Demand Responsive Grid Emissions Metrics Incorporating Renewable Intermittency and Scalable Space Heating Load Modeling using Physics-Informed Neural Networks

Abstract

Electrification is fundamental to decarbonization; however, its climate benefit is contingent upon the marginal generation source and the efficiency of the replacement technology. Conventional Average and Marginal Emissions Factors (AEF, MEF) do not adequately capture these effects. This thesis introduces three complementary tools to address this gap: the Extended Marginal Emissions Metric with Renewables (XMEF-R) metric, its technology-aware extension, the Comparative Emissions Factor (CEF), and Thermostat Hysteresis Embedded RC Model ODE (THERMO), a physics-informed deep-learning framework for building heating prediction.

CEF is a technology-sensitive metric that quantifies the emissions consequence of replacing a specific end-use with an electric alternative by explicitly linking outcomes to the marginal fuel and device efficiency. XMEF-R refines marginal-fuel identification by accounting for renewable intermittency and thermal generator dynamics, enabling real-time, regionally specific emissions assessments. Applying these methods shows that electrification is not intrinsically low-carbon: light-duty vehicle electrification typically reduces emissions, whereas space-heating yields reliable reductions only when paired with high-efficiency heat pumps; conversely, systems with useful secondary outputs (e.g., CHP) can achieve emissions reductions even with carbon based fuels if waste heat is utilized. These results support policy priorities, such as accelerating coal-to-gas transitions where necessary, scaling up renewables, and sequencing equipment replacement with grid decarbonization.

THERMO couples Neural Ordinary Differential Equations and a reduced-order resistance-capacitance (RC) thermal model within a Physics-Informed Neural Network to infer envelope and thermostat parameters from minimal metadata, producing physics constrained space-heating load forecasts and RC parameters for future studies at low computational cost.

Together, CEF/XMEF-R and THERMO create a scalable, data-driven framework that links building-level demand to grid emissions, enabling robust policy evaluation, real-time predictions, and strategic pathways for electrification.