Real-Time Data-Driven Electromechanical Oscillation Monitoring using Dynamic Mode Decomposition with Sliding Window

Abstract

Due to the complexity of the power system model, model-free or data-driven methods are promising for real-time electromechanical oscillations monitoring and allow grid operators to better manage the grid security and maximize transfer capacity during real-time operation. Dynamic mode decomposition (DMD) is a promising data-driven method and has been recently applied for electromechanical oscillations monitoring. However, it is still not clear what influence the length of time-window, power system eigenvalues and the use of data from pre-, during, and post-disturbances have on the estimation accuracy of the DMD. This work aims to investigate the above issues by performing a systemic analysis on three benchmark test systems. It is shown that the ultra-low frequency mode and large disturbances can negatively affect the estimation result of DMD method. In addition, it is found that the time-window length of 10 s is suitable in ensuring the best estimation accuracy/performance of the DMD with a sliding window.