Prediction of the Syngas composition of a gasification process

The scope of this work is the development of a mathematical model of a gasification process to be used for the prediction of the Syngas composition. The predictions are intended to support the gascromathographic measurements of the Syngas composition which are often not available due to periodic calibrations. This work represents the first step of broader project which scope is the development of a supervisory controller to perform process optimization and fault detection and isolation. The original contribution of the paper is the design of a new procedure for the selection of the dataset used to design and to identify the model; the approach is based on the use of the Fuzzy C-means (FCM) algorithm for the generation of the main clusters used as input data in the identification process. The presented results show the goodness of the proposed FCM method in filtering a large dataset and the reliability of the identified model in the prediction of the Syngas composition.

1. INTRODUCTION

In its widest sense the term gasification covers the conversion of any carbonaceous fuel to a gaseous product with a useable heating value. Currently the dominant technology is the partial oxidation of the fuel [1], which produces a synthesis gas (otherwise known as Syngas) consisting of hydrogen and carbon monoxide in various ratios, whereby the oxidant may be pure oxygen, air, and/or steam. Partial oxidation can be applied to solid, liquid, and gaseous feedstocks, such as coals, residual oils, and natural gas. Nowadays the gasification process is extensively present in modem Integrated Gasification and Combined Cycle (IGCC) plants used for the generation of electric power ([2]). In order to design complex control strategies for supervisory purposes [3] or Model-Based Fault Diagnosis systems, the development of prediction models capable of well fit process data is necessary. Several approaches for the prediction of the Syngas composition based on Principal Component Analysis (PCA) and Dynamic PCA [4], or Robust Neural Estimator [5] or physical-chemical modeling [6] have been proposed. Nevertheless most of them are eligible for solving the optimization problems or the Model-Based Predictive Control