Banca de QUALIFICAÇÃO: HIGOR ÍTALO DOS SANTOS

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
STUDENT : HIGOR ÍTALO DOS SANTOS
DATE: 18/11/2022
LOCAL: Laboratório de Telemática (fechada ao público).
TITLE:

UNSUPERVISED ALGORITHMS FOR THE CONSTRUCTION OF PROBABILISTIC AUTOMATA FROM TIME SERIES.

KEY WORDS:

Machine learning. Probabilistic finite state automaton. Graph minimization. Unsupervised modeling. Dynamical systems.

PAGES: 60
BIG AREA: Engenharias
AREA: Engenharia Elétrica
SUMMARY:

One of the commonly studied methods of analysis and modeling of dynamical systems involves processes of symbolization of time series associated with dynamic systems. Among the main advantages of this method are the reduction of the computational cost linked to the signal processing and robustness to noise. Through the use of Probabilistic Finite State Automata (PFSA) the discrete dynamics of the resulting series can be represented. These models are often employed in control algorithms and fault detection techniques. This work proposes a new algorithm for the determination of a PFSA related to a dynamical system from a sufficiently long realization of the dynamics, making use, from the analysis of the generated discrete sequence, of statistical methods, unsupervised machine learning and graph minimization, to obtain reduced PFSA models faithful to the discrete dynamics coming from the original system. The proposed method is finally submitted to dynamic systems with different characteristics, demonstrating its ability to generate models with a reduced number of states compared to other approaches in the literature and that maintain their quality.

COMMITTEE MEMBERS:
Presidente - 1851787 - DANIEL PEDRO BEZERRA CHAVES
Interno - 1882484 - JULIANO BANDEIRA LIMA
Externo ao Programa - 1288283 - RENATO MARIZ DE MORAES - null