Optimization using ANN Surrogates with Optimal Topology and Sample Size

Soumitri, M S and Majumdar, Saptarshi and Mitra, Kishalay (2015) Optimization using ANN Surrogates with Optimal Topology and Sample Size. In: IFAC International Symposium on Advanced Control of Chemical Processes, 7-10 June 2015, Whistler, Canada.

Full text not available from this repository. (Request a copy)

Abstract

Industrial scale process modelling and optimization of long chain branched polymer reaction network is currently an area of extensive research owing to the advantages and growing popularity of branched polymers. The highly complex nature of these reaction networks requires a large set of stiff ordinary differential equations to model them mathematically with adequate precision and accuracy. In such a scenario, where execution time of model is expensive, the idea of making the online optimization and control of these processes seems to be a near impossible task. Catering to these problems in the ongoing research, the authors presented a novel work where the kinetic model of long chain branched poly vinyl acetate has been utilized to find the optimum processing conditions of operation using Sobol sequence based ANN as meta models in a fast and highly efficient manner. The article presents a novel generic algorithm, which not only disables the heuristic approach of designing the ANN architecture but also allows the computationally expensive first principle model to determine the configuration of the ANN which can emulate it with maximum accuracy along with the size of training samples required. The use of such a fast and efficient Sobol based ANN as surrogate model obtained by the proposed algorithm makes the optimization process 10 times faster as compared to a case where optimization is carried out with the expensive first principle model.

[error in script]

IITH Creators: