FROM BASELINE TO FUTURE SCENARIOS: EVALUATING GRIDSEER TECHNOLOGY AND AVOIDED EMISSIONS AT DUKE UNIVERSITY

Abstract

This project evaluates the carbon reduction potential of machine learning–driven energy optimization at Duke University, with a focus on Scope 2 electricity emissions. Using 30-minute electricity consumption data and marginal emissions factors (MOER), the analysis compares observed demand with GridSeer feasible cases optimized for cost under current operating conditions. Net avoided emissions are calculated using a marginal emissions framework and decomposed into volume and timing effects to distinguish reductions from lower total electricity use versus load shifting across time. Results show a modest reduction in electricity use of about 1.26% and net avoided emissions of about 3,350 metric tons of CO2. However, the carbon benefit is not consistent across time, with 45.5% of intervals showing emissions above baseline under cost-based optimization. Most avoided emissions come from reduced total electricity consumption rather than from load shifting. Additional storage improves the carbon outcome under the analyzed scenarios. Overall, the results indicate a modest but positive emissions benefit under cost-based optimization.