A Semi-Supervised Method for Smart Irrigation Using Real Data In the Mekong Delta

Hoang-Loc La, Binh Pham Quang, Nguyen Tran Tho, Trung Dang Anh, Nam Thoai

Abstract: Besides effectiveness, agriculture’s efficiency has recently become a critical criterion when deploying smart irrigation systems. Several works focused on developing irrigation scheduling methods to reduce the human interventions in agricultural farming. These methods were mainly categorized into supervised learning and reinforcement learning. Conventional supervised learning methods were basically based on farmers’ experiences or experts’ suggestions. These methods may lead to increase the maintenance costs and, consequently, production costs. Moreover, human suggestions usually tend to over-irrigate, which leads to waste of freshwater resources and speeds up soil erosion. In another way, reinforcement learning can autonomously adapt to environmental changes without human interventions. However, this approach depends on the reliability of simulators, and in a production environment, running several trial-error steps in a simulator can be time-consuming and impractical. This paper proposed a novel semi-supervised learning method to autonomously adapt to the environment’s changes. Our approach is lightweight, simple, does not rely on complex emulation and reduces the human interventions in the irrigation process.