PhD dissertation: Wind Scenarios for Stochastic Energy Scheduling University of Washington
My doctoral research focused on how to create wind scenarios for the stochastic scheduling of power systems. Power systems are scheduled to reliably meet the next day's demand at lowest cost. This scheduling process is an optimization problem known as unit commitment. Wind and other uncertain renewables make this scheduling less efficient. So rather than scheduling based on a single forecast, stochastic unit commitment minimizes the expected cost over several wind scenarios for the next day.
My research focused on improving and testing algorithms for creating these scenarios. I evaluated these algorithms based on stochastic unit commitment simulations of the Texas power system (ERCOT). I created data-driven models of ERCOT's thermal and wind generation. The simulations themselves were a series of large optimization problems solved on Hyak, the UW's cluster.
Along the way I also created Minpower - a Python package for power systems optimization.
I have an interest in visualization and other digital creative pursuits - from transit visualization, to portfolio design and publishing. I've cataloged some of the more recent things I've worked on here.
Intern 3Tier, Software and Science Teams
Worked as part of the software team on a web-based interactive display of regional wind forecasts. Worked closely with the science team to evaluate the performance of machine learning techniques on hour-ahead wind forecasting and to assess the value of wind forecasts to energy traders.
Data Visualization Intern Habit Labs
Created a custom analytics dashboard for a mobile health startup working on helping people start healthy habits. Created user-facing visualizations within the mobile app.
Intern Alstom Grid, Optimization Software Team
Created a Python-based tool to automatically translate an evolving database specification for Alstom Grid's energy markets software into code for their optimization system.
Master's thesis: Speech Source Detection University of Washington
Built and tested a software system to learn and decide between "live" and recorded speech. Created theory-based models of acoustic transfer functions for use in speech detection and multimedia search. Presented the results at ICASSP'10.
Bachelor's Vanderbilt University
BS Mathematics and BE Electrical Engineering
Languages and Tools
- High Level
- Source Control
- Coopr, CVX, Gurobi, CPLEX
- Pandas, SQL, Redshift, Mongo
- SciPy stack, Mathematica, MatLab
- Flask, Rails, Django