Arthur Plínio de Souza Braga

Application of an artificial neural network in the use of physicochemical properties as a low cost proxy of power transformers DGA data

This paper is about the relationship between dissolved gases and the quality of the insulating mineral oil used in power transformers. Artificial Neural Networks are used to approach operational conditions assessment issue of the insulating oil in power transformers, which is characterized by a non-linear dynamic behavior. The operation conditions and integrity of a power transformer can be inferred by analysis of physicochemical and chromatographic (DGA - Dissolved Gas Analysis) profiles of the isolating oil, which allow establishing procedures for operating and maintaining the equipment. However, while the costs of physicochemical tests are less expensive, the chromatographic analysis is more informative and reliable. This work presents a method that can be used to extract chromatographic information using physicochemical analysis through Artificial Neural Networks. Its believed that, the power utilities could improve reliability in the prediction of incipient failures at a lower cost with this method. The results show this strategy might be promising. The purpose of this work is the direct implementation of the diagnosis of incipient faults through the use of physicochemical properties without the need to make an oil chromatography.

Filtered model-based predictive control applied to the temperature and humidity control of a neonatal incubator

This paper proposes a robust multivariable predictive control algorithm that improves the robustness of closed loop systems, even when they have multiple time delays between the inputs and outputs. The desired robustness is achieved by including an appropriate filter on the disturbances model. The proposed algorithm is applied to the control of humidity and temperature of a neonatal incubator. Simulation and experimental results show the advantages of the proposed algorithm compared to others proposed in the literature.

Artificial Neural Network Application in Estimation of Dissolved Gases in Insulating Mineral Oil from Physical-Chemical Datas for Incipient Fault Diagnosis

In this paper, Artificial Neural Networks are used to solve a complex problem concerning to power transformers and characterized by non-linearity and hard dynamic modeling. The operation conditions and integrity of a power transformer can be detected by analysis of physical-chemical and chromatographic isolating oil, allowing establish procedures for operating and maintaining the equipment. However, while the costs of physical- chemical tests are smaller, the chromatographic analysis is more informative. This work presents an estimation study of the information that would be obtained in the chromatographic test from the physical-chemical analysis through Artificial Neural Networks. Thus, the power utilities can achieve greater reliability in the prediction of incipient failures at a lower cost. The results show this strategy to be a promising, with accuracy of 100% in best cases. The application in the thermal fault diagnosis presents more than 91% accuracy in best cases.

Influence zones: A strategy to enhance reinforcement learning

Abstract Reinforcement Learning (RL) aims to learn through direct experimentation how to solve decision-making problems. RL algorithms often have their practical applications restricted to small or medium size problems—mainly because of their strategies for value function estimation demanding very high number of interactions. To overcome this difficulty, we propose to enhance RL performance by updating several state (or state–action) values at each interaction. Therefore, the influence zone algorithm, an improvement over the topological RL agent (TRLA) strategy, allows to reduce the number of requested interactions. Such a reduction is based on the topological-preserving characteristic of the mapping between states (or state–action pairs) and value estimates. The comparison of the influence zone approach with seven other RL algorithms suggests that the proposed algorithm is among the fastest to estimate the value function and that it takes less value function updatings to perform such an estimation. The influence zone algorithm also presents a remarkable flexibility in adapting its policy to changes of the input space topology.

Fuzzy logic controller implementation on a FPGA using VHDL

Fuzzy-logic-based control (FLC) systems have emerged as one of the most promising areas for research in Applied Computational Intelligence. These systems operate with knowledge represented in a linguistic form (IF-THEN rules) that describes relations which are not precisely known, but those effects are intuitively understood by humans. This fundamental feature makes FLC a powerful tool for industrial applications, since complex systems can be controlled by easily comprehensive rules. The growth in number of fuzzy logic applications led to the need of finding efficient and economic ways to implement them. The Field Programmable Gate Arrays (FPGAs) devices, with their reconfigurable logic, practicality, portability, low consumption of energy, high operation speedy and large datastorage capacity, are a great choice for FLC embedded systems project development and prototyping. In this paper, the design and implementation of a Mamdani-type Fuzzy controller is demonstrated using VHDL programming language.

