Research

Machine Learning Techniques for Anomaly Detection: Applications in Renewable Energy

This research project aims to develop and evaluate advanced methods for anomaly detection using machine learning techniques. Anomalies refer to events or patterns that significantly deviate from expected behavior in a dataset, and efficiently detecting these anomalies is critical in various applications, including cybersecurity, fraud detection, and the monitoring of critical systems such as energy generation systems. Effective monitoring and maintenance are crucial for ensuring sustainable energy generation in the context of energy systems. Early detection of anomalies, failures, or component degradation is essential to minimize disruptions, maximize efficiency, and extend the lifespan of renewable energy assets. This project will explore various machine learning algorithms, data preprocessing techniques, and performance evaluation strategies to improve the accuracy and effectiveness of anomaly detection in renewable energy systems, with a particular focus on wind energy.

Team: Diego Brandão, Rodrigo Toso (Microsoft), Angela Ferreira (IPB/Portugal), Rafaelli Coutinho (CEFET/RJ), Gustavo Semaan (UFF)

 

MultiSoils: a Framework for Soil Security

The MultiSoils´ objective is to host, connect, and share large amounts of curated soil data and knowledge at the Brazilian and South American levels. The e-infrastructure consists of several layers of services, a curated database of soil profiles, and a cloud-based computational framework to prepare, compute, and share soil data integrated with map visualization tools. MultiSoils is an open, elastic, provenance-oriented, and lightweight computational e-infrastructure that collects, stores, describes, curates, harmonizes, and directs to various soil resource types: large datasets of soil profiles, services/applications, documents, projects, and external links. The map viewer allows the users to navigate through key soil data for the region.

Team: Diego Brandão, Marcos Ceddia (UFRRJ), Jorge Soares (CEFET/RJ), Renato Mauro (CEFET/RJ), Frederico Machado (Petrobras)

 

Development of Computationally Efficient Models for Science and Engineering Applications 

This project aims to develop computationally efficient models for simulating problems in Science and Engineering. The reduction in computational costs for these problems will be achieved by applying optimization techniques, adaptive approaches, and high-performance computing methods. The first group of techniques involves representing the problem using linear, integer, or mixed-integer programming models, leveraging exact methods and/or heuristics for problem-solving. The second group employs a local refinement approach to the computational domain of the problem, with a focus on graph-based techniques. In terms of high-performance computing, parallel processing techniques will be explored to reduce the simulation time of the methods developed by the other two groups. This project incorporates heuristics, metaheuristics, exact methods, and numerical techniques. This project includes addressing problems such as network optimization, which emphasizes routing and wavelength assignment problems and divisible task scheduling; operations research, which focuses on optimization problems in the energy sector; capacitated clustering; and combinatorial optimization problems. Adaptive and High-Performance Approaches emphasize numerical methods for simulating environmental, financial, and traffic problems, damage detection, and cancer growth modeling.

Team: Diego Brandão, Laura Assis (CEFET/RJ), Kennedy Fernandes (UFSB), Pedro Henrique Gonzalez (COPPE/UFRJ), 

Computational Modeling Applied to Environmental Problems: Theoretical and Simulation Aspects

This project aims to develop computationally efficient models for simulating environmental problems, specifically focusing on models that describe the behavior of a pollutant in a water body. The computational costs of classical methods will be reduced through the use of adaptive mesh refinement and parallel processing techniques. The first set of techniques uses tree-based models to represent the computational domain. In these models, the cost of obtaining neighborhood information is higher than uniform domain representation. To address this, graph-based representation techniques are proposed to reduce this cost. Regarding parallelism, the objective is to employ the Hopmoc method with a numerical flow controller. Finally, pattern recognition techniques will be applied to determine the best parameters for calibrating the methods developed.
 
Team: Diego Brandão,  Sanderson Lincohn Gonzaga de Oliveira (Unifesp), Edcarlos dos Santos (UFSB), Carla Osthoff (LNCC)
 
 

Computational Modeling and Sensor Networks for Environmental and Engineering Problems

The main objective of this research is to develop a monitoring system using sensor networks. The data obtained through this system will be used to calibrate mathematical models that describe environmental and engineering problems. Wireless Sensor Networks (WSNs) are an emerging technology where small devices, called sensors, are primarily used to monitor hard-to-reach or inhospitable areas, such as oceans, deserts, forests, aquatic niches, and industrial complexes. The use of WSNs for monitoring has been strongly developed over the past decades. This approach is particularly appealing compared to traditional methods. For instance, data collection processes are almost manual in monitoring river water quality. The developed system will provide essential data for the simulation of computational models and the formulation and simulation of inverse problems in the environmental and engineering fields.

Team: Diego Brandão, Antonio José da Silva Neto (UERJ), Jader Lugon (IFF), Raphael Guerra (UFF), Fernando Carvalho (IFPE), Juliano Kazienko (UFSM), Silvio Quincozes (Unipampa)