A neural network based saturation model for dynamic modeling of synchronous machines

This paper presents a new approach for modeling saturated synchronous machines by employing multi-layer neural networks. In the proposed model, Park´s equations are used as fundamental state-space system in which stator quantities are referred to rotor reference frame and winding fluxes are considered as state variables. Magnetic saturation is also taken into account in both d- and q-axes. First, a conventional model is presented and afterwards, the new approach is described and developed. Using trial and error procedure, an appropriate neural network is employed and trained for modeling saturation using back-propagation algorithm. Several steady state and transient simulations such as short circuit and sudden changes in input quantities are done. Quantitative and qualitative results obtained from simulations show the accuracy and usefulness of the new model as well as valuable information on the topic.