A data-driven approach to soil moisture collection and prediction using a wireless sensor network and machine learning techniques

soil moisture;data driven;wireless sensor network;machine learning

Hong, Zhihao ; Iyer ; Ravishankar K. ; Kalbarczyk ; Zbigniew T.

关键词: soil moisture;    data driven;    wireless sensor network;    machine learning;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/88145/HONG-THESIS-2015.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Agriculture has been one of the most underinvestigated areas in technology, and the development of Precision Agriculture is still in its early stages. This thesis proposes a data-driven methodology that aims to address some of the current problems in Precision Agriculture development. Soil moisture, a key factor in the crop growth cycle, is selected as an example to demonstrate the effectiveness of our data-driven approach. The success of the data-driven approach depends on two factors: (1) the quality of the data gathered and (2) the effectiveness of its analysis and interpretation. Previous studies have focused on addressing these factors separately,by either developing hardware for collecting soil moisture data or building efficient data analysis models.In our work, we take a holistic approach by addressing problems on both ends and designing an integrated system for Precision Agriculture that uses a wireless sensor network and machine learning techniques. On the collection side, a reactive wireless sensor node is developed that aims to capture the dynamics of soil moisture while sampling at relatively low frequencyto save energy. The sensor node dynamically adjusts its sampling frequency based on soil moisture readings and can be easily configured to meet the specific needsapplications. The hardware is prototyped using MicaZ mote and VH400 soil moisture sensor. On the data analysis side, a site-specific soil moisture prediction framework is proposed based on models generated by the statistically sound machine learning techniques SVM (support vector machine) and RVM (relevance vector machine). The framework can integrate inputs from other reliable data sources to improve its accuracy. The proposed framework is evaluated under a historical dataset on 9 sites across Illinois. Itachieves low error rates (15%) and high correlations (95%) between predicted values and actual values when forecasting soil moisture about 2 weeks ahead.