Fifth meeting on SciML
Our fifth meeting will be in room ‘Gaia 1’ of the Gaia-building, on the 8th of June from 9-12 am. We will have interesting presentations by Sjoerd Boersma and by Fakirah Rifanti Maulana.
Abstract of presentation Sjoerd Boersma
Crops can develop under controlled conditions in a greenhouse. The control objective is to minimize energy use per kilogram yield (efficiency) by controlling the greenhouse’s climate. Nonlinear model predictive control (NMPC) is a paradigm that can be used for such an objective. For this, a greenhouse and crop model that is suitable for including in an optimization is required. These models are rare and mostly not well validated due to a lack of available data. There are however more precise and complex greenhouse and crop simulators available. These are not suitable for being included in an optimization though can generate relatively realistic data via scenario studies. In this preliminary study, such a simulator is used to generate data. A dynamical nonlinear neural network is fitted on this data and the validated model is consequently used to optimize the 1-day ahead future control signals that maximize the greenhouse’s 1-day ahead future efficiency.
Abstract of presentation Fakirah Rifanti Maulana
Greenhouse climate control hinges upon an accurate, concise, and easily interpretable mathematical model. However, the intricate nonlinearities inherent in greenhouse-crop dynamic systems pose significant challenges when designing an effective model-based controller. In this study, our objective is to explore advanced machine-learning approaches for identifying the dynamic model of a lettuce greenhouse. The presented approach is built upon the so-called Parallel-Implicit Sparse Identification of Nonlinear Dynamics (SINDy-PI) method, which leverages sparse regression to extract the most parsimonious governing physical equations from data.
