Accurate predictive modelling of electricity demand and generation will become increasingly important in a renewable dominated power system. In case of mis-forecasts, similar models can also be used to help activate demand-side flexibility and provide essential grid services ranging from congestion and voltage management to frequency control.

However, these models are not readily available for most flexible assets, due to the prohibitive costs of requiring domain experts to build them.

The rise of smart meters means such models can increasingly be learnt using machine learning algorithms. However, these models target predictive accuracy rather than learning correct causal relationships, which means they cannot be relied upon for downstream decision-making, even when post-hoc explainability tools are used.

In this talk, Hussain will discuss some of these issues in greater detail, and show how combining domain knowledge with observational data can lead to models that are not just accurate but also causally consistent with the help of several case studies.

Speaker

Hussain Kazmi is currently an assistant professor at KU Leuven, focusing on data science and decision support tools for the energy transition. He holds a PhD in electrical engineering from KU Leuven, MSc degrees in sustainable energy technology from Technical University of Eindhoven and Politecnico di Torino, and a B.E. in electrical engineering from National University of Sciences and Technology Pakistan. In the past, he has been a visiting researcher at Imperial College London, National University of Singapore, and KTH Royal Institute of Technology Sweden. He has also played key roles in several clean energy start-ups, and was granted the International Institute of Forecasters’ Annual Award for his work on value-oriented forecasting in 2021. Currently, in the context of an EIT InnoEnergy project, he coordinates a cross-European Working Group on Energy Data Science.