Decompositional Rule Extraction from Artificial Neural Networks and Application in Analysis of Transformers

The artificial neural networks represent efficient computational models that are widely used to solve problems of difficult solution in Artificial Intelligence. The greatest difficulty associated with the use of Artificial Neural Networks (ANN) is in obtaining knowledge about its behavior, because of that ANNs are also considered as black-box methods. This paper presents a brief history of methods of extraction of knowledge, and in detail a method of interpreting the behavior of an artificial neural network by establishing a relation of equality between certain classes of neural networks and systems based on fuzzy rules, with modifications that allow the acquisition of rules coherent with the domain of the variables of the problem. An example of application is used to illustrate the method, considering the identification of incipient faults in transformers by using data from gas dissolved in transformer oil.

A vehicle classification based on inductive loop detectors using artificial neural networks

This paper presents a classifier based on a net neural MLP (Multiple Layers Perceptron) using the algorithm of Levenberg-Marquardt to do the identification of the load of vehicles through the magnetic profile collected in equipment of control of traffic.

Dynamic loading of distribution transformers from models of dynamic thermal

It is known that to monitor intrinsic life of transformers variables has been the object of concern in regard to avoid economic losses caused by faults in such equipment. It is possible through the system management, maximize the lifetime and improve grid reliability, which could also reduce maintenance costs. This paper brings a contribution to the development of these systems by proposing and testing a model based on the enhancement of IEEE standard for modeling the top oil and the hot spot temperature of distribution transformers of Manaus Energia Utility to establish maximum loading limits for the equipment that minimize the loss of lifetime. The results were compared with a standard IEEE model, which presented more conservative results.

Adaptive Fuzzy Learning Vector Quantization (AFLVQ) for Time Series Classification.

Over the past decade, a variety of research fields studied high-dimensional temporal data for pattern recognition, signal processing, fault detection and other purposes. Time series data mining has been constantly explored in the literature, and recent researches show that there are important issues yet to be addressed in the field. Currently, neural network based algorithms have been frequently adopted for solving classification problems. However, these techniques generally do not take advantage from expert knowledge about the processed data. In contrast, Fuzzy-based techniques use expert knowledge for performing data mining classification but they lack on adaptive behaviour. In this context, Hybrid Intelligent Systems (HIS) have been designed based on the concept of combining the adaptive characteristic of neural networks with the informative knowledge from fuzzy logic. Based on HIS, we introduce a novel approach for Learning Vector Quantization (LVQ) called Adaptive Fuzzy LVQ (AFLVQ) which consists in combining a Fuzzy-LVQ neural network with adaptive characteristics. In this paper, we conducted experiments with a time series classification problem known as Human Activity Recognition (HAR), using signals from a tri-axial accelerometer and gyroscope. We performed multiple experiments with different LVQ-based algorithms in order to evaluate the introduced method. We performed simulations for comparing three approaches of LVQ neural network: Kohonen’s LVQ, Adaptive LVQ and the proposed AFLVQ. From the results, we conclude that the proposed hybrid Adaptive-Fuzzy-LVQ algorithm outperforms several other methods in terms of classification accuracy and smoothness in learning convergence.

MLP Back Propagation Artificial Neural Network for Solar Resource Forecasting in Equatorial Areas

Renewable energy (RE) resources such as solar are increasingly being used worldwide. Solar resource shows high availability, but presents an intermittent characteristic, causing oscillations in the electricity production. Intermittence is one of the main barriers for the use of solar plants in a system that needs to balance demand and electricity production. Aiming to contribute to a larger use of the solar resource in the world energy matrix, we propose a solar irradiance prediction methodology, developed from data collected in Fortaleza-CE (latitude: −03° 43′, longitude: −38° 32′). Predictions were developed using Multilayer Perceptron (MLP) Back propagation Artificial Neural Network (ANN) with the advance of 1 hour. In the best ANN performance, 41.9% of the predictions obtained up to 5% of error, 58.7% obtained errors lower than 10% and 68.6% obtained errors lower than 15%. MAPE (mean absolute percentage error) of 6.11% was found, which can be considered good, since errors found in previous works reached 20%.