Multi-Input Single-Output Neural Network Modeling of the System for Regulation of Synchronous Generator's Excitation

This paper deals with the development of the artificial neural network (ANN) models of the synchronous generator's (SG) excitation regulation system. The neural network models have 3 inputs –reference voltage value, the generator's active and reactive power, while the output is the terminal voltage of the generator. Furthermore, the models are developed under various operating conditions, i.e. for different values of the generator's active power. The training process of ANNs is carried out using Levenberg-Marquardt algorithm, while the criterion function is defined as sum of squared errors between real generator's voltage and estimated voltage. The operational data, used for training and validation of ANNs, are obtained from the simulation model of the complete system for excitation regulation of SG. The mentioned model is developed using technical documentation of 40 MVA generator from hydropower plant Perucica.