A Hierarchical State-Space Model for Short-Term Forecasting of Residential Electricity Demand

• Activity Number: 231
• Type: Topic Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Survey Research Methods Section
• Author(s): Jonathan Hosking*+ and Ramesh Natarajan and Soumyadip Ghosh and Shivaram Subramanian and Xiaoxuan Zhang
• Companies: IBM Research Division and IBM Research and IBM Research and IBM Research and IBM Research

Abstract

Electricity transmission and distribution systems are increasingly moving towards the use of a "Smart Grid", a densely instrumented network for electricity delivery that facilitates remote control, automation, and feedback to consumers. This evolution introduces new challenges in modeling and forecasting electricity usage data. We describe a model of the daily load curve for residential electricity consumption that includes the effects of dynamic price incentives on the demand response. The model represents the observed values of the daily load curve by a set of periodic smoothing-spline basis functions, with the coefficients of the basis functions evolving according to a linear Gaussian state-space model that incorporates mean shifts, day-of-week and holiday adjustments, temperature effects, as well as the dynamic price incentive effects. This modeling and forecasting methodology provides the ability to model intraday load substitution effects that are induced by the specified dynamic pricing schedules, the ability to use fine-grained (5 to 15 minute interval) observational data and enables fast updating of model forecasts as new usage and weather data arrive.