PEI PROGRAMA DE PÓS-GRADUAÇÃO EM ENGENHARIA INDUSTRIAL (PEI) ESCOLA POLITÉCNICA Teléfono/Ramal: No informado

Banca de DEFESA: BRENNER BIASI SOUZA SILVA

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
DISCENTE : BRENNER BIASI SOUZA SILVA
DATA : 19/06/2019
HORA: 09:00
LOCAL: Sala de Congregação da Escola Politécnica da UFBA
TÍTULO:

OPERATIONAL ANALYSIS IN INDUSTRIAL DRAINAGE SYSTEM: A MACHINE LEARNING APPROACH


PALAVRAS-CHAVES:

Overflow of industrial effluents. Unsupervised learning. Supervised learning.


PÁGINAS: 87
GRANDE ÁREA: Engenharias
ÁREA: Engenharia de Produção
RESUMO:

The occurrence of overflows in industrial effluent retention systems is an important environmental and operational aspect in most industries. The objective of this work was to analyze and classify the behavior of rainfall and overflows, and to propose a predictive model of overflows into a basin of the industrial effluent retention system in an oil refinery. The cluster analysis for precipitation and tank level time series is analyzed from the perspective of similarity between clusters and the detection of change points to determine the behavior of time series. A constructed predictive model was also proposed for rainwater drainage systems in industrial areas using supervised machine learning with the objective of indicating whether the containment basin will overflow within 24 hours, using the techniques used k-nearest neighbors (KNN) and Random Forest. The set of unsupervised machine learning methodologies used allows information to be obtained on hydrological and process events in scenarios with low data availability without the need to increase the information. It was identified that, in the absence of precipitation or occurrence of low daily precipitation volumes, the system failed, and the percentage of overflows is higher than the expected natural value. In addition, there are no overflows in rainy periods if there is satisfactory operation of the system. Scenarios and variations in sampling techniques for training the classification models were used. The best results were obtained from the Random Forest algorithm using the oversampling, undersampling and ROSE re-sampling techniques.


MEMBROS DA BANCA:
Presidente - 2574950 - KARLA PATRICIA SANTOS OLIVEIRA RODRIGUEZ ESQUERRE
Externo ao Programa - 2276528 - LEIZER SCHNITMAN
Externo ao Programa - 1741302 - LUCIANO MATOS QUEIROZ
Notícia cadastrada em: 17/06/2019 14:38
SIGAA | STI/SUPAC - - | Copyright © 2006-2024 - UFBA