Multi-machine power system dynamic equivalents using artificial intelligence (ANN)

Modern power systems have been large and more complex for interconnections of the countries or electric companies. The simulation of the large power systems for the stability analysis is not necessary to model in detail the entire system. It often requires the use of equivalent representation of circuits and generators. The method of constructing to obtain a dynamic equivalent for a large power system involves several steps: the partition of the system into coherent areas, the coherent area aggregation, and the aggregation of the coherent generators. For this purpose coherency based-dynamic equivalent techniques for large-scale power systems are urgent subjects of research transient stability studies. The performance of these equivalents has been evaluated by series of computational tests. This research is aimed at enhancing the transient stability simulation of large-scale power systems by using coherency technique to reduce the computing time of transient stability studies. The second goal is presentation of artificial neural networks (ANN) on dynamic equivalents for generators. The capability of the proposed ANN based dynamic equivalents, in capturing the nonlinear dynamic characteristics of equivalent generators in complex power system is illustrated, i.e. equivalent generators are replaced by several artificial neural networks (ANN) which are designed in this paper. A two layered, feed-forward, and back-propagated learning, capability ANN was used to emulate the nonlinear models for dynamic equivalents for generators in a large power system. Performance evaluation results due to three-phase faults associated with different locations and/or duration and different loading levels are compared with those obtained using time domain simulations. The results verify that the ANN generator dynamic models can emulate the dynamic equivalents for generators well and should be suitable as a representation method for dynamic stability analysis. The main outcome of this resea- rch is that improves the efficiency and accuracy of constructing a reduced order equivalent model of large-scale power systems into consideration applying the neural networks to save a huge computer time and memory allocations.