Banca de DEFESA: GILLIARD CUSTÓDIO

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
DISCENTE : GILLIARD CUSTÓDIO
DATA : 28/07/2022
HORA: 09:00
LOCAL: https://conferenciaweb.rnp.br/webconf/ronaldo-5
TÍTULO:

Soil Moisture Forecast as Time Series

PÁGINAS: 61
GRANDE ÁREA: Ciências Exatas e da Terra
ÁREA: Ciência da Computação
SUBÁREA: Metodologia e Técnicas da Computação
RESUMO:

Proper management of water resources in Brazil is a demand of public interest. Among the high consumers of the potable water resource, the irrigation process of crops stands out, which is responsible for the consumption of approximately 50\% of all consumed water in a year.  Irrigations systems waste around 60\% of the water resource, and for that reason,  there is room for improvement. The Smart Water Management Platform (SWAMP) aims to use technology to minimize water waste. The soil moisture is an important variable used by SWAMP to calculate the water needed for irrigation. In this work are used multivariate time series modelling techniques to predict the soil moisture value. It was used historical data from two years with multiple components such as soil moisture, soil temperature, weather components and others to predict the soil moisture value. During the modelling step were evaluated machine learning algorithms,  a deep learning algorithm and a traditional statistical technique. The algorithms \textit{Extreme Gradient Boosting} and \textit{Random Forest}  were selected among the machine learning algorithms. The \textit{Spectral Temporal Graph Neural Network} (StemGNN) neural network architecture was the evaluated deep learning algorithm. And the \textit{Vector Autoregression} was defined as a reference of expected performance in predicting soil moisture. Preliminary experiments show that the \textit{Random Forest} algorithm was the most efficient in the process of predicting soil moisture. Furthermore, it was observed that StemGNN was not able to outperform the reference model in most of the evaluated modelling scenarios.

MEMBROS DA BANCA:
Presidente - Interno ao Programa - 1673092 - RONALDO CRISTIANO PRATI
Membro Titular - Examinador(a) Interno ao Programa - 2196309 - CARLOS ALBERTO KAMIENSKI
Membro Titular - Examinador(a) Externo à Instituição - RAMIDE AUGUSTO SALES DANTAS