Real-time forecasting of short-term irrigation canal demands using a robust multivariate Bayesian learning model
In the lower Sevier River basin in Utah, the travel times between reservoir releases and arrival at irrigation canal diversions limit the reservoir operation in enabling delivery changes, which may not be compatible with the on demand schedule in the basin. This research presents a robust machine learning approach to forecast the short-term diversion demands for three irrigation canals. These real-time predictions can assist the operator to react promptly to short-term changes in demand and to properly release water from the reservoir. The models are developed in the form of a multivariate relevance vector machine (MVRVM) that is based on a Bayesian learning machine approach for regression. Predictive confidence intervals can also be obtained from the model with this Bayesian approach. Test results show that the MVRVM learns the input–output patterns with good accuracy. A bootstrap analysis is used to evaluate robustness of model parameter estimation. The MVRVM is compared in terms of performance and robustness with an Artificial Neural Network.
Ticlavilca, A. M., M. McKee and W. R. Walker. 2013, Real-time forecasting of short-term irrigation canal demands using a robust multivariate Bayesian learning model. Irrigation Science. 31(2):151–167. DOI 10.1007/s00271-011-0300-6.