Energy & Behavioral Data Analytics
We develop economic, statistical and physical tool boxes that have applications to climate science, the water-energy nexus, building energy use, demand response and smart transportation. We work in close collaboration with other research groups in the Energy Technologies Area at Berkeley Lab, including the Behavior Analytics team and Computational Research Division. Our goal is to pull customer behavioral insights from granular, high frequency data using a combination of machine learning, behavioral economics and causal inference.
Measuring Demand Response
In collaboration with researchers across the Energy Technologies Area (ETA) at Berkeley Lab, we develop statistical algorithms for forecasting the potential of various sectors (residential, commercial, agriculture, manufacturing) to provide responsive and flexible load. We develop methods to estimate Demand Response (DR) services, including load regulation, shed and shifting.
A recent output of our research was an analysis of the DR potential available in California, and valuation of such services. For this research we developed a bottom-up, customer end-use load forecasting model with tight integration between weather, loads and renewable generation patterns (constituting net load) using approximately 100,000 customer smart meter energy data. The resulting forecasts were combined with a detailed DR cost database to express DR supply curves for each grid service, showing how much DR is expected to be available at a range of costs.
Analytics for Behavioral Insights and Decision Science
We work in close collaboration with the Behavior Analytics team (Annika Todd and Anna Spurlock in the Electricity Markets & Policy (EMP) Group, and the Computational Research Division (CRD) (John Wu and Alex Sim) with a goal to pull customer behavioral insights from granular, high frequency data using a combination of machine learning, behavioral economics and causal inference.
Recent and ongoing collaborations with the Behavior Analytics and CRD team have resulted in several proof-of-concept real world applications highlighting data-driven analytics and ground truth verification through causal inference, including:
- A novel application of machine learning algorithms to derive meaningful household segmentation for time-based rate pilots that relate not only to program enrollment but also actual realized peak period energy reductions;
- A review and quantitative evaluation of machine learning methods applied to extracting residential customer energy attributes;
- A derived library of archetypal 24 hour load patterns from over 30 million residential daily load shapes;
- Technical support to novel program evaluation methods;
- Algorithms to develop and ground-truth detection of household air conditioner ownership and usage.
- Identification and application of innovative analytics and visualization on the survey responses to develop new insights into transportation behavior.