Black-Box Modeling of Synchronous Generators Using Feedforward Neural Networks

This paper provides a black-box modeling approach for grid connected synchronous generators using a specific type of Artificial Neural Network (ANN) known as the Feedforward Neural Network (FNN). The FNN is utilized to establish connection between excitation voltage, active power, reactive power and terminal voltage values. Meanwhile, the weight coefficients of the FNN are determined using the Levenberg-Marquardt algorithm. The data generation required for the experiment was conducted using 297 MVA synchronous generator model in Simulink, and a graphical programming environment based on Matlab. Various experiments were conducted for validation purposes, involving different step disturbance values. The proposed FNN model is accurate and provides a very high degree of matching with the experimental results. Moreover, the test procedure and model are easily implementable, requiring no decoupling of the generator from the grid or additional equipment for realization. The only required measurements are active power, reactive power, excitation voltage and stator terminal voltage. Obtained models can be applied to various testing scenarios related to the excitation system.