CAUSAL INFERENCE METHODS FOR ELECTRICAL CONSUMPTION’S ESTIMATION

The thesis was lead in Électricité de France’s (EDF) Lab, a leader in the energy sector, particularly in energy efficiency initiatives such as thermal renovations. EDF’s commitment to sustainable energy management and the need for accurate evaluation of its initiatives sets the stage for this research. The study delves into the realm of causal inference, a crucial statistical tool for understanding the impacts of interventions in situations where controlled experiments are not feasible. It particularly addresses the challenge of evaluating the real effect of thermal renovations on electric consumption while navigating potential biases in EDF’s extensive dataset. The thesis details the methodologies employed, which include advanced causal inference techniques and machine learning models such as Two-Way Fixed Effects, Rocket, Matching, Propensity Score, Synthetic Control, T-Learners, R-Learners, and Causal Forests. These methods are applied to simulated data, meticulously modeling the effects of thermal renovations on electricity consumption and addressing issues like unobserved confounding and selection bias.

Author dc:creator

• ZAYENI LANGROUDI ROXANA

• Causal Inference, Machine Learning Models, Synthetic Control, Two-Way Fixed Effect, Feature Selection, Propensity Score