Description
ABSTRACT
Power systems frequently experience variations in their operation, which are mostly manifested as transmission line faults. Over the past decade, various techniques of fault diagnosis have been developed to ensure reliable and stable operation of power systems. Power system faults are abnormal conditions that disrupt the normal operation and stability of the electrical network. They can cause damage to equipment, loss of supply, safety hazards, and economic losses. Therefore, detecting and diagnosing faults quickly and accurately is essential for maintaining the reliability and security of power systems. In this seminar, you will learn about some of the most effective methods for power system fault detection and diagnosis, and how they can help you improve your power system coordination.
TABLE OF CONTENTS
COVER PAGE
TITLE PAGE
APPROVAL PAGE
DEDICATION
ACKNOWLEDGEMENT
ABSTRACT
CHAPTER ONE
INTRODUCTION
1.1 BACKGROUND OF THE STUDY
- APPLICATION ANALYSIS OF ARTIFICIAL INTELLIGENCE IN POWER SYSTEM FAULT DIAGNOSIS
- FAULT DETECTION
- MACHINE LEARNING TECHNIQUES FOR FAULT CLASSIFICATION
- FAULT DETECTION AND DIAGNOSIS CHALLENGES
- FAULT DETECTION AND DIAGNOSIS BENEFITS
- FAULT DETECTION AND DIAGNOSIS TRENDS
- CONCLUSION
REFERENCES
CHAPTER ONE
1.0 INTRODUCTION
1.1 Background of the study
To prevent people and property from damage or injury, electrical faults in a power system must be cleared fast. In the early days of electrical power systems the fault clearing was administered by the maintenance staff, who visually detected the fault and manually operated a switch to clear the fault. As fault currents became larger and the operating requirements of the electric power system became more stringent, the need for automatic fault clearance became a necessity (Talent, 2017).
The power transmission network is the most vital link in the country’s energy system as it carries massive quantities of power from generators to substations at high voltages. The modern power system is a complex network that demands a high-speed, accurate, and reliable system of protection. Faults in the power system are inevitable, and there are usually higher overhead transmission line failures connected to other major components. Not only do they influence the system’s reliability, but they also have a widespread effect on end-users. Additionally, as the configurations become more complex, the complexity of protecting transmission line configurations increases, predicting faults (type and location) with considerable accuracy, therefore, improves the power system’s operational reliability and stability and helps prevent colossal power failure. Faults in power systems may arise due to various reasons however these faults must be predicted and diagnosed as early as possible if not it may sometimes lead to the blackout of the entire systems following which it affects the customer even though a lot of necessary protection devices are employed in the detection of faults. Still, it is necessary to predict the faults in advance to overcome the above-said problems. Digital technology was introduced with the introduction of a smart grid enabling the installation of sensors along the transmission lines that can capture live fault data because they present useful data that can be used to detect disruptions in transmission lines (Talent, 2017). A considerable amount of heterogeneous data continuously collected by the growing number of distributed low-cost and high-quality sensors, such as Remote Terminal Units, Phasor Measurement Units, and smart meters, along with those generated by other measuring devices is required for the operational control and performance analysis of smart grids. Conventional time-domain techniques are inefficient in computational terms and may not meet real-time application specifications (Jawad, 2023). Most researchers believe that the approach of machine learning (ML) such as artificial neural networks (ANNs), decision trees (DTs), deep learning models, etc. is capable of providing interesting information on safety in power systems (Jawad, 2023). This seminar analyzes the scientific literature and summarizes the most relevant approaches that can be applied in power transmission systems to fault identification and diagnosis methodologies.
1.2 APPLICATION ANALYSIS OF ARTIFICIAL INTELLIGENCE IN POWER SYSTEM FAULT DIAGNOSIS:
With the advent of artificial intelligence (AI) techniques, new approaches have emerged to enhance fault detection capabilities. This analysis aims to provide a comprehensive overview of fault detection in power systems using AI Power system fault diagnosis is a critical aspect of ensuring the reliable and safe operation of electrical power systems. It involves the identification, analysis, and mitigation of various types of faults that can occur in the system, such as short circuits, ground faults, and other abnormal conditions. Effective fault diagnosis is essential for minimizing downtime, preventing equipment damage, and ensuring the continuity of power supply to customers. One of the primary methods for fault diagnosis in power systems is the analysis of voltage and current signals obtained from various points in the system. These signals are typically acquired through specialized measurement devices, such as digital fault recorders, intelligent electronic devices (IEDs), and phasor measurement units (PMUs). The acquired data is then processed and analyzed using various techniques, including (Bouchiba et al., 2021):
- Fault location algorithms: These algorithms utilize the measured voltage and current signals, along with the system’s topological
information, to determine the location of the fault within the power system. Common fault location techniques include impedance-based methods, travelling wave-based methods, and artificial intelligence (AI) techniques like neural networks and fuzzy logic.
