Título: Probabilistic Forecasting with Seasonal Ensemble Fuzzy Time-Series
Autores: Lima e Silva, Petrônio Cândido;Alves, Marcos;Severiano Junior, Carlos Alberto;Linhares Vieira, Gustavo;Guimarães, Frederico;Javedani Sadaei, Hossein
The aim of this paper is to propose a method for probabilistic forecasting based on the aggregation of seasonal Fuzzy Time-Series techniques with ensemble learning. The proposed method generates different seasonal FTS models and the best ones are combined into an ensemble learning. The forecasting procedure consists in evaluating individual models and combining their outputs into a continuous probability distribution using Kernel density estimation. The method was applied to SONDA dataset considering three seasonal indexes on solar radiation data. The best Ensemble models were those with 15 minutes interval index and Entropy partitioning in their different parameters. The built ensemble forecasts were then compared with ARIMA and Quantile Auto-Regression models using Continuous Ranked Probability Score (CRPS) metric. The Ensemble FTS method presented a slightly larger CRPS, especially for the Epanechnikov, Tophat and Triangular kernels, which suggests a better model.
Fuzzy forecasting; Time-series; Fuzzy seasonality; Ensemble Learning.
Código DOI: 10.21528/CBIC2017-54
Artigo em pdf: cbic-paper-54.pdf
Arquivo BibTeX: cbic-paper-54.bib