Echo State Network (ESN) Based Generator Speed Prediction of Wide Area Signals in a Multimachine Power System

This paper presents an echo state network based prediction of generator speed using a two area four generator multimachine power system in real time. The echo state network (ESN) is trained using the generator speed deviation delayed between 20ms to 50ms as inputs. The output of the ESN is the predicted speed of the generators for steady state scenarios. The trained and test data were compared, the results show a good correlation between these data sets demonstrating that the ESN has effectively learned the nonlinear properties of the multimachine system. This paper also uses ESN to address the need to mitigate power system stability issues time steps ahead e.g. early detection of oscillations in the power systems network that could cascade into system failure, blackouts, and sudden system collapse. Echo State Networks have been used in this research due to its superior performance in terms of the effectiveness of learning nonlinear systems and its ease of training compared to other recurrent neural networks (RNN). The IEEE two area four generator test system was modeled in RSCAD software and tested in real time using the RTDS (real-time digital simulator). The real time implementation was carried out at the Real Time Power Intelligent Systems (RTPIS) laboratory, Clemson University, SC. USA.