- Fault type identification: By analyzing the characteristics of the voltage and current waveforms, fault type identification algorithms can classify the nature of the fault, such as single-phase-to-ground, phase-to-phase, or three-phase faults. This information is crucial for implementing appropriate protection and mitigation strategies.
- Fault cause analysis: In addition to locating and identifying the fault, it is essential to determine the underlying cause of the fault. This analysis may involve examining factors such as equipment failure, environmental conditions, or external events that could have triggered the
- Advanced signal processing techniques: Modern fault diagnosis systems often employ advanced signal processing techniques, such as wavelet analysis, Fourier analysis, and independent component analysis (ICA), to extract valuable information from the measured signals. These techniques can help in noise reduction, feature extraction, and pattern recognition, enhancing the accuracy and robustness of the fault diagnosis process (Bouchiba et al., 2021).
1.3 FAULT DETECTION
The techniques for detecting a faults and classifying them make use of changes in current and voltage signals in case of fault. Techniques range from hand-coded expert-defined rules based on certain thresholds to artificial intelligence-based techniques such as ANNs, vector supporting machines, and blurred decision systems. The methods vary from hand-coded and expert-defined rules based on certain thresholds to artificial intelligence-based techniques, such as support vector machines, fuzzy decision systems, ANNs (Song et al., 2022). Several characteristics and signal transformations were suggested and used for detection purposes, such as Fourier and wavelet transformations. While protection of critical lines and system buses is ensured with local protection equipment such as relays and circuit breakers, the data made available by PMUs offer the potential to increase understanding and situational awareness in a power management center as also suggested using the output of a PMU-only state estimator for detection and classification of faults. In this context, the approaches use decision trees, and employs support vector machines for this purpose. Such methods presume, as discussed above, the complete presence of all the measurements in full synchronization, given the promising results provided in these works. In the scope of this work, we have experimented with two fault detectors for the output of a PMU-only state estimator: one based on ANN and the other based on support vector machines. Because of the observed superior performance of ANN and space limitations, we restrict our discussion and findings with ANNs in the following to detect and identify faults. Further work is ongoing for a comparison of different machine learning-based techniques for power system fault detection and classification (Song et al., 2022).
1.4 MACHINE LEARNING TECHNIQUES FOR FAULT CLASSIFICATION:
Classification is a crucial step in fault prediction. The list of various classification algorithms that were widely used in the literature of decision support systems presented for different domains and have been used to develop our classifiers are
1.4.1 Support Vector Machine
A knowledge-based method is provided using support vector machines (SVMs) for ready post-fault diagnosis. SVMs are used as an intelligence tool to identify the faulty line arising from the substation and to find the distance from it. SVMs are also compared in datasets with radial-based neural networks that correspond to different transmission system faults. For post-fault evaluation of any relay mal-operation (a faculty or incorrect operation) following a disruption in the adjacent line connected to the identical substation, the approach is particularly important. This can help improve the process of fault monitoring/diagnosis, thereby ensuring secure power systems operation (Hajiyev et al., 2020). In this research, a single vector support machine is used to identify ten types of shunt faults and fault location regression model, which removes manual work.
1.4.2 Bayesian Learner (Naïve Bayes)
Bayesian classifiers are statistical classifiers that use supervised methods of learning to predict the probability of class membership. Bayesian classification is based on the theorem of Bayes, which offers practical learning algorithms that combine prior knowledge with observed data. The Bayesian theory of learning is a probabilistic model of learning. It is applied to decision-making and inferential statistics dealing with the inference of probability. Due to the mutual coupling between circuits, parallel transmission lines are difficult to protect. Fault detection and classification techniques based on the Naïve Bayes classifier may be used to secure a parallel transmission line with inter circuit faults. This is a suitable classification method for more massive data sets, as it takes less time and higher accuracy for the training process (Hajiyev et al., 2020).
1.4.5 Sequential Minimal Optimization (SMO)
Training a support vector machine needs the solution of the sizeable quadratic programming optimization problem. SMO splits this major quadratic programming problem into a series of minor quadratic programming problems. Such small quadratic programming problems are analytically solved, which prevents using a time-consuming optimization of numerical quadratic programming as an inner loop. Since matrix calculation is bypassed, SMO scales in training set size for different test problems around linear and quadratic. The standard chunking SVM algorithm scales in the defined size of the
learning between linear and cubic. The calculation time of SMO is determined by SVM analysis, which makes SMO the fastest for linear SVMs and sparse data sets. SMO can be stronger at least a thousand times quicker than the chunking algorithm in real-world sparse information sets (Hu et al., 2020).
1.4.6 Logistic Regression
In case the dependent variable is dichotomous (binary), logistic regression is the appropriate regression analysis to perform. Logistic regression is a statistical method, like all regression analyses. Logistic regression is rarely used in the diagnosis of power distribution failure, while the neural network has been widely used in reliability research on power systems. In logistic regression, the dependent variable Y with interested outcome values are 1 and 0.
