Abstract
Conference Title: 2017 Systems and Information Engineering Design Symposium (SIEDS) Conference Start Date: 2017, April 28 Conference End Date: 2017, April 28 Conference Location: Charlottesville, VA, USA The Water Distribution System (WDS) in the District of Columbia is one of the most complex in the world, spanning 1,440 dense urban miles. Providing water to 700,000 residents and 16.6 million annual visitors, the system consists of approximately 40,000 valves, 9,500 fire hydrants and 130,000 service connections. Originally developed in the 1800s, the median age of pipe in the WDS has reached 78 years. Ruptures in water main pipes carrying over 200 psi of pressure can buckle roadways, flood streets and cause widespread service disruptions. In winter months, this can create icy conditions which are extremely hazardous. The District's water utility reports approximately 500 water main breaks per year. Since the implementation of a priority system designed to expedite processing of high-impact emergency failures, the utility's operational data has identified high response times for incidents classified as low priority. The average is approximately 84 days, nearly six times higher than emergency incidents. The scheduling phase of the workflow is responsible for 98 percent of the total delay, indicating the need for a streamlined scheduling methodology. Emergency incident volume increases twofold in winter months, with sporadic failures caused by temperature fluctuations. This has forced the utility to increase reliance on contracted crews by 20 percent, and indicated an urgent need for weather-dependent failure forecasting tools. A queueing model of the utility's workflow was used to derive standard scheduling protocols for groups of low priority incident types. Results indicate that achieving performance goals is feasible without additional resources. Recommended workflow configurations and operational improvements reduced the low priority response time to 12.4 days. Additionally, an Artificial Neural Network (ANN) was trained and deployed to generate main break forecasts for scheduling purposes. Validation tests using the utility's observed main breaks reported R2 = 0.88 and R2 = 0.86 for the weekly and daily models, respectively. The prediction model is coupled to the workflow to predict surges in failure volumes and allow for a proactive workflow response.