A comparison of the ability of black box and neural network models of ARX structure to represent a fluidized bed anaerobic digestion process

The performance of three black box models which were parameterized and vali dated using data collected from a laboratory scale fluidized bed anaerobic digester, were compared. The models investigated were all ARX (auto regress ive with exogenous input) models, the first being a linear single input sin gle output (SISO) model, the second a linear multi-input multi-output (MIMO ) model and the third a nonlinear neural network based model. The performan ce of the models were compared using correlation analysis of the residuals (one-step-ahead prediction errors) and it was found that the SISO model was the least able to predict the changes in the reactor parameters (bicarbona te alkalinity, gas production rate and % carbon dioxide). The MIMO and neur al models both performed reasonably well. Though the neural model was shown to be superior overall to the MIMO model, the simplicity of the latter sho uld be a consideration in choosing between them. A simulation with an horiz on approaching 48 h was performed using this model and showed that although the absolute values differed significantly, there were encouraging similar ities between the dynamic behavior of the model and that of the fluidized b ed reactor. (C) 1999 Elsevier Science Ltd. All rights reserved.