A model-free sparse wide-area damping controller for inter-area oscillations

As the successor of local control, wide-area damping control (WADC) is considered to be an effective tool for damping inter-area oscillations. The state-of-the-art WADC approaches are either model-based or centralized, so they are subject to model inaccuracy, bad data, high communication costs, and other nontrivial design challenges. On the other hand, current data-driven WADC practices are affected by the selection of input/output measurements, need to be pre-tuned and trained offline, or lack clear physical interpretation of the power grid small-signal dynamics. In this paper, we propose a novel online purely data-driven WADC method using phasor measurement unit (PMU) data, which requires no power system model information and can damp multiple inter-area modes simultaneously using minimal control effort. To directly shape the closed-loop characteristics of poorly-damped inter-area modes and account for physical constraints on communication network, we develop a modal linear quadratic regulator (MLQR)-based sparsity-promoting optimal state feedback controller. The developed WADC strategy does not require offline training, is adaptive to the applied PMU dataset and can be mapped to the actual grid physics. Dynamic simulations on the widely studied IEEE 68-bus benchmark system verify the effectiveness of the proposed approach in spite of measurement noise, PMU losses, and communication network restrictions. Moreover, it is shown that the proposed controller can adapt to different operating conditions due to its data-driven nature. Finally, our data-driven sparse WADC has comparable performance to conventional model-based and centralized data-driven WADC, while overcoming their limitations.

Automated summary

In this paper, a novel online purely data-driven wide-area damping control (WADC) method using phasor measurement unit (PMU) data is proposed. It can damp multiple inter-area modes simultaneously and adjust the closed-loop characteristics under the constraints of the communication network. This method does not require offline training, is adaptive to the applied PMU dataset, and can be mapped to the actual grid physics.