Considering analysis from electrical distribution system authorities, it is predicted an increase in electrical energy demand as well a higher level of distribution generation. From this increase in energy matrix, the loads of different types of consumers and non-linear devices, the complexity of the system increases. Thus, the voltage sag in distribution systems is an important event to be monitored, as it can affect consumers several times a year, and may compromise the production of industrial consumers and other segments of society. Therefore, to overcome the existing limitations, Machine Learning techniques are currently being explored for the characterization of voltage sags, however, the theme has not yet reached the stagnation point. With the oscillography stored in intelligent electronic devices, present in the distribution system, characterization using such unconventional techniques becomes possible. This research aims to establish a method for characterization of voltage sags using Machine Learning techniques, for three common situations in distribution systems - energizing a transformer, starting a motor, or presence of a fault. The proposed methodology can differentiate the previously mentioned events from those observed during the normal operation of the electrical network, helping the distribution system operators. Furthermore, the proposed method was implemented with three different intelligent techniques, providing a performance comparison between these techniques, with signals in time domain, providing a performance comparison between such techniques. SVM (Support Vector Machine) was the most suitable technique to implement the proposed method, seeking its operation against more complex electrical systems. 

Autor: Bruno Stabile dos Santos

Acesso: http://biblioteca.ufabc.edu.br/index.php?codigo_sophia=127261  

This work analyzes the influence of photovoltaic solar generation, connected through inverters, on the performance of protection devices in a consumer system supplied by the local utility’s medium-voltage network. The study discusses key aspects related to the protection requirements for this type of installation and establishes criteria for modeling the inverters in the SINAP simulation software. The simulations were based on data from a real distribution network, as well as a detailed representation of the electrical installation of UFABC's Santo André campus. After accurately modeling the electrical network and simulating fault occurrences, it was possible to assess whether the presence of photovoltaic generation requires adjustments to the protection function parameterization criteria in order to meet coordination requirements, as desired and required by standards. Additionally, based on the observed results, important aspects were raised regarding the need to standardize protection schemes required by power utilities. Discussions and comments on the results are presented in the text, along with the simulations performed and some practical tests.  

Autor: Élio Vicentini 

Currently, loads from various types of consumers are increasingly sensitive to power variations and interruptions. In most cases, power supply failures result from faults occurring in distribution systems, which affect the reliability of the energy supply and can lead to voltage sags, momentary and sustained interruptions, and high operational costs. Utility companies are aware of this issue and are working on the development of efficient solutions to improve their reliability and availability indices. In this context, different techniques for fault detection and location are being studied, aiming at the rapid restoration of service in the event of disturbances in the distribution system. Along these lines, this work presents a method that can be used as a decision-support tool for system operators, indicating the faulted area of the distribution system and providing an approximate fault location in the event of a feeder failure. The proposed method uses the similarity between voltage signals and the Hausdorff Distance to achieve this goal. Preliminary tests showed that the method is quite promising, successfully identifying the faulted area in a distribution system composed of dynamic loads and distributed generation. Subsequently, to validate the proposed method, it was tested on the IEEE 34-bus system, once again demonstrating its effectiveness in identifying the faulted region. The solution discussed in this work was implemented using different input signals, data window sizes, and sampling rates, in order to determine the configuration that best balances accuracy and computational cost. According to the observed results, the proposed method is a highly promising option for supporting the operation of distribution systems. It is simple and accurate, and it allows for quick and straightforward adaptations

