Causaal machinaal leren voor thermische gebouwsystemen: modellering, interventieanalyse en beslissingsondersteuning

Abstract

Data science tools and data-driven decision-making are being increasingly used in the operation of critical infrastructure such as the electricity and transportation grid. This is driven by the increasing flexibility needs and opportunity arising from proliferation of distributed energy resources such as solar PV, electric vehicles and batteries. However, by definition, many data science tools (e.g. machine learning models) assume a stationary environment. When the environment changes, as is the case in today's dynamic energy sector, such data-driven models may no longer hold over time. This was evidenced in the aftermath of the Covid-19 pandemic, but is also more broadly visible due to the evolving energy mix and consumer behavior. These data-driven models also often solely minimize prediction loss, and do not necessarily learn the underlying dynamics of the system. This can lead to models which generalize poorly in previously unseen conditions, which can in turn cause dangerous interactions when they are used to control the operation of distributed energy resources. This can include the charging and discharging of electric batteries and vehicles in local energy communities, as well as the operation of HVAC (heating, ventilation and air conditioning) systems. As a result, in this project, the student will focus on investigating causal inference to make the data science and data-driven decision-making toolkit more robust. The resulting methodology will be applied to real-world energy communities to optimize the operation of distributed flexible resources to guarantee their support for critical infrastructure such as the energy grid.