Comparison of Weather Parameter Forecasting Results using Statistical and Artificial Neural Network Methods

Predicting hourly dry bulb temperature is important and it can use for different studies like forecasting building energy consumption, architectural, agriculture and etc. So, it is necessary to develop and compare forecasting methods. In this study two Forecasting methods, statistical (regression and exponential smoothing- ۵ cases) and neuralnetwork (۱۲ cases) have been used to predict hourly dry bulb temperature. Hourly dry bulb temperature (۸۷۶۰ hour) is available for period of ۱۹۹۸-۲۰۱۹ of Tabriz Iran. The goal of paper is predicting data of year ۲۰۱۹ using past available data (۱۹۹۸-۲۰۱۸) and compare mentioned methods. In statistical methods, linear regression method withlow error rate has been selected as the best method with error of ۴.۰۷ for Mean Squared Error (MSE), ۲.۰۲ for Root Mean Squared Error (RMSE), and ۰.۸ for Mean Absolute Error (MAE). In the neural network method, best prediction case has Errors ۰.۸, ۰.۹ and ۰.۷ for MSE, RMSE and MAE. Result shows that using regression method to predict data have high error. Also using neural network method only in some special case has good agreement with data (Bayesian Regularization training method: ۲ layers with higher than ۲۰ neurons).