Predicting Temperature Using Optimized Adaptive Neuro-fuzzy Interface System and Bayesian Model Averaging

Abstract

This study uses an optimized adaptive neuro-fuzzy interface system (ANFIS) and Bayesian model averaging (BMA) to estimate one-month-ahead temperature. The lagged temperatures were used as the inputs to the models. The dragonfly optimization algorithm (DRA), rat swarm optimization (RSOA), and antlion optimization algorithm (ANO) were used to set the ANFIS parameters. The results indicated that the BMA model outperformed the other models. Also, the DRA had the best performance among other optimization algorithms. The Nash–Sutcliffe efficiency (NSE) of the BMA, ANFIS-DRA, ANFIS-RSOA, ANFIS-ANO, and ANFIS models was 0.96, 0.91, 0.90, 0.89, and 0.87, respectively. The BMA and ANFIS-DRA had the highest NSE values at the testing level. It was observed that increasing time horizons decreased the accuracy of models.