A Robust Framework for Fault Detection in Smart Grid Power Systems

Thesis Abstract

The increasing complexity and critical importance of smart grid power systems in ensuring reliable and uninterrupted electricity supply necessitate advanced fault detection mechanisms capable of addressing diverse fault conditions with high accuracy and minimal false alarms. Traditional fault detection methods often suffer from limitations such as sensitivity to noise, inability to adapt to dynamic grid conditions, and inadequate identification of complex or synchronized faults, thereby compromising the resilience and operational efficiency of smart grids. This study aims to develop a comprehensive and robust fault detection framework that integrates machine learning algorithms with signal processing techniques to enhance detection accuracy and system reliability. The specific objectives include 1) analyzing the limitations of existing fault detection methods in smart grid environments; 2) designing a hybrid detection model combining wavelet transform for feature extraction with support vector machine (SVM) classifiers for fault identification; 3) optimizing model parameters through grid search and cross-validation; 4) evaluating the framework's performance under various fault scenarios through simulation and empirical testing; and 5) establishing a scalable methodology adaptable to real-time implementation. The research adopts a mixed-methods approach comprising both qualitative and quantitative analyses. The quantitative phase involves the collection of electrical signal data from a simulated smart grid environment modeled using MATLAB/Simulink, with a population of electrical signals representing normal and faulty conditions across 10,000 test cases. Fault scenarios include line-to-ground, line-to-line, and three-phase faults introduced at random intervals and magnitudes, covering both transient and persistent faults. Data acquisition employs high-frequency sampling (100 kHz) using digital oscilloscopes integrated within the simulation setup, ensuring comprehensive signal capture. The qualitative component involves expert interviews with electrical engineers and system operators to contextualize the practical challenges of fault detection. Data analysis employs advanced signal processing techniques, notably wavelet packet decomposition, for time-frequency feature extraction. The extracted features feed into a support vector machine (SVM) classifier trained with a radial basis function kernel. Model performance is assessed via K-fold cross-validation, with evaluation metrics including accuracy, precision, recall, F1 score, and receiver operating characteristic (ROC) curve analysis. Sensitivity and specificity analyses are conducted to ascertain the robustness of the framework under varying noise levels and fault conditions. Additionally, the study applies the Theory of Adaptive Signal Processing to justify the hybrid approach's capability to dynamically adapt to changing grid states and fault signatures. The framework’s computational complexity is analyzed concerning real-time applicability using computational time benchmarks. Expected findings indicate that the proposed hybrid model significantly outperforms traditional threshold-based and standalone signal analysis methods in accurately detecting and classifying faults with an estimated detection accuracy exceeding 98%. The framework demonstrates resilience to electromagnetic interference and system noise, maintaining high detection rates under diverse operational conditions. It is anticipated that the results substantiate the model's validity and effectiveness in enhancing smart grid fault management, thereby bridging crucial gaps identified in previous studies. This research contributes novel insights into integrating signal processing techniques with machine learning models tailored for smart grid fault detection, advancing the theoretical understanding of adaptive fault identification frameworks. It also offers a practical, scalable methodology suitable for deployment in real-time monitoring systems, improving grid reliability and reducing downtime. The study’s conclusions recommend further exploration of online learning algorithms for real-time adaptive fault detection, as well as the potential integration of this framework with existing supervisory control and data acquisition (SCADA) systems for comprehensive grid management. The findings serve as a foundation for future research aimed at enhancing the resilience and intelligence of modern power systems in the face of increasing operational complexities.