Energy Efficiency Monitoring and Root Cause Analysis

Thilo Weber
1. Okt.
1 Min. Lesezeit

Description:

In collaboration with HSLU and geoimpact, I developed a machine learning framework that estimates the real-world impact of building retrofitting measures — such as insulation and window replacement — on heat consumption. By combining predictive modeling with causal inference, the system quantifies the true energy savings achieved across diverse building types and regions. This scalable, data-driven approach enables policymakers and energy planners to track retrofit effectiveness, refine subsidy programs, and accelerate progress toward climate and efficiency targets.

Methods:

Heterogeneous treatment effects & causal inference: Heterogeneous treatment effects refer to the variation in the impact of a treatment across individuals or subgroups, and causal inference aims to estimate and understand these effects by identifying the causal relationship between interventions and outcomes.
Partial dependence: The plots below show the partial dependence of the heat consumption indicator (HCI) on construction year and heating degree days.
Combining domain knowledge & ML: In the plots below, the lines of the linear model and the neural network, which both integrate domain knowledge in the model architecture, show more realistic dependencies than the gradient boosting (without domain knowledge).

Specials:

Apart from efficiency monitoring, such a framework can be used for a general root cause analysis of physical processes. For example, I introduced the framework to a friend who is working for a big chemical industry company. He quickly gained a lot of valuable insights into their production processes from it. He is since known as “the data leech” at his company.