Model Predictive Sliding Mode Control with Neural Network for UAVs

To operate an unmanned aerial vehicle (UAV) within a set of maneuverability constraints, herein, we use the Model Predictive Control (MPC) method to find the optimal control input under various control input constraints. Although the MPC method can predict future states and reflect them it the present state optimally, in real-world scenarios, its computational load increases exponentially with the number of state variables and the length of the time window. To reduce the computational burden and obtain the predicted optimal control input by using the MPC technique, herein, we devise an approach involving neural networks. To evaluate the weighting parameters of the neural networks, a considerable volume of learning data is required, which can be generated by conducting numerical simulations. Herein, we generate the input and output data pairs for a given time window by using the MPC method and by means of simulations. This learning process is expected to mitigate the computational burden dramatically. Lastly, one of the drawbacks of MPC is that the model for evaluating the optimal control input for a given time interval must be extremely accurate to guarantee system stability. Therefore, to increase the robustness of the MPC method against external disturbances and internal uncertainties, we augment it by using the Sliding Mode Control (SMC) method. The effectiveness of the suggested neural-network-based Model Predictive Sliding Mode Control method is demonstrated by means of numerical simulations.