Sistemas evolutivos baseados em regras para previsão de séries temporais meteorológicas

Abstract

This work considers evolving intelligent methods for weather time series prediction. We evaluate the methods evolving Takagi-Sugeno (eTS), eXtended Takagi-Sugeno (xTS), Dynamic Evolving Neural Fuzzy Inference System (DENFIS), Fuzzy-Set Based evolving Modeling (FBeM), and a variation of cloud-based intelligent method known as typicality-and-eccentricity-based method for data analysis (TEDA). The objective is to develop data-centric nonlinear and timevarying models to predict mean monthly temperature. Past values of minimum, maximum and mean monthly temperature, as well as previous values of exogenous variables such as cloudiness, rainfall and humidity are considered in the analysis. A non-parametric Spearman correlation based method is proposed to rank and select the most relevant features and time delays for a more accurate prediction. The datasets were obtained from weather stations located in main Brazilian cities such as Sao Paulo, Manaus, Porto Alegre, and Natal. These cities are known to have particular weather characteristics. Additionally, an ensemble of cloud and fuzzy models and fuzzy aggregation operators is developed to give single-valued and granular predictions of the time series. Granular predictions convey a range of possible temperature values. Therefore, it provides a notion about the error and uncertainty associated with the single-valued predictions.