 Autor: Joaquim Siqueira de Lima

Acesso: http://biblioteca.ufabc.edu.br/index.php?codigo_sophia=124330  

The electric power system is changing. Changes in legislation made it possible to connect distributed generators (DGs) to the grid, and units that only consumed power began to inject it into the grid. Photovoltaic generation has been growing in recent years and is already the most used source as distributed generation in Brazil. One of the main concerns when connecting a DG to the grid is unintentional islanding, which occurs when a portion of the grid containing DG and loads remains electrified, but electrically isolated from the rest of the grid. In this case, the DG power supply is not supervised by the grid. When this situation is not identified by the existing protections, in addition to problems with the quality of energy, accidents can occur, since there is a part of the network that is improperly energized. This work aims to present and compare two methodologies for the detection of islanding of photovoltaic generators. The first methodology is based on artificial neural networks (ANNs), and the detection is carried out through an analysis of the voltage signal at the point of common coupling between the installation considered and the concessionaire. The second proposed methodology uses Discrete Wavelet Transform (DWT) to detect islanding, also analyzing the voltage signal at the connection point with the utility. At the end of this work, the results of the proposed algorithms in the face of islanding situations are presented, as well as tests to delimit the operating limits of each algorithm, making it possible to verify the good performance of both techniques. The algorithms responded correctly in 100% of the practical cases evaluated, even in situations of low power unbalance. The detection time was low for both techniques, between 0.06 s and 0.09 s. 

Autor: Luiza Buscariolli

Acesso: http://biblioteca.ufabc.edu.br/index.php?codigo_sophia=124592

Power systems (EPS) are responsible for providing electrical energy safely and reliably to consumers. In the event of contingencies, it is the protection system's role to act in order to minimize damage to the end consumers and to the PS itself. One of the most important devices in the protection system is the current transformer (CT), whose main task is to provide the protection relay with a scaled replica of the primary current of the PS. However, CTs have physical characteristics that make them susceptible to saturation, which results in distortion of their secondary waveform. Due to this distortion, the protection relay may fail, as its decision depends on the integrity of the signal from the CT. Such failure can lead to severe damage to PS equipment, with significant financial losses, not to mention the risk to human lives. To prevent incorrect decisions by the relay, several algorithms for detecting and reconstructing distorted waveforms from CTs have been proposed in the literature. A significant portion of these algorithms is based on function approximation techniques and the use of artificial intelligence. In this regard, this work proposes the use of artificial neural networks (ANNs) to correct the distorted secondary waveform from CTs. The ANNs used for waveform correction are of the MLP type, specified through a supervised training process. Various sampling rates and data window sizes were evaluated to specify an ANN capable of delivering high performance for the proposed algorithm. Furthermore, for performance comparison, two different computational environments were used to model the ANNs: MATLAB and Keras. The signals used in the training and testing process of the ANNs come from two distinct data sources: the first is based on the classical modeling of CTs, widely used by the scientific community (IEEE), while the second is based on the CT model in Simulink. The results clearly show that a highly distorted waveform can be fully reconstructed by the ANN, providing protection relays with a reliable signal for correct decision-making.

Autor: Bruno dos Santos Saraiva Silva

Acesso: http://biblioteca.ufabc.edu.br/index.php?codigo_sophia=121847   

Ferroresonance phenomenon is an oscillatory effect that occurs when a non-linear inductance of an Inductive Voltage Transformer (IVT) is connected in series with circuits that have equivalent capacitances, such as: transmission lines, underground circuits, capacitive loads, and other arrangements found in electrical power systems (EPS). Due the non-linearity of the IVT magnetic core, this equipment, when connected to grids with specific capacitances, is vulnerable to different behaviors and disturbances, produced by faults or switching operations. This condition is known as Ferroresonance. Considering the dynamic characteristics of this non-linear disturbance, the ferroresonant response may manifest in different ways, as periodic oscillations at the fundamental frequency (fundamental mode) or at sub-multiple values of the fundamental frequency (subharmonic mode), among others. The accurate detection of the ferroresonant effect may prevent equipment damage as well as prevent revenue loss for utilities. In turn, this work presents a method based on Artificial Neural Networks (ANNs) to detect the ferroresonant effect. The proposed algorithm for detecting the ferroresonant effect uses IVT secondary voltage samples to decide about the occurrence of a ferroresonant event. Moreover, this work also discusses and analyzes different scenarios that can generate the ferroresonance phenomenon, by using ATP and PSCAD to simulate all considered cases. The obtained results show that the ferroresonance conditions can be predicted and minimized by means of an accurate modeling and, later, simulation of all possible operational conditions of the EPS. After a large number of simulations and analyses, it was possible to observe that the proposed ANN-based method is able to detect different conditions of ferroresonance.

