Remarks on an Optimal Predictive Control Using a Quaternion Neural Network-based Identifier

This study explores the application of quaternion neural networks (QNNs) in control systems, specifically focusing on optimal predictive control. The approach introduces a model predictive control method for achieving optimal control of non-linear systems. In this method, a direct transfer function identifier is constructed using a QNN to estimate the future predicted output of the plant. The optimal control input is then determined by minimising the objective cost function which comprises the predictive control error and the increment of the control input. A derivative-free optimisation method is employed for this purpose. Computational experiments are conducted to assess the effectiveness of the optimal predictive control using the QNN-based identifier. These experiments are performed in the context of a tracking task involving a discrete-time non-linear plant. The simulation results demonstrate that the proposed method executes the control task with appropriate accuracy. Furthermore, the findings affirm the feasibility of employing QNNs for controlling non-linear systems.