A generalised approach to process state estimation using hybrid artificial neural network/mechanistic models

Abstract In this work, a hybrid model which combines mechanistic elements with Artificial Neural Networks (ANNs) is used as a basis for a generalised on-line state estimation technique. Balance equations, which are more clearly defined, constitute the mechanistic side of the model whilst the ANN element is exclusively applied to modelling the more complex non-linear rate relationships present. The black-box approach offered by ANNs avoids the call made on both manpower and laboratory resources in modelling such complex system features mechanistically. The standard extended Kalman filter algorithm is modified to accommodate the hybrid model and, along with the stochasitic process and measurement noises, handles intrinsic ANN error explicitly. Application of the approach is demonstrated in a simulation case study based on a pilot scale process involving three tanks in series. Results demonstrate the effectiveness of both the estimation scheme and an inferential estimate-based control scheme for the system.

[1]  Rimvydas Simutis,et al.  Hybrid modelling of yeast production processes – combination of a priori knowledge on different levels of sophistication , 1994 .

[2]  Kurt Hornik,et al.  Multilayer feedforward networks are universal approximators , 1989, Neural Networks.

[3]  Martin A. Riedmiller,et al.  Advanced supervised learning in multi-layer perceptrons — From backpropagation to adaptive learning algorithms , 1994 .

[4]  Lyle H. Ungar,et al.  A hybrid neural network‐first principles approach to process modeling , 1992 .

[5]  J. A. Wilson,et al.  Monitoring bioprocesses using hybrid models and an extended Kalman filter , 1996 .

[6]  J. W. Ponton,et al.  Alternatives to neural networks for inferential measurement , 1993 .

[7]  George Cybenko,et al.  Approximation by superpositions of a sigmoidal function , 1992, Math. Control. Signals Syst..

[8]  Richard P. Lippmann,et al.  An introduction to computing with neural nets , 1987 .

[9]  P. A. Minderman,et al.  INTEGRATING NEURAL NETWORKS WITH FIRST PRINCIPLES MODELS FOR DYNAMIC MODELING , 1992 .

[10]  Barry J. Wythoff,et al.  Backpropagation neural networks , 1993 .

[11]  Mark A. Kramer,et al.  Improvement of the backpropagation algorithm for training neural networks , 1990 .

[12]  Zorzetto Lfm.  Bioprocess monitoring with hybrid neural network/mechanistic model based state estimators. , 1995 .

[13]  Markus A. Reuter,et al.  A generalized neural-net kinetic rate equation , 1993 .

[14]  A. Jazwinski.  Stochastic Processes and Filtering Theory , 1970 .

[15]  Rimvydas Simutis,et al.  Bioprocess optimization and control: Application of hybrid modelling , 1994 .

[16]  J. W. Ponton.  Neural networks: some questions and answers , 1992 .