Double Deep Q-Learning Based Irrigation and Chemigation Control

Jianfeng Song1, Dana Porter1, Jiang Hu1, Thomas Marek2
1Texas A&M University, 2Texas A&M AgriLife Research and Extension


As crop production has become more mechanized and complex, and as sensors and data have become more accessible, limitations of production managers to effectively use the data and the need for automated integration of information into useful management decisions with automated controls have become more apparent. In this article, a double deep Q-learning technique based module for irrigation and chemigation control is proposed and evaluated. This module is designed to maximize net return at harvest by automatically managing the irrigation and chemigation scheduling processes during the crop growing season. Using this approach, the proposed module can automatically select the optimal or near-optimal irrigation and chemigation amount and application schedule. The proposed module is evaluated on various crops, climate conditions, and soil types. The results show that the proposed module can achieve an average of 50% higher net return compared to traditional strategies.