added 9 research items
Our goal is to rank the electricity service structures (manholes and service boxes) in Manhattan according to their vulnerability to serious manhole events such as manhole fires, explosions and smoking manholes. Manhole event prediction is a new ap-plication for machine learning, and we started this project with a large amount of data from several disparate sources, most of which is extremely noisy. Our data includes Con Edison ECS (Emergency Control System) trouble tickets, which are records of past events affecting the secondary (low voltage) electric distribution system such as a manhole fire, manhole explosion, smoking manhole, no-light event, low voltage event, flickering light event, side-off partial outage, or burnout. We describe our approach to dealing with this challenging dataset, involving a combination of targeted information extraction, statistics, and feature and label development for machine learning. Our ef-forts have resulted in a streamlined bipartite ranking model that is aimed at predicting future manhole events. This model has demonstrated promising results on a "blind" prediction test. The overall goal of our system is to assist Con Edison to prioritize the inspection of over 50,000 manholes and service boxes in Manhattan and to prioritize long-term follow-up repair works.
We present a visualization framework for analyzing the Consolidated Edison Company of New York (Con Edison) trouble tickets for the Manhattan electrical distribution system. The Con Edison Emergency Control System (ECS) is a work management tool that documents all events that occur in the electrical distribution system. The trouble ticket generated from ECS is a record of an event affecting the secondary (low-voltage) electrical distribution system, such as a manhole fire, manhole explosion, smoking manhole, no-light event, flickering light event, side-off partial outage, or burnout. The visualization tool outlined here is used alongside our preliminary statistical and machine learning work for predicting future manhole events. ECS tickets stored in our PostgreSQL database are displayed using Google Earth's satellite images of Manhattan as a backdrop. The ability of this tool to display events relative to the surrounding buildings has already yielded some highly promising directions for our ongoing analysis.
We present a knowledge discovery and data mining process developed as part of the Columbia/Con Edison project on manhole event prediction. This process can assist with real-world prioritization problems that involve raw data in the form of noisy documents requiring signicant amounts of pre-processing. The documents are linked to a set of instances to be ranked according to prediction criteria. In the case of manhole event prediction, which is a new application for machine learning, the goal is to rank the electrical grid structures in Manhattan (manholes and service boxes) according to their vulnerability to serious manhole events such as res, explosions and smoking manholes. Our ranking results are currently being used to help prioritize repair work on the Manhattan electrical grid.