کنفرانس منطقه ای سیرد

صفحه اصلی / همایش بهداشت حرفه ای، ایمنی و محیط زیست (بهام) منطقه شمال غرب کشور

Random-Forest-Based Framework for Outage Prediction in Low- and Medium-Voltage Power Distribution Networks

نویسندگان :

Vahid Sahebalzamani¹ Ali Sohrabi Bidar² Hamid Tahmasebi Khou³ Majid Sadeghi⁴

1- دانشگاه شهید بهشتی 2- دانشگاه صنعتی خواجه نصیرالدین طوسی 3- دانشگاه صنعتی خواجه نصیرالدین طوسی 4- شاهد تهران

کلمات کلیدی :

distribution network outage prediction،machine learning،Random Forest

چکیده :

Power distribution networks face increasing challenges in reliability, driven by aging infrastructure, more frequent extreme weather, and higher customer expectations for uninterrupted service. This study examines the application of supervised machine-learning techniques to predict imminent outages in low- and medium-voltage distribution feeders. A feature set is proposed, including meteorological data (e.g., wind speed, precipitation), equipment telemetry (e.g., transformer temperature, load), and short-term historical outage counts. A Random Forest classifier is trained on an hourly time-series dataset using a time-based train/test split to preserve real-world temporal ordering. Performance is evaluated with AUC-ROC, precision/recall, and a thresholded decision scheme for operational dispatch. Results on a synthetic but realistic dataset demonstrate meaningful discrimination (AUC ≈ 0.75) and show how probabilistic outputs can support risk-based prioritisation of feeders. Feature-importance analysis highlights the influence of wind speed, precipitation occurrence, transformer thermal load and past outage history. The framework is designed for ease of integration into utility operations, facilitating proactive crew dispatch, inspection planning and grid hardening. Future work will apply the method to real feeder datasets and explore calibration of probabilities and cost-based threshold optimization.