Application of the relevance vector machine to canal flow prediction in the Sevier River Basin

https://doi.org/10.1016/j.agwat.2009.09.010Get rights and content

Abstract

This work addresses management of water for irrigation in arid regions where significant delays between the time of order and the time of delivery present major difficulties. Motivated by improvements to water management that will be facilitated by an ability to predict water demand, it employs a data-driven approach to developing canal flow prediction models using the relevance vector machine (RVM), a probabilistic kernel-based learning machine. A search is performed across model attributes including input set, kernel scale parameter and model update scheme for models providing superior prediction capability using the RVM. Models are developed for two canals in the Sevier River Basin of southern Utah for prediction horizons of up to 5 days.

Section snippets

Introduction and background

One of the biggest challenges in areas with limited water is getting the necessary amounts of water to the desired places at the appropriate times, with the ever-present objective of providing the water with minimal loss in transmission and minimal excess. Meeting this challenge is problematic when, as is often the case, the amounts of water needed, the locations and times of need, and the losses that will occur are not precisely known at the time when water management and diversion decisions

Predictive function estimation and the relevance vector machine

In this section we very briefly introduce the relevance vector machine to establish some notation and concepts for following discourse. A considerably more complete discussion and derivation of the results appears in Flake (2007); the relevance vector machine appears initially in Tipping (2001).

Prediction is the deduction or estimation of a system condition based on some functional or intuitive relationship between that condition and other conditions in the system. The task of machine learning

Application to canal flows in the Sevier River Basin

For initial assessment of potential model inputs and in much of our experimentation we used flow data from the Richfield Canal, one of the largest canals in the Sevier River Basin. An example of hourly flow measurements taken at the head of Richfield Canal is shown in Fig. 1. The figure shows significant periods of flow during the growing season separated by short periods of zero or near-zero flow. This interrupted flow pattern is indicative of the water requirements of the major crops

Summary and conclusions

We have established a method for predicting canal flow for canals in the Sevier River Basin.

Through our experiments it has become apparent that the water use of farmers may not reflect well the crop water need, especially with the limited serviceability of the evapotranspiration input. It is disputable that the prediction models formed are more evidence of the RVM learning patterns of farmer and canal operator behavior than of a system with physical input–output implications. In this way,

References (9)

  • Allen, R.G., Pereira, L.S., Raes, D., Smith, M., 1998. Crop evapotranspiration—guidelines for computing crop water...
  • B. Berger et al.

    Developing a virtual watershed: The Sevier river basin

  • Berger, B., Hansen, R., Jensen, R., 2003. Sevier river basin system description. Tech. rep., Sevier River Water Users...
  • Flake, J.T., 2007. Application of the relevance vector machine to canal flow prediction in the Sevier river basin....
There are more references available in the full text version of this article.

Cited by (24)

  • An integrated method based on relevance vector machine for short-term load forecasting

    2020, European Journal of Operational Research
    Citation Excerpt :

    Based on a probabilistic Bayesian learning framework, relevance vector machine (RVM) can produce highly sparse models to overcome the overfitting problem triggered by SVM methods when dealing with noisy data (Tipping, 2001). So far sparse Bayesian learning and RVM approach has been successfully applied to many fields such as machining tool control (Yuan et al., 2019; Yuan, Zhang, Wu, Zhu, & Ding, 2017), water demand forecast (Flake, Moon, McKee, & Gunther, 2010), and wind power prediction (Yan, Liu, Han, & Qiu, 2013). However, only a limited number of studies have focused on employing RVM for load forecasting.

  • Wind power prediction using a three stage genetic ensemble and auxiliary predictor

    2020, Applied Soft Computing Journal
    Citation Excerpt :

    A linear hyper plane is then fixed on the data in higher dimension which is then used to predict the target values for test inputs. Every target value consists of kernel function on input with presence of noise [40], which helps of training set overfitting. Natural selection method by Koza [42] is known as Genetic Programming which is used to automatically solve the computer problems.

  • A hybrid model of EMD and multiple-kernel RVR algorithm for wind speed prediction

    2016, International Journal of Electrical Power and Energy Systems
  • Wind speed prediction using the hybrid model of wavelet decomposition and artificial bee colony algorithm-based relevance vector machine

    2015, International Journal of Electrical Power and Energy Systems
    Citation Excerpt :

    Sancho et al. [11] presented an evolutionary-SVM algorithm for short-term wind speed prediction, and applied the evolutionary-SVM algorithm to wind speed prediction in wind turbines of a Spanish wind farm. Relevance vector machine (RVM) is an intelligent learning technique based on sparse Bayesian framework, the number of relevance vectors in RVM is much smaller than that of support vectors in SVM, which makes RVM have a sparser representation compared with SVM [12,13]. Moreover, there is no need to set the penalty parameter in RVM, which makes RVM more convenient to use than SVM.

  • Modeling tunneling-induced ground surface settlement development using a wavelet smooth relevance vector machine

    2013, Computers and Geotechnics
    Citation Excerpt :

    Because of these advantages, the results obtained from an RVM are often superior to those obtained from an SVM for the same inputs [12,13]. RVMs have been successfully applied to fault diagnosis [14], canal flow prediction [15] and monitoring network analysis [16]. Smooth relevance vector machines (sRVMs) are an extension of RVMs, to some extent [17].

View all citing articles on Scopus
View full text