Estimation of the Basin Outflow by Wavelet Neural Network, Conjunctive Use of Wavelet Analysis and Artificial Neural Network: Fid-Bau Portal

Estimation of the Basin Outflow by Wavelet Neural Network, Conjunctive Use of Wavelet Analysis and Artificial Neural Network

Naderirad, Iman / Saadat, Mohsen / Avokh, Avid / Mehrparvar, Milad

Abstract

Estimation of the basin outflow by machine learning (ML) methods is increasingly attracted attention in recent years. Some subsets of ML methods considered in this paper include the artificial neural networks (ANN), conjunctive use of wavelet analysis and ANN (W-ANN) and wavelet neural network (WNN) as a hybridization of ANN and wavelet analysis. To improve the performance of these methods, one, two and three-month delays in rainfall records as well as the maximum and minimum of the air temperatures are employed as input data. The efficiencies of the methods are compared based on the Pearson correlation coefficient (R) according to the test data. In order to evaluate the methods more accurately, 20 runs were performed in the test phase. The average and standard deviation of the obtained R within these runs were taken into account. As a novelty incorporated in this paper, implementation, simplicity and accuracy of WNN are considered and compared to other methods. WNNT1, WNNT2, WNNT3, ANNT1, ANNT2, ANNT3, W-ANNT1, W-ANNT2 and W-ANNT3 represent methods that use 1-, 2-, and 3-month delays in input data to model outflow. According to the results, the R value of WNNT is obtained 0.928 with a three-month delay used as input data. The W-ANN with Daubechies mother wavelets (db4), three-month delay input and level two of resolution (W-ANNT3-db42) has the R value equal to 0.939. It is concluded that W-ANNT-db42 has approximately the same R value with WNNT3. However, the WNN method does not need a trial-and-error process to choose the mother's wavelet function and the resolution level; therefore, it is recommended as the best method to model the basin outflow.