Analysis and evaluation of regression model for centrifugal chiller

Abstract

The prediction accuracy of three kinds of centrifugal chiller regression models (Multivariable Polynomial Regression Model, BP-Artificial Neural Network Regression Model and Support Vector Regression Model) is analyzed using ASHRAE 1043-RP data, and the prediction performance in small sample is also discussed. Experimental results show that the Linear Regression Model and Support Vector Regression Model have excellent prediction performance, while BP Neural Network Regression Model has serious over-fitting problem. These results can provide some reference for chiller fault diagnosis model selection.