Temperature Prediction using Time Series Time-Delay Neural Networks

Abstract

Having an accurate temperature predictions, provides a very positive impact on various fields such as agricultural based industries, aviation, tourism and also taking precautionary measures, in case of extreme weather conditions like heatwaves. This paper propose a temperature prediction model that covers the city of Agadir, The data used to train and test the network are real data obtained by the Moroccan Meteorological Administration (METEO MAROC), which represent temperature values on 30 minutes basis, taken from sensors based in the AGADIR EL MASSIRA Airport. The model is based on the use of Time Series Time-Delay Neural Networks. A type of networks considered to be one of the most straightforward dynamic network, which consists of a feedforward network with a tapped delay line at the input. The designed prediction network model showed good performance, as shown in the calculated mean squared errors (MSE) for the testing data set is about 0.4149, and its correlation coefficient R is about (0.99). The correlation of the predicted error with time showed that almost all the autocorrelation function values fall within the bound of the confidence interval.