Autor: Ricardo Sotero da Silva

Acesso: http://biblioteca.ufabc.edu.br/index.php?codigo_sophia=123732

The growth of renewable energy exploitation has driven the expansion of distributed generation. In this context, High Voltage Direct Current (HVDC) transmission systems have emerged as an interesting alternative for integrating these renewable sources. When integrating many renewable sources, HVDC systems tend to evolve into Multi-Terminal Direct Current (MTDC) systems. However, the high fault current levels, the unavailability of direct current (DC) circuit breakers that operate quickly to prevent damage to the system, and the lack of protection schemes that isolate only the faulted segment limit the implementation of MTDC systems. This work presents three different protection proposals, based on the use of Artificial Neural Networks (ANNs), for the protection of MTDC systems. Initially, the feasibility of the proposed protection schemes is verified through a simplified MTDC system, and later validated through simulations on the CIGRÉ three-bus MTDC system. In this scenario, the proposals are compared in terms of response times and the accuracy observed for fault detection, classification, and location tasks. All three proposals use ANNs; however, the first one considers the direct processing of DC current signals, while the second and third proposals use, respectively, the Discrete Fourier Transform (DFT) and the Discrete Wavelet Transform (DWT) for pre-processing the DC current signal. The ANN-based algorithm is developed in Matlab, and the MTDC systems are implemented in the ATP-Draw and PSCAD software, allowing the generation of signals to evaluate the performance of the proposed protection schemes. The results show that all ANN-based proposals are promising, regardless of the pre-processing technique used.

Autor: Júlio Octavio Arita

Acesso: http://biblioteca.ufabc.edu.br/index.php?codigo_sophia=111271  

The basic function of the Electric Power System (EPS) is to provide high-quality power supply to consumers whenever demand occurs. However, fault events on the EPS can cause problems of quality and reliability in the power supply requiring schemes to correctly detect, classify and locate these events. In addition to the inherent complexity in performing these schemes, the penetration of Distributed Generation (DG) in the EPS brings new challenges to the protection functions due to the resulting bi-directional power flow capabilities. Thus, in order to contextualize this scenario, this research presents some methods proposed in the literature for the detection, classification and location of faults in the EPS, considering or not the presence of DG. In this sense, as the main focus of this work, it is proposed an intelligent system based on artificial neural networks (ANNs) to detect, classify and locate faults in the EPS. The proposed scheme uses the post-fault voltage values in the system buses to determine, in the event of a fault, the following information: a) the transmission line (TL) where the fault occurred; b) the type of fault; and c) the location point of the fault in the previously identified TL. The ANNs of the proposed scheme were trained for different types of fault that can occur along the TLs and considering several values of fault resistances. The main characteristic of this algorithm is to dispense with complex formulations and power flow analysis for detection, classification and location of faults in TLs of the EPS, even in the face of challenges brought by the presence of DG. To confirm the validity of the proposed scheme, hundreds of new fault cases were used considering several operating situations from three different EPSs (IEEE 4-bus system, IEEE 9-bus system and IEEE 14-bus system). The results show that the proposed scheme is able to properly identify the line segment with fault, to classify the type of fault that occurred, as well as to indicate the exact location point of the fault, using as input only the voltage values measured in the buses of the electric power system

Autor: Fernanda Soares Vitor Petite

Acesso: http://biblioteca.ufabc.edu.br/index.php?codigo_sophia=109170

This work provides a methodology to determine the levels of reliability/availability in electrical substations, based on the need to improve the efficiency of maintenance operation reducing negative environmental, social, economic and technical impacts, caused by power outages. The methodology is based on two methods typically used individually in reliability studies. The method called Fault Tree that provides a logical model of possible failure combinations for a major event, and the Monte Carlo simulation used to determine the power system index by random generation of the different states of the system (operation, failure or maintenance). Considering this context, in this work are identified vulnerabilities points, the probability of failure and the unavailability of each substation, in order to increase the reliability indices, increase the service life of components and provide a better preventive maintenance scheduled. Consequently, this works seeks to decrease the frequency of uncontrolled power cuts and their environmental, social and economic impacts produced by nonsupply of electricity. In this sense, a discussion about the impacts of electrical faults to society is also conducted. 

