Vol. 12, Issue 3 (2023)

Author(s):
Avanish Yadav, MA Alam and Shakti Suryavanshi

Abstract:
A daily rainfall runoff time-series estimate is crucial to water resource planning and development. The aim of this research to comparatively examine the applicability of Artificial Intelligence (AI) methods (i.e., multilayer perceptron (MLP), Support Vector Machine (SVM), multilayer perceptrons neural network coupled with genetic algorithms (MLP-GA) and the Support Vector Machine coupled with genetic algorithms (SVM-GA) model for estimating and simulating the daily runoff discharge. The models investigated at Jondhara stations, Seonath stream in Bilaspur district, Chhattisgarh, India. Three performance criteria, including correlation coefficient (CC), root mean square error (RMSE) and percent bias (PBIAS), were applied for model performance assessment. Based on the result findings, SVM-GA algorithms showed superior performance to other models for stations with low correlation coefficients (CC), RMSE and PBIAS (CC = 0.9751, RMSE = 0.3230, and PBAIS = -0.0080 in training data set and CC = 0.9575, = RMSE = 0.3709 and PBAIS = -0.0117 in testing data set in Q-4 model and CC = 0.9743, RMSE = 0.3354 and PBAIS = -0.0084 for training data set and CC = 0.9733, RMSE = 0.3390, and PBAIS = -0.0104 for testing data det in Q-16 model). As compared to statistical criteria all methods, specially SVR-GA performed exceptionally well. The study indicated that SVR-GA could handle and simulate daily runoff based on limited information.