Neuro-optimal controller for three-phase induction motor based on Levenberg-Marquardt training algorithm

The features of a neural network make neuro-controllers suitable for use in electrical drives, in which the higher speed computation is required. The core of the neurocontrol system is the neurocontroller, and the key technique is the training algorithm. In this paper the features of the Levenberg-Marquardt algorithm are emphasized. The obtained neurocontroller provides an approximate solution of the matrix Riccati differential equation (MRDE). Hence, the obtained neuro-optimal controller performs, during dynamic regimes, the main tasks: smooth response, no oscillations on the control interval, no overshoot, fast compensation of the load torque, and input energy minimization. In order to obtain the accuracy of the optimal solution, the input pattern of the neuro-optimal controller was the choice to include all the nonlinearities of the three-phase induction motor in rotor field oriented coordinates fed by optimal current.