Online Predictive Control for Daily Irrigation in Open-Field Agriculture
C.N. Goedhart (TU Delft - Mechanical Engineering)
M. Guo – Mentor (TU Delft - Team Meichen Guo)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
Agriculture accounts for roughly 72% of global freshwater withdrawals, but irrigation is often still based on intuition or fixed schedules. Model-based and data-driven approaches have shown promise, but they mostly rely on historical datasets or crop-specific knowledge that are unavailable in many settings. This thesis addresses that gap by proposing an irrigation control framework that requires no prior data and can learn the system dynamics online during the growing season. The framework combines three components: a Zone Model Predictive Control strategy, an online model estimator based on Recursive Least Squares (RLS), and a Readily Available Water (RAW) estimator. The framework is evaluated using the AquaCrop simulator on a sugarcane case study in southern Mozambique. Results show that the proposed framework performs comparably to a static controller preconfigured with historical data and crop-specific knowledge, while requiring neither. A sparse measurement strategy is also evaluated, reducing the number of required soil moisture measurements by around 65% with little impact on performance. That said, learning the system online comes with trade-offs, and the framework does not consistently outperform a well-configured static model. Together, the results suggest that effective predictive irrigation control is achievable in data-scarce environments.