| Abstract |
The rotor field oriented induction motor (IM) is controlled at constant flux, the optimal control synthesis consisting of the determination of the statoric three-phase currents system, based on the longitudinal and transversal components of the statoric phasor current. Thus, the obtained linear mathematical model of IM becomes more accessible in view of the adaptive control [1], optimal control techniques [2], and so on. The goals of the authors to the electrical drive have been provided by using a quadratic energetically performance criteria offered by the optimal control theory [2][3][4]. Therefore, as the problem formulation was done with fixed time, the matrix Riccati differential equation [MRDE] must be solved. This task was performed in [5][6], but it is a more expensive time. To overcome this problem it will be shown the approximation of the optimal control solution by using a feed forward neural network (NN) in the paper. To improve the backpropagation learning algorithm, the momentum method and adaptive learning rate were used. The architecture of the NN is presented in this paper. After the successfully tested NN, this could be implemented with an adequate weights connections and bias downloaded in the learning frame, by using a modest digital controller. The optimal control law oriented to the vector controlled induction machine provides dynamic regimes with minimal energy consumption. |
