EIMC2 Talk

Tuesday, November 12, 2024 - 4:00pm

Event Calendar Category

Other LIDS Events

Speaker Name

Deepjyoti Deka

Affiliation

MIT Energy Initiative

Building and Room number

45-500A

"Fast optimization of generator dispatch and associated risk estimation are crucial to ensuring reliable and economical delivery of electricity in power grids."

AC Optimal Power Flow (OPF),  with non-linear AC power flow equations and constraints, is the primary optimization framework used for power grid operational and planning problems. In this talk, we present results on two Bayesian data-driven proxies to learn AC power flow equations and AC -OPF surrogates using limited data, with a focus on probabilistic predictions. 

The first data-driven model uses graph-structured Gaussian process (GP) model for risk assessment of critical voltage constraints. The proposed GP is based on a novel kernel, named the vertex-degree kernel (VDK), that decomposes the voltage-load relationship based on the network graph, and is estimated using Active Learning. Simulations demonstrate that the proposed VDK-GP achieves a reduction in computational time by 15 times over Monte- Carlo simulations in estimating system risk. 

The second model termed BayesOPF is a Bayesian Neural Networks (BNN) based proxy for AC-OPF using limited OPF training time and samples. BayesOPF is trained in a sandwiched manner that alternates between a supervised learning step for minimizing cost, and an unsupervised learning step for enforcing feasibility.  We show that BayesOPF outperforms conventional neural network architectures on OPF problems, without requiring any correction or projection step. 

Deepjyoti Deka is a research scientist at MIT Energy Initiative. His research interests lie at the intersection of machine learning and optimization for tractable sensing and secure operation in renewable rich power grids. From 2018-2024, he was a staff scientist in the Theoretical Division at Los Alamos National Laboratory and served as a PI/co-PI for DOE and internal projects on machine learning in power systems and optimization of interdependent networks. He received the M.S. and Ph.D. degrees in Electrical and Computer Engineering (ECE) from the University of Texas, Austin, TX, USA, in 2011 and 2015, respectively. He completed his undergraduate degree in Electronics and Communication Engineering (ECE) from IIT Guwahati, India with an institute silver medal as the best outgoing student of the department in 2009. Dr. Deka is a senior member of IEEE, and has served as an editor on IEEE Transactions on Smart Grid.

The newly formed Energy Systems & Infrastructures: Modeling, Computing and Control (EIMC2) LIDS research group comprises 10 LIDS subgroups working in this field unique to LIDS–modeling, control, and computing. EIMC2 is launching a biweekly seminar series. The seminars will be help in 45-500A on Tuesdays, 4-5pm unless otherwise noted.

For any questions, please reach out to seminar organizers Rahman Khorramfar (khorram[at]mit[dot]edu) and Luis Carlos (luiscvm[at]mit[dot]edu)