Development and application of reduced‐order neural network model based on proper orthogonal decomposition for <scp>BOD</scp>5 monitoring: Active and online prediction

Abstract

This study aims at providing a reduced‐order neural network model (RONNM) based on proper orthogonal decomposition (POD) for online prediction of the 5 days biochemical oxygen demand (BOD 5 ). To achieve this goal, the POD method is used to reduce the input vectors to neural network (NN) model. Evaluation of the selected RONNM and its comparison with the model fed with all input vectors (NN model) shows that it reduces not only the output error but also computational cost. Results depict that Pearson correlation coefficient (R) and root mean square error for the best fitted RONNM are 0.94 and 7.75, respectively. Furthermore, accuracy analysis of the output results of models based on developed discrepancy ratio statistic reveals that RONNM is superior. Thus, developed methodology can be considered as an effective tool to train the water resources managers for online evaluation of water quality related to BOD 5 . © 2011 American Institute of Chemical Engineers Environ Prog, 32: 120–127, 2013.