|
Orji, F. N.
Michael Okpara University of Agriculture Umudike, Abia State, Nigeria
Ahaneku, I. E.
Michael Okpara University of Agriculture Umudike, Abia State, Nigeria
Awu, J. I.
National Centre for Agricultural Mechanization, Ilorin, Kwara State, Nigeria
Cyprain, N. T.
Rivers State University, Portharcourt, Rivers state, Nigeria
Ubah, J. I.
Nnamdi Azikiwe University, Awka, Anambra State, Nigeria
ABSTRACT
Hydrological modeling plays a pivotal role in understanding water distribution and movement within specific regions over time. This study focuses on the optimization of the sigmoid transfer function hyperparameter "a" in Artificial Neural Networks (ANNs) for hydrological modeling, with the aim of enhancing model performance. The study employs hydro-meteorological data sourced from Nigerian Hydro-meteorological Service Agency (NIHSA) in Abuja and Anambra-Imo River Basin Authority and collected between 1992 and 2022. A supervised backpropagation training algorithm was employed, with a sigmoid hyperparameter "a" interval of 0.2, 0.4, 0.6, 0.8 and 1. The results of the NARX-NN model shows that the R-squared values for the hyper parameter optimization values of 0.2, 0.4, 0.6, 0.8 and 1 are 0.9988, 0.9996, 0.9976 and 0.9987, respectively. The results of the sigmoid hyper parameter “a” R-squared values ranges from 0.9988 to 0.9987, with the best performance achieved at a hyper parameter “a” value of 1. The results reveal that hyper parameter optimization significantly influences model performance. These findings emphasize the importance of optimizing the sigmoid transfer function hyper parameter to enhance the accuracy and applicability of hydrological models.
Keywords: Artificial Neural Network; Transfer Function; Optimization; Modeling
https://doi.org/10.33922/j.ujet_v10i1_9
|
View: 89 | Download: 7
Published
Tuesday, April 23, 2024
Issue
Vol. 10 No. 1, June 2024
Article Section
GENERAL
The contents of the articles are the sole opinion of the author(s) and not of UJET.
|
