Visualization:

• Implemented novel software interfaces for visualizing anomalous events in grid networks.

• Designed a novel visualization technique called a 'cluster dendrogram' for analyzing the spread of anomalous data from a source node in the grid network.

Structure learning on graphs: 

• Devised novel landscape theory and algorithms for quadratic feasibility problems -- a foundational paradigm for estimation in power systems. 

• Devised novel algorithms and theory for the problem structure learning of general graphical models when measurements are corrupted noise. 

• Created a novel optimization framework for learning the topology of the distribution grid with unobserved nodes

• Pioneered novel methods for localization of signals on a network efficiently (and near optimally) from noisy measurements

• Developed a label-efficient algorithm for finding and leveraging the homology of decision boundaries

• Devised a novel framework for efficiently learning the topology of networks that obey conservation laws from potential measurements

Event detection, Identification, and Security:

• Designed an event identification platform by combining physics-based methods with intelligent classification techniques. These techniques leverage statistical techniques of bootstrapping and explainable models to identify events in the settings of limited event data yet with high-dimensional feature space for each event. 

• Developed semi-supervised learning techniques to identify power systems events with a small number of labeled samples in addition to unlabeled samples. These semi-supervised techniques, including label spreading and self-trained SVMs, with only a few labeled samples achieve performance similar to that of hundreds of labeled samples in the supervised setting.

• Rigorous testing using both on large synthetic power system networks and on proprietary data from a US utility.

Synthetic Data Generation:

• Focused on generating synthetic load data profiles for the grid

• Used 100TBs of available PMU data to do so

• Designed both simple linear and complex deep-learning based synthetic data generation techniques. 

• Showed that linear methods work for low granularity limited sampling settings while generative adversarial networks based models work very effectively for high temporal resolution settings

• Verified the efficacy of such datasets for many practical power systems applications where load data is essential 

Parameter Estimation

• A novel approach (called EGLE) for phasor measurement unit (PMU)-based transmission line parameter estimation in presence of non-Gaussian measurement noise 

• Time-synchronized topology identification (TI) and unbalanced three-phase distribution system state estimation (DSSE) in real-time unobservable primary distribution networks using deep learning 

• Statistical characterization of the measurement errors introduced by a phasor measurement unit (PMU) 

• A low-cost, inductively-powered, time-synchronized, micro point-on-wave (PoW) recorder for real-time monitoring of modern distribution systems

• A new graph theoretic approach for analyzing whether a contingency will create a saturated cut-set in a meshed power network 

• A coordinated wide-area damping controller (CWADC) for mitigating low frequency oscillations (LFOs) using an enhanced selective modal analysis (SMA) and linear matrix inequality (LMI)-based polytope 

The funded research has led to many publications, software tools, and broader tools for the power systems community to use. It has also allowed the training of nearly 10 PhD students and three postdoctoral fellows at ASU.

Last Modified: 01/11/2024
Modified by: Lalitha Sankar