Date of Award

Summer 2001

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Chemical Engineering - (Ph.D.)

Department

Chemical Engineering, Chemistry and Environmental Science

First Advisor

Norman W. Loney

Second Advisor

Basil Baltzis

Third Advisor

Robert Benedict Barat

Fourth Advisor

Dana E. Knox

Fifth Advisor

Denis L. Blackmore

Sixth Advisor

Daniel Joseph Wasser

Abstract

Evolutionary Polymorphic Neural Network (EPNN) is a novel approach to modeling chemical, biochemical and physical processes. This approach has its basis in modern artificial intelligence, especially neural networks and evolutionary computing. EPNN can perform networked symbolic regressions for input-output data, while providing information about both the structure and complexity of a process during its own evolution.

In this work three different processes are modeled: 1. A dynamic neutralization process. 2. An aqueous two-phase system. 3. Reduction of a biodegradation model. In all three cases, EPNN shows better or at least equal performances over published data than traditional thermodynamics /transport or neural network models. Furthermore, in those cases where traditional modeling parameters are difficult to determine, EPNN can be used as an auxiliary tool to produce equivalent empirical formulae for the target process.

  • Feedback links in EPNN network can be formed through training (evolution) to perform multiple steps ahead predictions for dynamic nonlinear systems.
  • Unlike existing applications combining neural networks and genetic algorithms, symbolic formulae can be extracted from EPNN modeling results for further theoretical analysis and process optimization.
  • EPNN system can also be used for data prediction tuning. In which case, only a minimum number of initial system conditions need to be adjusted. Therefore, the network structure of EPNN is more flexible and adaptable than traditional neural networks.
  • Due to the polymorphic and evolutionary nature of the EPNN system, the initially randomized values of constants in EPNN networks will converge to the same or similar forms of functions in separate runs until the training process ends. The EPNN system is not sensitive to differences in initial values of the EPNN population. However, if there exists significant larger noise in one or more data sets in the whole data composition, the EPNN system will probably fail to converge to a satisfactory level of prediction on these data sets.
  • EPNN networks with a relatively small number of neurons can achieve similar or better performance than both traditional thermodynamic and neural network models.

The developed EPNN approach provides alternative methods for efficiently modeling complex, dynamic or steady-state chemical processes. EPNN is capable of producing symbolic empirical formulae for chemical processes, regardless of whether or not traditional thermodynamic models are available or can be applied. The EPNN approach does overcome some of the limitations of traditional thermodynamic /transport models and traditional neural network models.

Share

COinS