Investigating the performance of ANFIS model to predict the hourly temperature in Pattani, Thailand

Abstract

The aims of this study are (i) to investigate the performance of adaptive neuro-fuzzy inference system (ANFIS) model for predicting hourly temperature in Pattani, Thailand, and (ii) to compare its performance with statistical methods by using root mean square error (RMSE) and mean square error (MSE) as tools to evaluate. The observation data used in this study are from automatic weather station (AWS) in Pattani, Thailand, collected during January, 18th 2014 to January, 18th 2015 (in total 8,640 data series). The data will be separated into training and checking data sets (ratio 70% for training and 30% for checking dataset). ANFIS can optimize the performance of fuzzy model by tuning the parameter in membership function. In ANFIS method, the model will combine the learning capabilities of a neural network and reasoning capabilities of fuzzy logic in order to increase predicting ability. The ANFIS model has been built by using seven of the generalized bell-shaped membership functions with the linear output. The statistical techniques for time series forecasting used in this study are ARIMA and Exponential Smoothing which are powerful model and general used for time series forecasting. Results showed that ANFIS model had smaller RMSE and MSE than other models.