Analysis and Prediction of Dammed Water Level in a Hydropower Reservoir Using Machine Learning and Persistence-Based Techniques

C Castillo-Botón,B Morales-Díaz,J Sanz-Justo,L M Moreno-Saavedra,D Casillas-Pérez,S Salcedo-Sanz,P A Gutiérrez,C Casanova-Mateo

doi:10.3390/w12061528

C Castillo-Botón, B Morales-Díaz + Show 6 more

Open Access

https://doi.org/10.3390/w12061528

Copy DOI

Journal: Water	Publication Date: May 27, 2020
Citations: 29	License type: CC BY 4.0

Affiliation: University of Alcalá, University of Valladolid

Abstract

This paper presents long- and short-term analyses and predictions of dammed water level in a hydropower reservoir. The long-term analysis was carried out by using techniques such as detrended fluctuation analysis, auto-regressive models, and persistence-based algorithms. On the other hand, the short-term analysis of the dammed water level in the hydropower reservoir was modeled as a prediction problem, where machine learning regression techniques were studied. A set of models, including different types of neural networks, Support Vector regression, or Gaussian processes was tested. Real data from a hydropower reservoir located in Galicia, Spain, qwew considered, together with predictive variables from upstream measuring stations. We show that the techniques presented in this paper offer an excellent tool for the long- and short-term analysis and prediction of dammed water level in reservoirs for hydropower purposes, especially important for the management of water resources in areas with hydrology stress, such as Spain.

Highlights

The use of river water resources by means of reservoirs and dams is of primary importance for energy generation, water supply, navigation, and flood control, among others [1]
We evaluated the performance of a number of the Machine Learning (ML) regressors such as multi-layer perceptrons, support vector regression, extreme learning machines, or Gaussian processes in this problem of short-term prediction
This experimental evaluation is closed with the short-term results of dammed water level, focused on a weekly time-horizon prediction problem, and how ML algorithms perform on this problem

Summary

Introduction

The use of river water resources by means of reservoirs and dams is of primary importance for energy generation, water supply, navigation, and flood control, among others [1]. The management of water reservoirs in rivers is a critical problem, including many possible tasks that depend on the specific aim of the dammed reservoir. Among these tasks, the prediction of the dammed water level in the reservoir is critical in many cases, for evaluating structural problems in dams [3], water supply and resource availability [4,5], water quality [6,7,8], bio-diversity conservation [9], navigation management [10], disaster prevention [11], and hydropower production optimization, which is the problem we are interested in this paper [12,13]. Many authors [14,15] have considered hydro-meteorological data, but alternative input data for prediction are available, such as images from video cameras [16]

Results

Discussion

Conclusion