Ferroresonance analysis in electric power systems artificial intelligence based detection and matlab simulation

Ferroresonance is a complex and potentially hazardous phenomenon that can occur in high-voltage electrical power systems, often leading to significant equipment damage, prolonged outages, and safety risks. It is typically associated with the interaction between nonlinear components of the system, particularly transformers, capacitors, and circuit breakers, when the system's components resonate at certain frequencies. Ferroresonance is triggered by specific conditions such as sudden load changes, faults, or switching operations, leading to abnormal voltage and current oscillations. These oscillations can cause severe damage to electrical equipment, disrupt the power supply, and lead to financial losses due to repair costs, downtime, and system failure. The unpredictability of ferroresonance poses a challenge to traditional protection systems, making it difficult to detect and prevent in real-time. The effects of ferroresonance manifest as high voltage spikes and distorted current waves that can cause permanent damage to key electrical components, including transformers and capacitors. These effects are often unpredictable and can last for extended periods, further exacerbating the risks to the power grid. When ferroresonance occurs, it can result in catastrophic consequences such as insulation failures, overheating of equipment, and even catastrophic failures of transformers and other critical infrastructure. The damaging electrical surges that result from ferroresonance can also propagate throughout the system, affecting nearby substations and potentially leading to widespread power outages. Such events can cause power loss, equipment failure, long recovery periods, and significant financial losses. As a result, accurate detection and prevention of ferroresonance are crucial for the reliable operation of power systems. The development of advanced detection techniques can help mitigate the impact of ferroresonance and improve the resilience of power networks. Traditional methods for detecting ferroresonance typically rely on techniques such as frequency analysis, time-domain simulations, and system modeling. However, these methods often fall short in terms of real-time detection, as they tend to have high computational requirements and lack the flexibility needed to account for the dynamic nature of power systems. Furthermore, conventional approaches may not provide timely alerts, leading to delays in intervention and further damage to equipment. In contrast, artificial intelligence (AI)-based techniques, such as machine learning (ML), deep learning, and time-series analysis, offer promising alternatives for more accurate and faster detection of ferroresonance. AI-based methods can analyze patterns and anomalies in system data to predict the onset of ferroresonance and provide early warnings for potential failures. These approaches not only improve detection speed but also reduce costs and downtime by enabling quicker responses to incidents. By xxiv automating the detection process, AI-based monitoring systems reduce reliance on manual interventions and improve the overall efficiency and reliability of power systems. This study explores the use of three different AI-based techniques for detecting and analyzing ferroresonance events: K-Nearest Neighbors (KNN), Support Vector Machines (SVM), and Long Short-Term Memory (LSTM) networks. These techniques were chosen for their ability to address different aspects of ferroresonance detection and their suitability for time-series data. The goal of the study is to evaluate the performance of each machine learning model in detecting ferroresonance events based on real-time voltage and current measurements. The dataset used in this study consists of simulated power system data representing both normal and ferroresonance conditions, with the data split into training and testing sets for each model. The use of simulated data allows for controlled experiments where the models can be trained and tested under various ferroresonance scenarios, ensuring a comprehensive evaluation of their effectiveness. Additionally, data augmentation techniques were applied to increase the robustness of the models and enhance their ability to generalize across different scenarios. KNN, a simple yet powerful classification algorithm, was selected as the first classifier due to its effectiveness on small to medium-sized datasets. KNN classifies data points based on their proximity to nearby data points, making it well-suited for detecting patterns in power system data. However, KNN's computational complexity increases as the dataset size grows, potentially limiting its scalability for large datasets. SVM, on the other hand, was chosen for its ability to classify nonlinear data and its proven accuracy in handling complex classification tasks. SVM creates hyperplanes that separate data points into distinct classes and is particularly effective at dealing with intricate decision boundaries. It performs well even with complex ferroresonance patterns, where nonlinearities are present. LSTM was employed to model the dynamic and sequential nature of ferroresonance events, particularly for time-series data. LSTM networks are capable of learning long-term dependencies in sequential data, making them ideal for capturing the temporal aspects of ferroresonance events and detecting early warning signs of potential failures. The performance of each model was evaluated using various metrics, including accuracy, precision, recall, and F1-score. The results showed that KNN performed well on smaller datasets, achieving high accuracy rates in identifying ferroresonance events. However, as the dataset size increased, KNN's computational complexity became a limitation, reducing its efficiency. SVM, by contrast, demonstrated superior performance on larger, more complex datasets, with high accuracy and robust classification capabilities. SVM's ability to handle nonlinear decision boundaries allowed it to classify ferroresonance events accurately even in challenging scenarios. LSTM networks, however, outperformed both KNN and SVM in detecting ferroresonance events in time-series data. LSTM was able to capture long-term dependencies and patterns in the data, allowing it to provide early warning signals of ferroresonance events before they became critical. The ability of LSTM to model the temporal dynamics of ferroresonance made it an invaluable tool for real-time monitoring of power systems. These findings confirm the potential of AI-based approaches for improving the detection of ferroresonance in power systems. The results show that AI models, especially LSTM, can detect ferroresonance events more accurately and efficiently than traditional methods, providing valuable real-time alerts for operators. The ability to detect these events early can significantly reduce the risk of equipment damage and system outages. Future research could focus on expanding the dataset to include realworld measurements and exploring additional AI models to further enhance detection accuracy. As AI techniques continue to evolve, they are expected to play a critical role in reducing the impact of ferroresonance events on power systems, improving overall system reliability, and enabling more proactive maintenance strategies. The integration of AI-based monitoring systems into power grids will automate the detection process, reducing the reliance on manual interventions and allowing for faster response times to potential ferroresonance events. Furthermore, these systems can continuously learn from new data, improving their performance over time. As the field of AI progresses, more sophisticated models and techniques will emerge, further enhancing the ability to detect and mitigate ferroresonance, ultimately leading to more resilient and efficient power systems.