Clemson University
Application of Artificial Neural Network Algorithms for Irrigation Scheduling
Abstract
dc:description.abstractNeural networks have been extensively used in predicting soil water tension for improved irrigation scheduling and management. However, their lack of interpretability constrains their efficacy in grasping the nuanced patterns prevalent in soil water tension time series data. The first goal of this research was to develop interpretable deep neural network models for soil water tension prediction across multiple soil depths (0.15m, 0.3m, 0.46m and 0.6m) and prediction horizons (1h, 6h, and 12h). The Neural Hierarchical Interpolation for Time Series (N-HiTS) and Neural Basis Expansion Analysis Time Series (N-BEATS) models were used in this research. Historical soil water tension data collected at the University of Georgia’s C. M. Stripling Irrigation Research Park (SIRP) and in Blackville, South Carolina, were used to train and test the models. The results were benchmarked with the Long-Short-Term Memory (LSTM) to compare the models with a recurrent soil water tension prediction method. All the algorithms were coupled with a probabilistic multi-quantile loss function to quantify the uncertainty associated with predictions. Analysis suggested that the N-HiTS and N-BEATS models outperformed the LSTM model in the two testbeds by maintaining accuracy over the extended horizons and depths. The prediction uncertainty was more controlled for N-HiTS and N-BEATS, with narrower uncertainty bands across horizons and soil depths, while LSTM exhibited widening intervals, reflecting reduced predictive confidence. We demonstrated how the proposed architecture can be augmented with uncertainty quantification to provide probabilistic soil water tension predictions that are interpretable without considerable loss in accuracy. Accurate and interpretable soil water tension predictions can reduce water use, provide timely irrigation, and increase crop yield. The second goal of this study was to apply deep reinforcement learning algorithms namely proximal policy optimization (PPO) and deep Q-network (DQN) to generate irrigation schedules of an irrigated field located at the University of Georgia’s C.M SIRP. Results indicated that PPO outperformed DQN by converging faster and maintaining stable rewards across the training steps. Crop water use efficiency also indicated that PPO provided a superior performance, highlighting its potential in optimizing both yield and water use. Experiments on deterministic and probabilistic policy decisions demonstrated that the DRL techniques can efficiently learn strategic policies for irrigation optimization in environments with a discrete or continuous action space.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy (PhD)
- Level thesis:degree_level
- Dissertation
- Discipline thesis:degree_discipline
- Agricultural and Applied Economics
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Umutoni, Lisa
- Contributors dc:contributor
-
- Vidya Samadi
- George Vellidis
- Charles Privette III
- Jose Payero
- Bulent Koc
Subjects
dc:subject × 7Identifiers
dc:identifier.*- Repository record dc:identifier
- https://open.clemson.edu/all_dissertations/4071