Application of Neural Networks to the Identification of Steady State Equivalents of External Power Systems

This paper suggests an approach based on artificial neural networks to identify steady state equivalents of external power systems. The underlying idea is to train an artificial neural network ANN to learn the behaviour of an external power system. After training, the ANN can be attached to the study system at its boundary buses, replacing the external power system and reproducing its behaviour. The equivalent model proposed in the article expresses the relationship between the power flows in the interconnection lines and the phase and voltage of the boundary buses. Thus, no external system information is required for constructing the equivalent. The model is functional in both directions, i.e using power flows as inputs and phasor voltages as outputs or using phasor voltages as inputs and power flows as outputs. The method was implemented and evaluated on the IEEE-30 bus system and on the Chinese Jiangxi province power system containing 294 buses. The results show that, given appropriate training, the ANN can serve as a model for a power system in an accurate and robust manner. Contrary to the classical methods, the non-linear character of the ANN enables it to accurately model the functioning of the external system also after major operating condition changes such as branch and generator outages.