Measurement of quality parameters related to power outages i̇n the electricity distribution network via machine learning: Performance assessment for Çinar and Ergani districts of Diyarbakir province

The reliable, high quality, and uninterrupted transmission of electrical energy from the point of generation to the point of consumption is essential. In the distribution network section, which is located, just before electricity reaches the consumer, various faults occur due to miscellaneous reasons. Among these reasons, electrical, mechanical, atmospheric, external interventions, completion of economic life, and incorrect maneuvers stand out. Interruptions in electrical power supply due to faults can disrupt communication, transportation, and security systems; they can also lead to unwanted problems and social incidents in critical facilities such as hospitals, airports, and border posts where electricity is crucial. Proper understanding of the causes of power outages can lead to reduced adverse effects through accurate analysis. A significant portion of the electrical distribution network providing electricity to consumers in the Çınar and Ergani districts of Diyarbakır province consists of overhead lines. The distribution network composed of overhead lines is directly affected by weather conditions, thereby directly influencing the number and duration of faults. Therefore, knowing to what extent electrical faults are affected by weather conditions is important for a healthy performance assessment. In this study, the predictability of supply continuity quality parameters in electrical distribution network faults in the Çınar and Ergani districts of Diyarbakır province based on weather conditions is investigated using machine learning models. Machine learning models are trained using features such as daily maximum temperature, minimum temperature, temperature difference, precipitation amount, and wind speed. The data utilized in this study were trained based on deep learning models such as Long Short-Term Memory (LSTM) and decision tree models including XGBoost, CatBoost, LightGBM, and Random Forest. In the study, it is found that the CatBoost model is more successful compared to other methods. The average accuracy performances achieved by the CatBoost model for the three strategies were observed to be 75.1%, 80.07%, and 78.49%, respectively. Through the CatBoost model, the sensitivity of the electrical distribution network to weather conditions can be measured. With this proposed model, the aim is to calculate performance scores more accurately, identify which weather conditions are more sensitive to power outages, and determine more effective maintenance, repair, and investment strategies based on this identification. Additionally, detecting increases in fault numbers and durations independent of weather conditions is also aimed.