Natural Gas Forecasting

From MSCS Research Experiences for Undergraduates Wiki
Jump to: navigation, search

How much natural gas will customers of WE Energies need each day for the next week?

Residential, industrial, and commercial customers use natural gas for space heating, cooking and water heating, as feedstock in industrial processes, as fuel for industrial and commercial operations, and hundreds of other purposes. Theoretically, heat replaced is lost through convection and conduction, with delays, so one could do first-principles architectural and thermodynamic modeling of buildings. Viewed as a function of temperature (adjusted for wind, humidity, and other weather conditions), consumption is a very noisy signal.

Data availability is critical for any modeling activity. GasDay has over 500,000 days worth of data on weather and consumption for operating areas across the US and over 7 M days worth of weather and consumption data for individual customers. Current models are multiple regression and neural networks, combined using ensemble forecasting techniques. Outlier detection and removal is critical. The application seems appropriate for Bayesian techniques.

GasDay licenses software to 25 local distribution companies. Each day, we help forecast about 20% of the natural gas delivered to residential, industrial, and commercial customers in the US. We use modern software architecture and tools, including databases, multi-tiered systems, distributed computing, automated testing, and user interface design.

We offer research opportunities in mathematical modeling, statistical analysis of data, and software engineering. Previous REU participants' projects have involved water demand forecasting, high performance computing, data mining, and hourly natural gas demand forecasting. If you work with GasDay, your interests and skills will be matched with a suitable project, you will have plentiful attention from your mentor(s), and you will enjoy working in an active community of scholars.

Personal tools