Sensorless ANN-Based Speed Estimation of Synchronous Generators: Improved Performance through Physically Motivated Pre-filters

This paper deals with an improved approach for estimation of generator rotor speed in power systems. We propose a hybrid speed estimator, which combines classical approach, based on a mathematical model of the electrical machine, with AI-based approach, implemented by an artificial neural network (ANN). The classical estimation model processes the machine stator voltages and currents and provides intermediary information on the generator speed. The ANN component acts as a function approximator, mapping the intermediary speed components to a more reliable estimation of the generator speed. Multi-layer feedforward ANNs are used for this purpose. Data for training the ANN are obtained through off-line simulations of a generator operating in a two-machine model. After ANN design and training, the performance of the hybrid estimator is tested with simulated on-line measurements in a wide range of operating conditions. Results obtained with the hybrid estimator are compared against those provided by a classical mathematical model-based estimator and an AI-based estimator.