Decentralized secondary voltage control based on neural network and its effect on generator interaction

In the paper, a completely decentralized approach to secondary voltage control is presented utilizing inherent properties of the power system behavior. A modular secondary controller based on an artificial neural network is introduced. The controller is trained on power system data defined by the optimal power flow. It may be attached to a primary voltage controller of a generating unit to which it supplies adequate reference voltage. In the power system, specific problems can arise from interaction between two neighboring voltage regulated buses, where a slight change in operating conditions could lead to inadequate reactive power loading of adjacent generators. It is shown that the presented approach to secondary voltage control can successfully control operation of two adjacent generators. The desired generator operation is obtained without any exchange of data within the power system. The efficiency of the decentralized approach is evaluated in terms of minimizing losses and reactive power production. The tests were carried out on a practical power system model