Real‐time identification of electromechanical oscillations through dynamic mode decomposition

Abstract

Nowadays, traditional synchronous generators are replaced by distributed renewable energy sources (DRESs), which are connected to the grid via power converters. This shift towards non-synchronous generation leads to low inertia power systems and affects considerably the frequency control procedure. To provide an inertial response and to enhance grid stability, DRESs can be equipped with fast discharging energy storage systems, such as ultracapacitors. This feature allows distribution system operators (DSOs) to provide an inertial response as ancillary service to transmission system operators (TSOs) by coordinating the operation of many DRESs. For this purpose, DSOs should develop tools to quantify and control the provision of inertial response. Towards this objective, in this study, two new methodologies are proposed. The first one aims to evaluate the maximum aggregated inertial capability of active distribution networks (ADNs). The second one aims to dispatch DRESs to provide specific, TSO-defined, inertial response with the minimum cost. Both methods are tested on a medium voltage grid. Several cases are examined highlighting the impact of different parameters, such as converter limitations, line congestions, on the inertial capability of ADNs. Results indicate that the proposed methods can fully exploit operational limits of ADNs and maximise their inertial capability.

An improved dynamic mode decomposition for real‐time electromechanical oscillation monitoring in power systems: The impact of ultra‐low frequency modes and its removal strategy

“…DMD is one of the data-driven methods that has the potential to be used in the power system application due to its robustness and estimation speed. Specifically, DMD has been used for electromechanical oscillation monitoring [2][3][4][5][6][7][8][9][10][11][12][13][14][15][16] estimation of inertia [17] and frequency [18,19], harmonic distortion monitoring [20,21], fault detection [22], load forecasting [23][24][25] and data-driven control for enhancing rotor angle stability [26,27]. Regarding the oscillation monitoring using DMD, there are still some challenges related to the amount of input data that need to be addressed for achieving real-time oscillation monitoring.…”

“…Feature addressed Target Data type New technique proposed [3][4][5] Modal identification ability Analysis Ringdown None [6] Robustness [7,8] Computational speed [9] Accuracy [2], [10][11][12], [14][15][16] Robustness Improve optimization, data stacking, randomization, data processing, energy evaluation [13] Accuracy Non-linear mapping FIGURE 1 The behaviour of speed signal during real-time operation (ambient and ringdown data).…”

“…On the other hand, [15] and [16] are motivated by the robustness improvement of the DMD. For example, [15] focuses on the detection of the most dominant modes by adding additional step into the DMD and calculating the energy of the modes. Ref.…”

“…ULFMs can also be observed in Figure 9 of [11]. However, the characteristics of ULFMs and trends contained in the ringdown data are not discussed and analysed in [2][3][4][5][6][7][8][9][10][11][12][13][14][15][16].…”

“…However, as the scope of the paper is on improving the robustness and accuracy of the DMD method, the issue of reducing the window length while maintaining a good estimation's accuracy is not discussed in the paper. On the other hand, [15] and [16] are motivated by the robustness improvement of the DMD. For example, [15] focuses on the detection of the most dominant modes by adding additional step into the DMD and calculating the energy of the modes.…”