Open AccessComparative study of neural network based and white box model predictive control for a room temperature control application
Author(s) -
Phillip Stoffel,
Max Berktold,
Arman Gall,
Alexander Kiimpel,
Dirk Müller
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/2042/1/012043
Subject(s) - model predictive control , controller (irrigation) , computer science , black box , white box , artificial neural network , energy consumption , process (computing) , quality (philosophy) , control (management) , control engineering , control theory (sociology) , artificial intelligence , machine learning , engineering , philosophy , electrical engineering , epistemology , agronomy , biology , operating system
On a global scale, buildings are a major cause for primary energy consumption. Since buildings are complex multiple input multiple output systems and characterized by slow dynamics, model predictive control is a promising approach to reduce building energy consumption. Due to the high individual modeling effort model predictive control lacks practical applicability. For that reason black box process models are gaining more and more interest in scientific literature. In this work we evaluate the performance of an ANN based controller against a white box controller with perfect knowledge. We show that the data driven controller achieves a similar control quality as the white box controller. We initially train the data driven controller in 20 days and then employ an online learning strategy to continuously improve the control quality.