Autor: Jair Dias Barbosa

Acesso: http://biblioteca.ufabc.edu.br/index.php?codigo_sophia=77339 

Since the power demand is increasing, additional electrical power sources are required to properly supply all consumers. To meet this new increased demand small generators (DGs) are installed in the power grid, thus allowing to use electrical power sources available around the country. However, to connect such DGs in the power grid the islanding event must be considered. An islanding condition takes place when a section of the grid is inadvertently isolated from the utility, thus resulting in problems related to the power quality delivered to the consumers, among others. Depending on the operational condition, it is a complex task to distinguish an islanding situation from other possible events in the power system, new methods are needed to develop this task. This work presents an islanding detection algorithm based on artificial neural network (ANN), more specifically by using a multilayer perceptron (MLP). Considering the presented algorithm a digital relay architecture is proposed to be implemented in hardware by using the software LabVIEW and its commercial platforms (cDAQ-9172 and cRIO-9073). The proposed algorithm was evaluated in hardware using real signals and its expected behavior was checked. Besides, a study considering different sampling rates and different types of hardware (real time and not real time) platforms was developed, thus allowing to precisely verify the relation "hardware x accuracy". 

Autor: Nolman Barroso Hartmann

Acesso: http://biblioteca.ufabc.edu.br/index.php?codigo_sophia=77570

Distributed generation systems are becoming an increasingly common alternative to meet the growing energy demand, as they aim to reduce losses in the electrical system. However, the implementation and expansion of distributed generation systems also bring the need for more efficient protection of these systems, with the goal of preventing equipment and installation damage, minimizing power quality issues, and reducing the risk of personal accidents in case of network anomalies. One of the key elements for the protection of distributed generation systems is the detection of islanding in generators. The purpose of this detection is to disconnect the distributed generators, isolating them from the system in the event of a utility outage, thus preventing transmission and distribution lines from remaining unnecessarily energized by the distributed generators. This work aims to present a methodology for defining an algorithm based on artificial neural networks for islanding detection in synchronous generators. This detection is performed by analyzing the variations in the voltage signal at the bus of the generator to be protected. The voltage is sampled and structured into a data window to be processed by the algorithm. At the end of this work, the results of the proposed algorithm under various operational conditions of the electrical system are presented, demonstrating its good performance. The algorithm responded correctly in 100% of the practical cases evaluated, even in situations of low power imbalance.

Autor: Victor Luiz Merlin

Acesso: http://biblioteca.ufabc.edu.br/index.php?codigo_sophia=118659

This paper has as final objective the hardware implementation of an artificial neural network based system for the remote monitoring voltage of interest points in the power system. In a first phase, which is outside the scope of this work, studies and simulations of the power system in question were made to specify an artificial neural network. In a second phase, which is the focus of this work, the neural network initially specified is implemented in hardware and evaluated against real signals, using the software LabVIEW and the data acquisition platform CompactDAQ and CompactRIO. Every step used to model in software and evaluate in hardware one artificial neuron and, posteriorly, one neural network, will be discussed in this work. The practical results obtained are always compared with the results derived of simulations performed in the Matlab and LabVIEW environments, in order to support the methodology proposed in this project. As observed in laboratory, the obtained results indicate the effectiveness of the algorithm developed, based in neural networks, as well as the software and the data acquisition platform used to the implementation of the referred algorithm. 

Autor: Alex Soto da Silva

Acesso: http://biblioteca.ufabc.edu.br/index.php?codigo_sophia=72672  

Electric power is extremely important for all sectors of society. Therefore, it must be delivered to the final consumer within predefined limits of quality and reliability. Among other factors, to achieve this objective, it is essential that the electrical system has an efficient protection scheme, whose failure can cause serious damage to the system and the subsequent interruption in the electricity supply to the final consumer. Given the importance of protection devices in the performance of the electrical system, this work is dedicated to the design of differential protection for power transformers, focusing primarily on its implementation in a programmable logic device (PLD). The algorithm for the proposed protection is based on a multilayer perceptron (MLP) artificial neural network (ANN), which has the main characteristics of pattern recognition and function approximation. These characteristics enable the ANN to distinguish, through continuous evaluation of signals from the transformer, when it is operating under normal conditions or in a critical condition. Considering that performing certain functions using an ANN can lead to complexity, this work uses a high-capacity field-programmable gate array (FPGA) PLD for implementing the differential protection. The adopted FPGA is the EP1C12Q240C8 from the Cyclone family, which is part of the UP3 development platform. Implementing a complex digital system in PLD, such as an ANN, requires the use of development software and a hardware description language. For this work, the Quartus II development software and the VHDL hardware description language were chosen.

