A Distributed Algorithm for Scalable Fuzzy Time Series

Petrônio Cândido de Lima e Silva, Patrícia de Oliveira e Lucas, Frederico Gadelha Guimarães

Abstract: Time series forecasting, an essential task in management of Smart Cities and Smart Grids, becomes challenging when it needs to deal with big data. The development of highly accurate machine learning models is yet harder when considering the optimization of hyper-parameters, an expensive computational task. To tame these challenges this work proposes the Weighted Multivariate Fuzzy Time Series method, a simple and non-parametric forecasting method with high scalability and accuracy. A stack of methods is presented, which comprises a sequential training and forecasting procedure and a Map/Reduce extension for distributed processing. The stack of proposed methods was evaluated using a cluster with commodity hardware and a big data of solar energy time series, achieving good performance in feasible processing time.

External IDs:dblp:conf/gpc/SilvaLG19