Development and implementation of neural network observers to estimate synchronous generators' dynamic parameters using on-line operating data

Abstract

This paper illustrates a new application of artificial neural network (ANN) observers in identifying and estimating synchronous generator dynamic parameters via time-domain, on-line disturbance measurements. To prepare the training database for an ANN observer, the transient behaviours of synchronous generators have been determined through off-line simulations of a generator operating in a one-machine-infinite-bus environment. The Levenberg–Marquardt optimization utilising very fast back propagation algorithm has been adopted for training feed-forward neural networks. The inputs of ANNs are organized in coordination with the data from the observability analysis of synchronous generator parameters in its dynamic behaviour. A collection of ANNs with same inputs but different outputs is developed to determine a set of the parameters. The ANNs are utilized to estimate the above parameters by the measurements for every kind of fault separately. The robustness tests are executed by on-line measurements to identify the parameters. Simulation studies not only indicate that the observer is capable to identify the dynamic parameters of synchronous generator but also show that the tests which have given better results in identification of each dynamic parameter can be acquired.