Faults Detection in Power System Grids using Isolation Forests

The study examined faults detection in power system grids using Isolation Forests. In carrying out the study, a descriptive survey research design was adopted. The study was carried out in Akwa Ibom State. The target population of the study comprised all the licensed Electrical Engineers in Akwa Ibom State. A stratified sampling technique was used to select licensed Electrical Engineers from the three senatorial districts in Akwa Ibom State, represented by Uyo, Ikot Ekpene and Eket, giving a total sample size of 180 respondents for the study. The instrument used for data collection was a structured questionnaire entitled “Fault Detection in Power System Grids and Isolation Forests Questionnaire” (FDPSCIFQ). The face and content validity of the instrument were established by experts in Electrical Engineering to ensure the accuracy, appropriateness, clarity and completeness of the items in relation to the objectives of the study. The reliability of the instrument was determined and a reliability coefficient of 0.81 was obtained. This coefficient was considered sufficiently high enough to justify the use of the instrument for the study. The data generated from the study were analyzed using appropriate descriptive statistical techniques to answer the research questions. The results showed that Single Line-to-Ground Fault (L-G) was the major type of fault commonly encountered in power system grids, accounting for 23.89%, while Three-Phase-to-Ground Fault (L-L-L-G) recorded the least percentage of 16.67%. The results also showed that Extended Isolation Forest (EIF) was the major type of Isolation Forest identified as useful for detecting different forms of anomalies in datasets, with 24.44%, while Deep Isolation Forest (Deep IF) recorded the least percentage of 15.56%. Furthermore, Axis-Parallel Splitting Artifacts and Feature Space Constraints constituted the major challenge associated with the use of Isolation Forests for fault detection, with 28.33%, while High False-Positive Rates in High-Noise Sensor Environments recorded the least percentage of 21.67%. Regarding mitigating strategies, Extended Isolation Forest (EIF) with Random Hyperplanes recorded the highest percentage of 23.89%, while Physics-Informed Feature Engineering and SHAP Explanations recorded the least percentage of 15.00%. The study concluded that Isolation Forests offer a promising intelligent approach for improving the speed, flexibility, and effectiveness of fault detection in modern power system grids and can contribute to the development of more reliable, resilient, and intelligent electrical power networks. One of the recommendations made was that Power system operators should integrate Isolation Forests with existing protection systems to provide an additional intelligent layer for detecting abnormal electrical conditions and identifying potential faults before they develop into more serious grid disturbances.