Power system protection is essential for maintaining the reliability and stability of electrical grids, ensuring continuous service, and preventing catastrophic failures. As power systems evolve to incorporate renewable energy and increasingly complex configurations, the role of Artificial Intelligence (AI) in enhancing protection mechanisms has become indispensable. This paper reviews the integration of AI in power system protection, highlighting its potential to improve fault detection, adaptive protection strategies, predictive maintenance, and real-time monitoring. AI techniques, including machine learning, deep learning, and expert systems, offer significant advancements in overcoming the limitations of traditional protection schemes. Furthermore, the integration of AI contributes to the development of resilient and sustainable infrastructure, supports innovation in intelligent urban systems, and enhances the reliability of modern power grids. Despite its promising potential, challenges such as data scarcity, model scalability, and real-time processing need to be addressed for effective implementation. This review synthesizes the current literature on AI applications in power system protection, comparing them with conventional methods, and provides information on future research directions and practical applications to improve energy reliability, sustainable urban development, and industrial innovation.

Determination of relay pairs from breakpoints in a given network is essential for maintaining the current protection system. Pairs of relays as primary or backup maintain the operation of the protection scheme within its zone of protection in tandem. All calculations and assumptions that are made in protection systems are based on breakpoints. It is inadequately documented that machine learning can be used to determine breakpoint sets and relay pairs. This paper presents the implementation of supervised decision tree machine learning approach for determining directional overcurrent relay breakpoint set in 3-bus networks. Using the one-hot encoding method, 45 input features are extracted from a matrix derived from 3-bus, 5-line network data. Bayesian optimization is used to further optimize the hyperparameters of each model for each of the break point set outputs. Tree diagrams are also provided here to assist in the interpretation of the decision rule resulting from the regression analysis. Experiment tests indicated that the proposed method shows promising results in determining breakpoint set in terms of RMSE.

The scope of this study folds out the terms of electrical power output generated by the combustion of methane gas in the production of the portable anaerobic digester in kilo Watthour. When the gas flow meter indicated that there is sufficient gas to enter the generator, gas chromatography will again be conducted this time to ensure that the entering gas is really methane. Since there is small load to be powered and only 0.3 m3 of methane that would enter the generator, 1 kW generator would be used to yield sufficient energy which is equivalent to theoretical value of 4.5811 kWh (usable energy) as further explained.