Dipl.-Ing. Ole Ziessler joined the EASD team in 2017. Still a student at the time, he developed and completed both his project thesis, “Model-Based Testing of Dispatch Planning Methods in District Heating Networks,” and his diploma thesis, “Application of Machine Learning Methods on Building Monitoring Data.” Both addressed very engaging topics. We were therefore particularly pleased when the Verein zur Förderung der Ingenieurausbildung der Gebäude- und Energietechnik Dresden e. V. awarded Mr. Ziessler’s diploma thesis second prize for 2019. The 2019 awards were presented at the eighth specialist symposium on January 16, 2020.
Overview of the general toolchain for system analysis
“Application of Machine Learning Methods on Building Monitoring Data” – A Brief Overview
Ecological, political, and technological developments are leading to increasingly extensive data collection in building energy systems. New data-driven methods for analyzing and optimizing these technical systems are therefore gaining importance. This thesis investigates the relevance and applicability of machine learning (ML) in addressing application-specific challenges, using monitoring data from a modern building energy system as a case study. An ML-based toolchain was developed around three general research objectives: learning sensor associations to understand the system, data validation, and specific system analysis. Individual ML pipelines using the general pipeline architecture were developed with association learning (AL), classification, and regression algorithms. Evaluation showed that AL was applicable only to a limited extent, while classification- and regression-based pipelines performed well and fulfilled their data validation and system analysis objectives. Random Forest achieved the best results because it could learn the underlying system dynamics. It therefore has the potential to reduce manual effort in system analysis. The results show that ML can be applied to building monitoring data and that the methods used offer automatable, data-based solutions to interdisciplinary challenges.