Online Learning and Optimization in Distributed Energy Systems: Some Problems and Opportunities
summary
Shows practical algorithms for two problems: topology learning using a mutual-information Chow-Liu maximum-weight spanning tree and complex-variable group LASSO regression, tested on real PG&E and NREL data including a 3,000-bus reconstruction run thousands of times with zero errors; and delayed-cloud coordination where a cloud MPC publishes a nominal trajectory plus per-home bounds, implemented in lab and RPE trials, which cuts voltage violations and captures about 80% of the perfect-foresight arbitrage even with 24-hour delays.
AI-written summary · about
chapters (13)
- 0:00Intro
- 0:41Changing Electricity Grid
- 2:09How to schedule distributed energy resources?
- 3:57In Practice
- 6:38Problem Formulation
- 9:28Proof: Assume Data Generation Process
- 17:18New Probabilistic Description
- 17:52Lasso-based Algorithm
- 19:03Neighbor Set Estimation
- 21:55DER Scheduling Case Study
- 23:18Electric Power Quality (EPQ) Metric
- 25:44Coordination Challenge
- 31:36Conclusions
related talks
15mComputational Mobility: Learning, Prediction, and Algorithms
36mTechnical Research Fueling NextGen Actions
29mTowards GridFM: GENCO – a Unified Neural Solver for Steady-stat...
17mTopoOpt: Co-optimizing Network Topology and Parallelization Strategy for Distributed...
01h 07mIndustry Presentations
25mPCC Vivace: Online-Learning Congestion Control
47mRecent trends in combinatorial optimization augmented machine learning: A graph learning perspective
29mState Management in Distributed Online Learning with Ray