Empirical investigation of regression models for predicting system behavior in air handling units

摘要

Modeling of the system behavior is a key step for better management and accurate fault detection and diagnosis of air handling units (AHUs). This paper presents an extensive empirical investigation on a typical AHU. A dataset from an active unit is analyzed using different existing forecasting techniques. Performances of these techniques are compared to identify reliable models for distinct physical processes. To achieve that, linear regression models are calibrated by selecting appropriate sets of features and by incorporating, where appropriate, auto-regressive terms. Based on the outcomes of this study, we recommend specific features resulting from nonlinear combination of measurements for the air mixing process. For the heat exchange process, where the supply air is conditioned to the desired air comfort level, auto-regressive models with exogenous variables are found to be appropriate. A root mean squared error of about 0.10 degrees C for both air mixing and heat exchange processes is estimated for the proposed models.