| dc.description.abstract | State-of-the-art mechanically driven model of the anaerobic digestion process was
compared to data driven methodology of System Identification and Machine Learning.
Through collaboration with OsloMet, Veas WWTP and DHI was data collected for analysis and to compose a dataset that DHIs West software could use
to run their mechanically driven simulation based on the ADM1 model.
Live data from the same plant was also collected to create a dataset where two
data driven models were created; one using System Identification methodology
and the other using Machine Learning. Both data driven models were attempted
for closed-loop integration with controllers.
DHIs model output tracked the actual output, showing similar behaviour to
the actual plant. The data driven models showed less stable behaviour, although
at times closer tracking to actual output than in West. Closed-loop integration
with PI-controller and MPC on the SI-model showed close setpoint tracking with
predictive inputs to the MV. Controller implementation on the ML-model showed
less stable behaviour and further development of controller tuning is needed.
Keywords: Renewable Energy, Wastewater Treatment, Biogas Production, Anaerobic Digestion, Veas, DHI, West, Simulation, Control,
Machine Learning, Recurrent Neural Network, Long-Short Term Memory, System Identificatio | |
