Temperature time series: Pattern analysis and forecasting

Abstract

This paper uses time-frequency methods and neural networks for the analysis and forecasting of indoor temperature time series. In a first phase, the time series are processed by means of the Fourier transform and the empirical mode decomposition methods to unveil temporal patterns embedded in the data. In a second phase, neural networks are adopted for forecasting future values. The results obtained illustrate the effectiveness of the tools used and motivate further developments based on time-frequency techniques for designing the NN forecasting approach.