Novel approaches for air temperature prediction: A comparison of four hybrid evolutionary fuzzy models

Abstract

AbstractThe application of a novel method of adaptive neuro‐fuzzy inference system (ANFIS) for the prediction of air temperature is investigated. The paper discusses the improvement of the ANFIS when used with genetic algorithm (GA), particle swarm optimization (PSO), ant colony optimization for continuous domains (ACOR) and differential evolution (DE). For this purpose, three input of multiple variables are selected in order to predict monthly minimum, average and maximum air temperatures for 34 meteorological stations in Iran. The co‐efficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE) and Nash–Sutcliffe efficiency (NSE) are used as evaluation criteria. A comparison of suggested fuzzy models indicates that the ANFIS with the GA has the best performance in the prediction of maximum temperatures. It decreases the RMSE of the classic ANFIS model in the validation stage from 1.22 to 1.12°C for Mashhad, from 1.26 to 1.01°C for Zahedan, from 1.20 to 0.98°C for Ahvaz, from 1.76 to 1.24°C for Rasht and from 1.21 to 0.95°C for Tabriz.