, is able to suppress mismatched disturbances, and has significant stability and robustness.