Enhancing Smart Grid Resilience through Machine Learning: A Case Study in Urban Electricity Networks

Thesis Abstract

The increasing complexity and interconnectedness of urban electricity networks have heightened the vulnerability of smart grids to disruptions caused by cyber-attacks, equipment failures, and extreme weather events, necessitating advanced strategies for enhancing resilience. This study aims to investigate how machine learning techniques can be employed to improve the reliability and robustness of smart grids within urban environments, specifically focusing on the metropolitan electricity network of Rivertown. The specific objectives are to identify key factors affecting grid resilience, develop predictive models for fault detection and response, evaluate the effectiveness of machine learning algorithms in real-time resilience assessment, and propose an integrated framework for resilience enhancement. The research adopts a mixed-methods approach, combining quantitative analysis through a case study strategy with qualitative insights from stakeholder interviews. The population comprises 250 grid sensor datasets collected over three years from the Rivertown utility company’s smart infrastructure, including data on voltage fluctuations, load variations, fault occurrences, and maintenance logs. A stratified random sampling technique was employed to select 150 representative data samples for analysis. Data collection instruments encompassed sensor logs, system reports, and semi-structured interviews with engineers and operational staff. To ensure validity and reliability, the sensor data were preprocessed for consistency, and the interview protocols underwent pilot testing and expert review. Data analysis involves the application of supervised machine learning models, including Random Forest, Support Vector Machines, and Neural Networks, to develop fault prediction and anomaly detection algorithms. Regression analysis and receiver operating characteristic (ROC) curves are utilized to evaluate model performance, while feature importance measures identify critical resilience indicators. Additionally, thematic analysis interprets qualitative data to contextualize technical findings within operational realities. The analytical framework is grounded in the Resilience Theory and the Fault Tolerance Model, providing theoretical underpinnings for assessing adaptive capacity and system robustness. Expected findings suggest that machine learning models significantly improve early fault detection, allowing proactive responses that minimize downtime and prevent cascading failures. It is anticipated that Neural Networks outperform other models in accuracy, with an average precision of 92%, identifying resilience-critical factors such as load variance thresholds and network topology vulnerabilities. The integration of quantitative predictions with qualitative insights is expected to yield a comprehensive resilience framework that guides operational decision-making and strategic planning. The study contributes new empirical evidence on the utility of machine learning for resilience enhancement in urban smart grids, extending existing models with context-specific insights from Rivertown’s network. Theoretically, it advances resilience scholarship by demonstrating how predictive analytics can augment traditional engineering approaches, grounded in the Resilience Theory and Fault Tolerance Principles. Practically, it offers a scalable, data-driven framework adaptable to similar urban contexts, facilitating smarter responses to disruptions and fostering sustainable urban electricity management. The main conclusion underscores the potential of machine learning as a transformative tool for smart grid resilience, emphasizing the importance of integrating technological advancements with operational strategies. It recommends integrating predictive models into grid management systems, investing in sensor infrastructure, and fostering capacity-building among operational staff. Future research should explore adaptive learning systems that evolve with emerging network complexities and extend these models to incorporate renewable energy sources, contributing to resilient, sustainable urban power systems.