Information Geometry for Parameter Estimation in Large Energy Systems

Време17. јун 2016. 15:00
Предавачдр Александар Станковић
Местосала 61, Eлектротехнички факултет, Београд


Dynamic models of power systems (for example, electromechanical models used in transient stability) have grown in size to thousands of generators and tens of thousands of nodes. However, this growth in quantitative terms has largely not been accompanied with improvements in fidelity. Specifically, models have been largely unable to replicate major events like the 2003 blackout in the U.S. Eastern interconnection, even with of the widespread sensor coverage. Three recent trends have highlighted the model fidelity problems:

  • Market-driven operation leads to new flow patterns, unfamiliar to the operators,
  • Renewable energy sources such as solar and wind lead to new (and volatile) flow patterns,
  • New types of sensors such as Phasor Measurement Units (PMUs).

The premise of the information geometric approach is that a (nonlinear) model with many parameters is a mapping from a parameter space into a data space. A key difficulty in dealing with models of complex systems is the highly anisotropic mapping between the parameters and data spaces. Recent advances focusing on data space rather than parameter space have proven beneficial for understanding the global properties of models and for advancing numerical techniques for exploring them. The first part of the talk will review local methods for assessing reliability of parameter estimates. Next, we present semi-global and global methods that complement local analyses. We then present a model reduction procedure based on the Manifold Boundary Approximation Method (MBAM) and compare it with ideas standard in energy systems such as singular perturbation. Finally, we outline a path toward applying information geometric concepts in large interconnections.


Aleksandar M. Stankovic obtained the Ph.D. degree from Massachusetts Institute of Technology in 1993 in electrical engineering. He serves as the A.H. Howell Professor at Tufts University; he was with Northeastern University, Boston 1993-2010. He is a Fellow of IEEE (2005) and serves as an Associate Editor for IEEE Transactions on Power systems and for Annual Reviews in Control. He previously served Transactions on Smart Grid and on Control System Technology in the same capacity (1996-2010). He has held visiting positions at the United Technologies Research Center (sabbaticals in 2000 and 2007) and at L'Universite de Paris-Sud and Supelec (in 2004). He is a co-editor of book series on Power Electronics and Power Systems for Springer.