Open Access Thesis
Office buildings -- Energy consumption -- Data processing; Office buildings -- Heating and ventilation -- Data processing; Neural networks (Computer science)
Sensitivity analysis and Multiple Linear Regression (MLR) are the most extensively used techniques in studying the input-output relationship in building thermal systems. However, both MLR and most methods for sensitivity analysis do not account for nonlinear components embedded in building energy systems. Thus, their results might be distorted.
In this study, the Artificial Neural Networks (ANN) technique was applied to sensitivity analysis and modeling of an imaginary small office in order to (a) examine how the annual energy consumption responded to 40 building design parameters and evaluate relative contributions of these parameters to the variation of the building energy performance, and (b) develop models to represent the relationship between the annual energy usage with input parameters and then use these models to predict energy consumption. The data used for sensitivity analysis and modeling were generated by DOE-2.1 E simulation program.
Both Differential Sensitivity Analysis (DSA), the most conventional sensitivity analysis method, and ANN techniques were employed to analyze the sensitivity of building annual energy consumption to 40 design parameters. The relative importance of these parameters to the energy usage was ranked by the sensitivity coefficients coming from both DSA and ANN methods.
The relationship between building energy consumption and input parameters was then modeled by both MLR and ANN techniques using the most important 5, 10, or 15 parameters yielded in the above sensitivity analysis experiments. A comparison of the results demonstrated that: 1. ANN models were better than MLR models in predicting energy consumption because the error between DOE-2.1 E simulation and ANN model prediction was smaller than that from MLR models. 2. ANN sensitivity analysis was better than DSA because models developed with ANN-derived important parameters more precisely predicted building energy consumption, implying ANN sensitivity analysis more efficiently evaluated the relative importance of input parameters.
The results of this project illustrated that ANN technique can be adopted to perform sensitivity analysis and develop models to quantify the input-output relationship in building energy systems. The results showed that the ANN method had better performance than both DSA and MLR, which have been extensively used in building thermal system studies.
Year of Submission
Master of Arts
Department of Industrial Technology
1 PDF file (108 pages)
©2002 Quan Tang
Tang, Quan, "Application of artificial neural networks to sensitivity analysis and modeling of small office buildings" (2002). Dissertations and Theses @ UNI. 1268.