ANALYSIS OF AUTOREGRESSIVE ENERGY MODELS OF A RESEARCH HOUSE

摘要

Energy consumption from buildings is a major component of the overall energy consumption by end-use sectors in industrialized countries. In the United States of America (USA), the residential sector alone accounts for half of the combined residential and commercial energy consumption. Therefore, efforts toward energy consumption modeling based on statistical and engineering models are in continuous development. Statistical approaches need measured data but not buildings characteristics; engineering approaches need building characteristics but not data, at least when a calibrated model is the goal. Among the statistical models, the linear regression analysis has shown promising results because of its reasonable accuracy and relatively simple implementation when compared to other methods. In addition, when observed or measured data is available, statistical models are a good option to avoid the burden associated with engineering approaches. However, the dynamic behavior of buildings suggests that models accounting for dynamic effects may lead to more effective regression models, which is not possible with standard linear regression analysis. Utilizing lag variables is one method of autoregression that can model the dynamic behavior of energy consumption. The purpose of using lag variables is to account for the thermal energy stored/release from the mass of the building, which affects the response of HVAC equipment to changes in outdoor or weather parameters. In this study, energy consumption and outdoor temperature data from a research house are used to develop autoregressive models of energy consumption during the cooling season with lag variables to account for the dynamics of the house. Models with no lag variable, one lag variable, and two lag variables are compared. To investigate the effect of the time interval on the quality of the models, data intervals of 5 minutes, 15 minutes, and one hour are used to generate the models. The 5 minutes time interval is used because that is the resolution of the acquired data; the 15 minutes time interval is used because it is a common time interval in electric smart meters; and one hour time interval is used because it is the common time interval for energy simulation in buildings. The primary results shows that the use of lag variables greatly improves the accuracy of the models, but a time interval of 5 minutes is too small to avoid the dependence of the energy consumption on operating parameters. All mathematical models and their quality parameters are presented, along with supporting graphical representation as a visual aid to comparing models.