Comparative Analysis of the Efficiency of Linear ARMA Model with ANN and ANFIS Methods for Neretva River Mean Monthly Flow Rate Estimation

Abstract

Estimation of the characteristic values of water flow in rivers is extremely important both for planning and construction, as well as for protection against harmful effects of water (e.g. floods) and environmental protection of water-course. This paper analyzes and compares standardized methods in modeling mean monthly river flows using linear stationary Autoregressive–Moving-Average (ARMA) models and two machine learning methods, Artificial Neural Network (ANN) and Adaptive-Network-based Fuzzy Inference System (ANFIS) for the prediction of mean monthly river flows on the hydrologic measure station MS “Žitomislići” on the Neretva River (Bosnia and Herzegovina-BiH). The models are based on data collected over the 51-year period from 1926 to 1977. It was found that ANFIS models overcome the traditional ARMA model according to the mean relative square errors of the deviation, and the ANN model has a comparative performance with traditional ARMA models. All three models need to be tested with more recent data for different river levels in Bosnia and Herzegovina to assess the accuracy of predictions for different flows, as well as for a larger set of input parameters.