Abstract This article is aimed to the application of artificial neural networks (ANN) for constructing dynamic equivalents. Neural networks are used to estimate voltages at some special nodes. The use of dynamic ANN allows to quickly reduce the size of the electrical grid due to the rapidity of the training process; this facilitates to study different operating conditions. To quantify the accuracy of results, a root mean square measure is applied to compare signals from the full and the reduced model. The displayed figures and tables prove the applicability and robustness of the proposition.