Work

Adaptive ARX pressure modeling for smart variable-rate agricultural sprayers

Year: 2026

Type: article

Abstract: Accurate pressure prediction is required for model-based control of variable-rate agricultural sprayers, whose dynamics vary with pump commands, valve openings, and changes in nozzle activation. This work investigates hydraulic-regime-dependent AutoRegressive with eXogenous input (ARX) pressure models for a Smart Weed Sprayer and compares fixed, scheduled, feedback-corrected, and adaptive structures for control-oriented prediction. Eighty-four experiments were collected, of which 79 valid operating conditions were retained after data-quality and hydraulic-resistance screening. Local ARX models provide accurate in-sample fits but exhibit limited portability across hydraulic regimes, while hydraulic resistance alone was insufficient for reliable direct model scheduling. Validation is performed using repeated grouped five-fold cross-validation, with complete experiments assigned to either the training or test sets. Under adaptive output-feedback prediction, a fixed model with causal output-bias correction reduced mean RMSE from 6.709 to 0.810 psi, while scalar-gain RLS achieved 1.423 psi using only one adaptive plant parameter. After a 20% calibration interval followed by measurement-independent recursive prediction, scalar-gain RLS achieved an RMSE of 1.185 psi, compared with 6.309 psi for the fixed model and 1.475 psi for the bias-corrected model. Identifiability analysis showed that the available excitation weakly supports independent adaptation of the two numerator coefficients, supporting scalar-gain adaptation as a parsimonious alternative for future predictive-control implementation.

Source: Smart Agricultural Technology

DOI: 10.1016/j.atech.2026.102675

Authors: Deniver Reinke Schutz, Antonio O. Neto, Yiannis Ampatzidis, Vilma A. Oliveira, Paulo E. Cruvinel