Building-level power demand forecasting framework using building specific inputs: Development and applications

Abstract

Abstract In this paper, the development of a general framework for building level power demand forecasting and its applications to supervisory control and demand management are presented. Models of thermal loads, while rigorous and insightful, do not directly extrapolate to measures of power consumption and cannot be easily applied to a variety of buildings. Ultimately, building operators are interested in managing power consumption as energy costs and opportunities are directly related to the power variable. Our work develops Auto-Regressive models with eXogeneous inputs (ARX) to forecast power demand in conjunction with existing physics based modeling approaches and enhances the current control framework for building energy management. The main contributions of this work are identifying and incorporating building level measurements as inputs, and evaluating the use of power forecast models for supervisory control and demand response (DR). The move towards a smarter grid is expected to provide extensive data on building conditions and power consumption, which we can include in the model development. Options for model inputs and outputs are investigated depending on possible measurements, and their effect (or sensitivity) on the modeling and decision making processes are evaluated. It is shown that an appropriate selection of exogenous inputs related to the control action is necessary to capture the effect of common demand management practices such as precooling. The forecasting capabilities are also demonstrated on a simplified building model and on data collected from a real building.