Dynamic optimization of a pilot-scale entrained-flow gasifier using artificial recurrent neural networks

Abstract

The aim of this study is to present the construction and validation of a recurrent neural network (RNN) that can efficiently predict the dynamic performance of a pilot-scale gasifier unit. The RNN model consists of a set of sub-networks that predict the transient behavior of the each of the key process outputs as a function of the input parameters of the pilot-scale gasifier. The performance of the RNN was compared to a ROM model, which was validated using experimental data and computational fluid dynamics. The RNN model was embedded within an optimization formulation to investigate the optimal operation of a gasifier under key operational constraints. The results from these optimization studies illustrate the benefits of the RNN, which were able to identify optimal time-dependent profiles in key input variables that improve the efficiency and availability of the gasifier under load-following and co-firing.