Comparison of linear and non-linear methods for runoff prediction using, co-active neuro-fuzzy inference system (CANFIS) and multi linear regression (MLR) technique for Narmada river basin, Gujarat

Abstract

Runoff prediction is one of the most important and challenging task in the modern world. In this study, we attempted to forecast the daily runoff on the basis of Coactive Neuro-Fuzzy Inference System (CANFIS) and Multi Linear Regression (MLR) techniques. The performance of the developed models, on the basis of training and testing, was judged on the basis of four statistical measures such as Root Mean Squared Error (RMSE), Coefficient of Efficiency (CE), Correlation Coefficient (r) and Coefficient of Determination (R2) during monsoon period (June to September) for Chhota Udaipur area in Gujarat, India. The daily data of rainfall, minimum temperature, maximum temperature and wind speed were used for runoff prediction. The appropriate parameter combination of input variables for CANFIS was used to predict runoff. The Neuro Solution 5.0 software and Microsoft Excel were used in analysis and the performance evaluation of developed models, respectively. The architecture of CANFIS was designed with Gaussian membership function, Takagi-Sugeno-Kang fuzzy model, hyperbolic tangent activation function and Delta-Bar-Delta learning algorithm. Ten CANFIS models and MLR were selected based on the performance evaluation indices during testing period. CANFIS models were found to be much closer to the observed values of runoff as compared to MLR.