Log it(Y=1)=ln[P(Y=1)/P(Y=0)]=α+βX (1)
Α and β are unknown parameters to be estimated/defined using the training data using the maximum likelihood method (Petersen, 2019). P(Y=1)=1/1+e−(α+βX) (2)Finally, by comparing the measured probability with the predefined threshold, the class label is applied to that test case.
1.4.7 Decision Tree (DT)
The design of the decision tree is straightforward, and we can easily follow a tree structure to explain how to make a decision. The fault type is recognized utilizing a decision-tree algorithm (DT) (Liu et al., 2018). DT may be the most advanced technology to divide sample data into a collection of decision rules. Decision trees are often referred to as category trees for classification problems. The innumerable extent of power systems DT has recently been found to be highly successful in applications such as online dynamic safety evaluation, transient stability, and islanding detection. DT can identify and recognize transmission line failures reliably. It is applied in the power transmission network for fault detection. This defines the exact starting time of the fault with the moving waves triggered by the detector of fault and fault. Information from one side of the protected line is required for this process, and decision-making was carried out in just 2 ms, which is the best time of earlier approaches.
1.4.8 K nearest neighbor (KNN)
The K-nearest Neighbors algorithm is a secure, supervised algorithm for machine learning which can be used to solve problems of classification and regression. In these methods, the time of error occurrence and the defective phases are determined by calculating the interval separating each sample and its nearest neighbor in a pre-default frame. The maximum distance value is compared with predefined threshold values for detection and classification procedures. Simplicity, low calculation pressure, reasonable precision, and speed are the key advantages of these methods (Zhu et al., 2015).
1.5 FAULT DETECTION AND DIAGNOSIS CHALLENGES
Fault detection and diagnosis methods in power systems are complex and diverse, presenting several challenges. Power systems are large-scale, nonlinear, and dynamic systems with multiple components, interactions, and uncertainties, making it difficult to model, measure, and analyze fault behavior and effects accurately and comprehensively. Additionally, there are different types and configurations of faults that require different detection and diagnosis methods to handle their diversity and specificity. Furthermore, power systems generate huge amounts of data from various sources that may be noisy, incomplete, inconsistent, or delayed, which affects the performance and accuracy of the detection and diagnosis methods.
To overcome challenges in power system fault detection and diagnosis, several effective methods can be employed. Wavelet transform and Fourier transform help analyze transient signals and detect fault patterns, offering accuracy in identifying fault locations. Artificial Intelligence (AI) techniques like machine learning and neural networks can handle vast, noisy data, improving prediction and classification accuracy. Phasor Measurement Units (PMUs) provide real-time, synchronized data, enabling faster fault detection. Hybrid models that combine AI and traditional methods improve fault classification. Additionally, digital twin technology and advanced data analytics ensure early fault detection and effective system diagnosis (Zhu et al., 2013).
1.6 FAULT DETECTION AND DIAGNOSIS BENEFITS
Fault detection and diagnosis methods offer several advantages for power system coordination, such as improved reliability and availability of the power supply and service, enhanced security and stability of the power system operation and performance, and increased efficiency and cost-effectiveness of the power system resources and assets. These methods can reduce fault duration, frequency, and impact by enabling faster and more accurate fault isolation, clearance, and restoration. Additionally, they can help prevent or mitigate the fault propagation, escalation, or cascading by providing early warning, protection, and control actions. Finally, they can optimize the fault management, maintenance, and planning by providing useful information, insights, and recommendations.
1.7 FAULT DETECTION AND DIAGNOSIS TRENDS
Fault detection and diagnosis methods are undergoing a transformation as new technologies, standards, and applications are developed in power systems, such as smart grids, big data, and artificial intelligence. Smart grids incorporate advanced technologies and enable bidirectional communication, automation, and interoperability among the power system components and stakeholders. Big data involves the use of cloud computing, data mining, or analytics to process, store, and extract value from the data. Artificial intelligence is the science and engineering of creating intelligent machines or systems that can perform tasks that normally require human intelligence. Each of these developments provide new opportunities and challenges for fault detection and diagnosis methods. For example, smart grids offer the possibility of leveraging distributed generation, storage, and control capabilities. Big data offers the potential to exploit rich information, patterns, and correlations in the data. Artificial intelligence can enhance learning, adaptation, and generalization capabilities. However, each of these developments also present difficulties for fault detection and diagnosis methods such as dealing with increased complexity or overcoming data quality issues (Zhu et al., 2013).
1.8 CONCLUSION
Fault identification and classification are the essential safety functions of transmission lines. Considering applications for machine learning, the complex difficulties of the power system have become more comfortable to handle. Traditional methods are not computationally feasible solutions as they have an inadequate ability to handling massive amounts of data (including bits of different data sets) from units of measurement such as smart meters and units of phasor analysis. Protection of the power system includes the method of identifying and correcting faults until fault currents create damage to utility facilities or property of the consumer. Using powerful machine learning techniques to predict fault could result in improving the power transmission system protection procedures. It will also reduce the time necessary to clear the faults, especially for a long transmission line, thereby increasing the reliability and efficiency of the overall power system.
