Optimization and Statistical Learning for Energy Systems

Team leader: Dr. Yury Maximov

The group: aims & structure

  • Involves perspective young mathematicians/computer scientists as research interns in engineering research
  • young team members work under joint supervision of mathematicians / computer scientists and energy systems experts (physicists, power flow engineers, etc) to achieve better performance
  • We focus on mathematical problems of energy systems and problems with the same mathematical nature arise in other areas
  • Our updated web site“Optimization and Statistical Learning for Energy Systems” https://oslgroup.wordpress.com/about/
  • Research seminar: https://oslgroup.wordpress.com/research-seminar/
  • Admission rules: https://oslgroup.wordpress.com/the-team/

 

Principal research directions:

  • AC Optimal Power flow optimization: 
    • Tighter relaxations for small scale instances
    • Scalable approximations of Optimal Power Flow problem through statistical learning/sampling and stochastic optimization techniques
  • AC Power flow feasibility:
    • The problem is to find the all possible values of parameters under those the system of AC power flow equations is feasible under line/bus constraints
    • The goal is to develop machine learning techniques for computing feasibility set (set of parameters stands for the normal operating of the grid)
  • Statistical Learning and Stochastic optimization: theory and applications include
    • model selection, dimension
    • excess risk bounds on binary and  multi-class classification
    • mirror descent, its variants and applications in learning theory and energy systems
  • Optimization for Transportation Science and Network Analysis:
    • industrial projects on traffic optimization (among them joint projects with Huaweiand Genplan Moscow)
    • optimization and control for transmission grids

In all the problems above there is a bunch of applied mathematics: from approximation algorithms, statistics and optimization to complex analysis and algebraic geometry.

 

Recent Representative Papers:

  • Bikash Joshi, Massih-Reza Amini, Ioannis Partalas, Franck Iutzeler, Yury Maximov. Aggressive Sampling for Multi-class to Binary Reduction with Applications to Text Classification. arXiv:1701.06511
  • Roman Pogodin, Alexander Katrutsa, Sergei Grudinin. Quadratic Programming Approach to Fit Protein Complexes into Electron Density Maps. arXiv:1701.02192
  • Andrii Riazanov, Mikhail Karasikov, Sergei Grudinin. Inverse Protein Folding Problem via Quadratic Programming. arXiv:1701.00673
  • Kovaleva, Yu. Maximov, S. Nechaev, O. Valba. Peculiar spectral statistics of ensembles of trees and star-like graphs. arXiv:1612.01002
  • Yury Maximov, Massih-Reza Amini, Zaid Harchaoui. Rademacher Complexity Bounds for a Penalized Multiclass Semi-Supervised Algorithm. arXiv: 1607.00567
  • Daria Reshetova. Multi-class Classification: Mirror Descent Approach. arXiv:1607.00076
  • Yury Maximov, Daria Reshetova. Tight Risk Bounds for Multi-Class Margin Classifiers. arXiv:1507.03040

 

 Team:

 1 PhD student

  • Kirill Polovnikov, started at CES at the end of 2016

3 master students at Skoltech

  • Mikhail Karasikov (CDISE), Marina Danilova (CES), Alena Shilova (CDISE)

8 part-time research interns

  • Nikita Doykov, Oleh Horodnitskii, Mikhail Krechetov, Valerya Kovaleva, Igor Molibog, Roman Pogodin, Andrii Riazanov, Sergei Volodin

–Team co-leaders Elena Gryazina, scientific researcher at CES Skoltech and Yuriy Dorn (Moscow Institute of Physics and Technology)

Selected research topics: 

  • Sergei Volodin (Intern), “Power Flow Feasibility through Certificate Cutting”, under joint supervision with Prof. Anatoly Dymarsky & Dr. Elena Gryazina (Skoltech)
  • Igor Molibog (Intern) “Power Flow Steady State through Active Loss Regularization”, under joint supervision with Prof. Mikhail Davidson (Skoltech & Carana)
  • Mikhail Krechetov (MSc) & Roman Pogodin (Intern), “Schatten (pseuso)-Norm Relaxation of Optimal Power Problem”
  • Oleh Horodnitskii (Intern), “Learning feasibility domain with quadratic proxies”, under supervision of Yury Maximov, Prof. Dymarsky, Prof. Turitsyn (MIT), Prof. Chertkov
  • Alena Shilova (Intern), “Importance Sampling for Power Flow Feasibility”, helps to query less number of points to master up the performance of learning method.
  • Valerya Kovaleva (Intern), “Rare Event Statistics and Applications”, under joint supervision with Prof. Sergei Nechaev(CNRS/Poncelet)

 

Resent visitors:

  • Ekaterina Krymova, U. Duesburg Essen, October & December, 2016 (topic: Principal Component Analysis for Time Series)
  • Massih-Reza Amini, U. Grenoble Alpes, October 2016 (topic: Multi-class Classification and Applications)