Majority-Voting Ensemble Learning for Autonomous Multi-Actuator Control in Solar Greenhouses Under Continental Climate Conditions
Year: 2026
Type: article
Abstract: Solar greenhouses in continental climates require simultaneous regulation of heating, ventilation, shading, and irrigation to sustain optimal crop microclimate across high-amplitude diurnal and seasonal cycles. Rule-based and single-classifier control strategies are prone to decision errors when boundary conditions shift with sensor noise, actuator lag, or seasonal drift. This study proposes a majority-voting ensemble framework that aggregates binary actuation decisions from five heterogeneous classifiers—Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Gradient Boosting (GB), and Decision Tree (DT)—into a single robust actuation signal for each of four actuator channels simultaneously. The system was deployed on a low-cost ESP32–Firebase stack in a 54 m² experimental solar greenhouse at Karshi State Technical University and evaluated over a 60-day hybrid cucumber growing cycle. The voting ensemble achieved an overall classification accuracy of 94.2 % and a macro-averaged F1-score of 0.94 across all actuator channels, outperforming the best single classifier (GB, 89.1 %) by 5.1 percentage points. Total energy consumption fell by 21.4 % relative to a rule-based baseline, with a concurrent 2.8 % increase in marketable yield. The results confirm that majority voting is a computationally inexpensive yet highly robust coordination strategy for autonomous multi-actuator greenhouse control under the unpredictable disturbance patterns characteristic of Uzbekistan's continental climate.
Authors: Islomnur Ibragimov, Khudaynazarov Utkir
Institution: Karshi State Technical University
Language: English
Subfield: Plant Science
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7165664071
