Data-driven spatio-temporal modeling with cellular automata and fuzzy time series methods

• Ano de defesa: 2022
• Autor(a) principal: Lucas Malacarne Astore
• Orientador(a): Frederico Gadelha Guimarães
• Banca de defesa: Tiago Garcia de Senna Carneiro, Walmir Matos Caminhas, Pedro Paulo Balbi de Oliveira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Minas Gerais
• Programa de Pós-Graduação: Programa de Pós-Graduação em Engenharia Elétrica
• Departamento: ENG - DEPARTAMENTO DE ENGENHARIA ELÉTRICA
• País: Brasil
• Palavras-chave em Português: Fuzzy times series; Cellular automata; Spatio-temporal modeling; Land cover land usage; Dynamics modeling
• Link de acesso: http://hdl.handle.net/1843/51934

There have been several applications of computer simulations in studies of spatio-temporal dynamic systems, including epidemiological models. Among the strategies capable of reproducing and predicting future states and behaviors over time, Cellular Automata (CAs) are often applied in geospatial environmental modeling. The core concept of a typical and well-defined CAs model is the development of local rules set that describe the future cell states considering the neighboring cells. The process of building this set demands technical knowledge and years of scientific research. Machine learning-based techniques can be applied in order to automate it, although hyper-parameter optimization algorithms are required. Therefore, this work presents a data-driven approach for CA transitional rules set definition, based exclusively on historical data of a given spatio-temporal phenomenon. The local rules of the automaton are learned and represented using a Multivariate Fuzzy Time Series (MVFTS) method. The MVFTS model is then integrated into the CA simulation, working similarly to a traditional set of CA rules. The proposed methodology was tested using two study cases: Spatial Spread of Chagas Disease and Land Cover/Use Change in Delhi, India. In both sets of data, there was great potential for using the FTS model as a state transition strategy in CA.