Assessment of Machine Learning Techniques for Monthly Flow Prediction

Zahra Alizadeh,Jafar Yazdi,Abobakr Al-Shamiri,Joong Kim

doi:10.3390/w10111676

Zahra Alizadeh, Jafar Yazdi + Show 2 more

Open Access

https://doi.org/10.3390/w10111676

Copy DOI

Journal: Water	Publication Date: Nov 17, 2018
Citations: 24	License type: CC BY 4.0

Affiliation: Shahid Beheshti University, Korea University

Abstract

Monthly flow predictions provide an essential basis for efficient decision-making regarding water resource allocation. In this paper, the performance of different popular data-driven models for monthly flow prediction is assessed to detect the appropriate model. The considered methods include feedforward neural networks (FFNNs), time delay neural networks (TDNNs), radial basis neural networks (RBFNNs), recurrent neural network (RNN), a grasshopper optimization algorithm (GOA)-based support vector machine (SVM) and K-nearest neighbors (KNN) model. For this purpose, the performance of each model is evaluated in terms of several residual metrics using a monthly flow time series for two real case studies with different flow regimes. The results show that the KNN outperforms the different neural network configurations for the first case study, whereas RBFNN model has better performance for the second case study in terms of the correlation coefficient. According to the accuracy of the results, in the first case study with more input features, the KNN model is recommended for short-term predictions and for the second case with a smaller number of input features, but more training observations, the RBFNN model is suitable.

Highlights

Future river discharge predictions have been widely used for flood control, drought protection, reservoir management, and water allocation
Five other machine learning algorithms, i.e., feedforward neural network, time delay neural network, radial basis neural network, recurrent neural network, and K-nearest neighbors (KNN) were established to verify the capability of these algorithms for discharge prediction
Through a comparison of the results based on the value of the coefficient of correlation, it was found that the KNN model, without any training process in which a part of noisy data is used to train the model, and RBFNN model can provide more accurate predictions for Alavian and Dez Basins, respectively

Summary

Introduction

Future river discharge predictions have been widely used for flood control, drought protection, reservoir management, and water allocation. Owing to a lack of adequate knowledge regarding the physical processes in the hydrologic cycle, traditional statistical models, such as the auto regressive moving average (ARMA) and auto regressive integrated moving average (ARIMA) models [1], have been developed to predict and generate synthetic data. Such models do not attempt to represent the nonlinear dynamics inherent to the hydrological process, and may not always perform well [2]. ANN models suffer from overfitting or overtraining, which decreases the capability of the prediction for data far from the training samples

Methods

Results

Conclusion