Autor: Isaque da Silva Almeida 

Fault Location (FL) in Transmission Lines (TLs) is an essential function to ensure continuity of service in Electric Power Systems (EPS). In general, a complete FL scheme is formed by two previously steps, that is Fault Detection (FD) and Fault Classification (FC). Conventional methods for FL may present some limitations, such as the use of current signals, high computational cost, dependency on communication links, performance loss against different systems or fault characteristics. The objective of this thesis is to propose a complete and reliable FL method, addressing some limitations aforementioned. For this purpose, the FD and FC were designed by using Euclidian distance, while the FL was developed using Independent Component Analysis (ICA). To improve the reliability of the proposed method for FL, an intelligent Disturbance Classification (DC) based on Convolutional Neural Network (CNN) was also developed. The proposed methods for FD, FC, FL and DC work in time domain and they were all evaluated against different EPS, fault characteristics (in PSCAD), always presenting good results and advantages when comparing to conventional methods. Moreover, a comparison considering different possibilities for implementing the DC method is presented, proving that ICA is the best option to achieve an accurate and robust performance. 

Autor: Guilherme Torres de Alencar

Acesso: http://biblioteca.ufabc.edu.br/index.php?codigo_sophia=127286  

This work presents an intelligent algorithm for fault detection and location in High Voltage Direct Current (HVDC) systems based on Line-Commutated Converters (LCC), also known as LCC-HVDC. For fault detection are used AC (Alternating Current) and DC (Direct Current) signals from the rectifier substation, while for fault location only DC post-fault voltage signal is used. Thus, the proposed intelligent scheme does not require any communication link, using only local signals. Both mentioned functions are performed by Artificial Neural Networks (ANNs), being the fault location function based on similarity concept. Firstly, the proposed solution is developed and evaluated by means of the HVDC CIGRE benchmark and, secondly, it is validated by using a new LCC-HVDC system. The developed algorithm was evaluated considering hundreds of different fault cases, always presenting an accurate and fast response, even against different fault locations in the DC line, pre-fault currents and fault resistances. After its validation process in Matlab, the proposed algorithm was implemented in hardware and tested by using RTDS (Real Time Digital Simulator), showing to be a feasible alternative for practical purposes.

Autor: Alex Soto da Silva

Acesso: http://biblioteca.ufabc.edu.br/index.php?codigo_sophia=123721

High voltage direct current (HVDC) systems are becoming an usual option for supplying the increase of energy demand, since these systems may supply more efficient connection from power generators to remote load centers. Moreover, HVDC systems have been considered a viable alternative to connect renewable energy sources as offshore windfarms. However, the implementation of HVDC systems brings the need for a fast and efficient protection for those systems, in order to avoid damages to equipment and infrastructure and minimize problems of power quality in case of power system faults or disturbances. HVDC protection presents many challenges on fault detections, classification and location, since they may occur upstream the rectifier station (on AC), in the transmition line (on DC) or downstream the inverter substation (also on AC). A fast fault detection in HVDC systems have is important, since these systems transmit large amounts of power and due to the technical limitations of disconnection devices. The purpose of this work is to present a methodology based on artificial neural network for fault detection, classification and location in VSC-HVDC and LCC-HVDC systems. This detection is made analyzing AC/DC voltage signals present on the rectifier substation without using communication links. Even so, the algorithms are able to cover faults in the converter substations as well as in the DC line. In the end of this work, the test results of the proposed algorithm for two different HVDC systems, one VSC and other LCC, are presented. In both systems, the algorithm was widely tested in many different fault cases in order to confirm its good performance. In order to develop and evaluate the proposed algorithms, a large number of fault cases were simulated, varying the fault type, the fault location and the fault resistance as well as the operational conditions of AC systems. The results show that, through the proposed methodology, the obtained algorithms are effective for protection tasks in the VSC-HVDC and LCC-HVDC systems adopted in this work. 

Autor: Victor Luiz Merlin 

Acesso: http://biblioteca.ufabc.edu.br/index.php?codigo_sophia=118659