Eventos Anais de eventos
COBEM 2017
24th ABCM International Congress of Mechanical Engineering
A GENETIC ALGORITHM BASED CLUSTERING APPLIED TO MULTIVARIATE TIME SERIES
Submission Author:
Karine Ribeiro , BA
Co-Authors:
CRISTIANO FONTES, Karine Ribeiro
Presenter: Karine Ribeiro
doi://10.26678/ABCM.COBEM2017.COB17-0404
Abstract
This paper presents a method based on Genetic Algorithm (GA) and Fuzzy C-Means (FCM) for clustering multivariate time series. The method is applied to a real industrial case study which comprises pattern recognition for detecting operation failures in a gas turbine. The time series were collected from the Plant Information Management Systems (PIMS) and are associated with turbine starting events. In the proposed algorithm, each chromosome is an individual or solution, which encodes the clusters' centroids (patterns). A bi-criterion constrained clustering is proposed aiming to maximize both the similarity of objects in the same cluster (based on the SPCA metric) and the distance between the centers of the clusters. The proposed genetic algorithm obtained better results when compared to the traditional clustering method, the fuzzy c-means, according to the misclassification results. The recognized patterns (fault and normal operation) represent a potential for using in control systems or FDD (Fault Detection and Diagnostics) strategies, enabling the monitoring of the distance from the real process to the fault (or normal) operation condition.
Keywords
genetic algorithm, Clustering, Multivariate Time Series, pattern recognition, Fault Detection
