Department of Electrical Engineering of Tsinghua University, Beijing 100084, China

The paper proposes a mode identification method for low frequency oscillations in power systems based on exact mode order-exponentially damped sinusoids neural network (EMO-EDSNN). Firstly, the mode order is estimated via singular value decomposition. An EMO method is employed to solve the key problem of order determination. It comprehensively considers the variation laws of singular values change and the values themselves, thus overcoming shortages of artificial thresholds and enhancing the accuracy of order determination. Secondly, the EDSNN is constructed to translate the parameter estimation into an optimization problem. After training the neural network via the self-adaptive Levenberg-Marquardt algorithm aiming for a minimum square error between output and real signals, all the mode parameters can be obtained simultaneously. Finally, simulations of numerical signals, EPRI-36 system and actual signals are carried out. The results show that the proposed method can identify the mode parameters in an accurate and reliable manner.