Bachelor Thesis - Data-Driven Building Modelling
A study of physics-based and data-driven approaches to thermal building modelling and short-horizon forecasting.
Overview
Bachelor thesis at MCI Innsbruck on thermal building modelling, comparing two fundamentally different ways of approaching short-horizon indoor temperature forecasting: physics-based grey-box modelling and data-driven neural time-series models.
Approach
Physics-based modelling
Grey-box RC (resistance-capacitance) network models, describing a building’s thermal behaviour through resistances, capacitances and solar gain, with model parameters fitted numerically.
Data-driven modelling
Neural time-series architectures including RNN and LSTM models, implemented and evaluated in Python with TensorFlow/Keras, with attention to how much temporal context a model needs to be useful.
What it built
Hands-on experience across thermal modelling, time-series forecasting, comparative model evaluation and scientific reporting - including the discipline of designing an evaluation that can distinguish between methods rather than flattering one of them.
A note on detail
The thesis is under an academic access restriction, so this page stays at topic and method level.