Customized Neural Network System for Dynamic Security Preventive Control

Abstract

This paper proposes a new methodology for dynamic security assessment and preventive control. In the first phase, an Artificial Neural Network (ANN) is trained to provide the security status. ANN inputs are settled by a feature selection approach that takes into account the requisites of the control algorithm, to be applied in the second phase. The adaptive control methodology is based on the Steepest Descent method, where the usual explicit math functions to be dealt with are emulated by the trained ANN. In other to illustrate the developed approach, the methodology was applied to the control of dynamic security of Madeira island power system. Results attained so far show that the proposed approach was able to find the optimal control actions.