Prediction of distribution feeder outages caused by storms

每年秋天和冬天，不列颠哥伦比亚省（BC）的低陆平原都会经历暴风雨和极端天气。严重的风暴可能会吹倒树木和配电馈线，导致数千名客户在几小时到几天内停电，迫使学校关闭，公共交通系统中断，生产线关闭。这个问题在冬天尤其严重，因为在寒冷的天气里，停电意味着没有暖气。由风暴引起的配电馈线中断是由多种因素引起的，包括天气、环境和馈线的物理条件。强风是风暴事件中馈线中断的主要原因，但不是唯一的原因。配电馈线可能因闪电、雪、结冰、大雨和树木倒下而失效。树木倒下与风有关，也与植被的邻近和健康有关。老化的馈线在暴风雨中可能有更高的停电概率。

Every year, the Lower Mainland in British Columbia (BC) will experience storms and extreme weather in fall and winter. Severe storms could knock down trees and distribution feeders, causing power outages for thousands of customers over several hours to several days, forcing school closure, interrupted public traffic systems, and shutting down production lines. This problem is especially serious in winter because power outages mean no heating in cold weather. A distribution feeder outage caused by storms is due to multiple factors, including weather, environment and physical feeder conditions. High wind is a major cause of feeder outage in a storm event but is not a unique one. A distribution feeder can fail due to lightning, snow, icing, heavy rain and tree falling. Tree falling is related to wind and also to the proximity and health of the vegetation. An aged feeder may have a higher outage probability in a storm. The objective of this project is to develop a modeling method and discuss relevant data issues for the prediction of distribution feeder outages ahead of a future storm. Such an advanced prediction will give field managers a lead time from several hours to several days for emergency preparation and response, consequently reducing social, economic and environmental damages of storms. This is a challenging task because of the number of factors involved and the uncertainty of these factors. BC Hydro and SFU will collaborate to apply data mining techniques and statistical methods to solve this problem. The approach is to leverage a combination of historical outage data, weather databases, and feeder information to extract correlations between feeder outages and multiple factors in a quantitative manner. Such correlation is then used in combination with the weather forecast to predict feeder outages prior to the next storm. This project will contribute new knowledge to data mining and statistics research in dealing with unstructured data that are scattered in multiple data sources and are inherently uncertain and noisy.