06/01881 On-line building energy prediction using adaptive artificial neural networks: Yang, J. et al. Energy and Buildings, 2005, 37, (12), 1250–1259

Abstract

Multivariate regression calibration using multiple linear regression (MLR), principle component regression (PCR) and partial least squares regression (PLSR) algorithm was performed on 238Pu, 239Pu, 240Pu and 241Pu atom% abundances to predict 242Pu isotopic abundance. The MLR algorithm was found to be the best among these three algorithms. The effect of 238Pu composition on the 242Pu abundance prediction was found to be small but significant especially for achieving high accuracy of <0.5%. PCR and PLSR generated nearly identical results and were inferior to the MLR results. A comparison of MLR results with those obtained by employing seven previously reported empirical methods revealed far superior prediction capability of MLR model. Among the seven empirical models, the best prediction capability was found for Bignan correlation containing 238Pu isotopic data. The study clearly demonstrates that the production of 238Pu and 242Pu has some small correlation and the use of 238Pu in isotopic correlation for 242Pu prediction is important to get accurate results.