Machine Learning Applications in Power Systems Operation and Management

The transformation of power systems is increasingly driven by the need for smarter, more resilient, and efficient energy infrastructures. With growing energy demands and the shift toward decentralised generation, power systems are evolving to integrate advanced optimisation techniques and digital technologies. These innovations enable real-time monitoring, predictive maintenance, and dynamic load balancing, which are essential for enhancing grid reliability and performance. Optimisation strategies, such as demand side management, Optimal Power Flow (OPF), and distributed generation coordination, help minimise energy losses, reduce operational costs, and maximise the use of available resources. As a result, modern power systems are becoming more adaptive and capable of meeting sustainability goals while maintaining high standards of stability and efficiency. Machine learning (ML) plays a vital role in the transformation and optimisation of modern power systems by enabling data driven decision making, predictive analytics, and adaptive control. With the vast amount of real time data generated by smart grids, ML algorithms can forecast electricity demand, renewable energy generation, and equipment failures with high accuracy. These predictive capabilities allow grid operators to optimise power flow, balance supply and demand more effectively, and prevent outages before they occur. Overall, ML empowers power systems to become more intelligent, autonomous, and resilient in the face of increasing complexity and renewable integration. The first aspect of this research deals with the comparative analysis of traditional power flow algorithms such as Newton Raphson and Gauss Seidel with ML based models using python. This study involves implementing classical numerical techniques to establish accurate baseline results for voltage magnitudes, angles, and power losses across the system. These outcomes are then systematically compared with the predictions generated by various ML models trained on synthetic power system data. The goal is to evaluate the performance, accuracy, and computational efficiency of ML approaches in approximating or enhancing power flow studies. The findings reveal that the ML algorithms not only maintain high accuracy but also significantly improve computational efficiency, achieving up to a tenfold reduction in both time and space complexity compared to traditional methods. Building on this foundation, this research introduces a set of deterministic OPF programs that can handle base case as well as changes in power demand over time and unexpected failures in the grid. A fully automated OPF program is developed which provides optimal operating conditions for efficient and stable power flow. To address different operational conditions, a multiperiod OPF program is developed that uses load profiles across several time intervals, giving detailed insights into how voltage and generation change over time. Additionally, a Security Constrained Optimal Power Flow (SCOPF) program is proposed to ensure the grid remains secure against predefined contingencies (N-1 failures), enhancing reliability under uncertainty. The proposed program, using the IPOPT solver, solved the OPF in 2.24 seconds, while the industry-standard Pandapower library took 2.72 seconds, demonstrating that the proposed algorithm is faster. A key aspect of this research is the development of a novel SCOPF framework that integrates Explainable Artificial Intelligence (XAI) to enhance decision-making transparency and operational reliability. Traditional machine learning based SCOPF models which, while capable of delivering high accuracy and computational efficiency, often operate as black boxes producing results without offering insight into the underlying reasoning. In contrast, this research leverages ML models enhanced by XAI tools to not only optimise the optimal power flow under security constraints, but also to provide interpretable justifications for critical contingencies. This allows operators to understand the reasoning behind model outputs, detect potential risks, and build trust in ML assisted grid management. The proposed XAI-SCOPF framework thus represents a significant step toward more transparent, adaptive, and resilient power system optimisation in the era of smart grids and renewable variability. The results show that the most vulnerable lines suggested by the XAI models along with the rationale for their selection substantially reduce the time and space complexity of the SCOPF framework by five times while maintaining optimal performance. System analysis confirms that these lines cause the largest voltage drops across the nodes, validating their classification as the most vulnerable and reinforcing the credibility of the XAI-driven recommendations, making the approach a viable step toward trustworthy AI assisted power system operation.