Date of Graduation


Document Type

Problem/Project Report

Degree Type



Statler College of Engineering and Mineral Resources


Lane Department of Computer Science and Electrical Engineering

Committee Chair

Natalia Schmid

Committee Co-Chair

Muhammad Choudhry

Committee Member

Muhammad Choudhry

Committee Member

Sarika Khushalani-Solanki


An active research topic is the detection of various oscillations that may lead to instability and potential disruption in the operation of a power network. Forced Oscillations (FOs) play a unique role in power system stability among various oscillations. They are perturbances that change the system’s state and are caused for many reasons, including but not limited to persistent load changes and oscillatory load or generation, fault, triplane, and other mechanical anomalies. These factors can hugely affect the power grid by either increasing or decreasing the amplitude, causing corrupt modes leading to blackouts, affecting the equipment involved, delivering poor power quality, generator tripping, and impeded efforts to monitor modal oscillations. If detected early, FOs can be isolated and mitigated; however, their detection must happen within the first several seconds of the origin of the oscillations. Thus, the development of fast and effective detection algorithms is a key to preventing a power outage.

The state of the modern power grid is monitored using a network of Phasor Measurement Units (PMUs). EachPMUkeeps track of complex-valued voltage, current, and frequency as time progresses. These time-series measurements are sensitive enough to detect and localize the cause of oscillations in a power network. In this Project Report, we implement an asymptotic generalized likelihood ratio test statistic (GLRT) to detect FOs in a power network and use an autoregressive process to model the power spectral densities of both the ambient noise and FOs. The estimated power spectral densities, using the magnitude of both current and voltage for a subset of 12 PMU sensors in the network, are substituted in the GLRT statistic to analyze if the known FO is detected. We also focus on the effects of AR model order selection on detection performance.

In previous research performed by Pierre, it was hypothesized that the maximum of the detection statistic provided accurate localization of a FO present in the system. Our results demonstrate that the AR model-based GLRT statistic shows excellent reliability in detecting FOs; however, the maximum of the detection statistic is not always observed in the data of the PMU sensor closest to the origin of the FO. Thus, the statement that the location of a FO can be identified based on the location of the maximum of the detection statistic must be investigated further, leaving us with a bulk of future work.