NOx Concentration Prediction With a Flexible Cascaded Echo-State Network in a Cement Clinker Calcination System

Abstract

Measuring nitrogen oxide (NOx) concentration accurately and timely is critical for pollution control during cement clinker calcination. However, due to the harsh environment and complex reactions, NOx concentration in calcination systems cannot be directly measured. In this article, a flexible cascaded echo-state network (FCESN) is developed for online measurement of the NOx concentration during the cement clinker calcination. First, a flexible cascaded network structure is proposed, where some echo-state networks (ESNs) as basic modules are added to the network in the form of cascade one by one until the learning error is satisfied. Then, based on the idea of gradual approximation, a novel training algorithm is proposed to pretrain ESN modules for extracting the dynamic feature of the NOx sequence. With the increase of ESN modules, the learning error is monotonically decreasing. Finally, the proposed approach is implemented for online prediction of the NOx concentration in an actual cement clinker calcination system. The prediction results indicate that FCESN is superior to other popular prediction methods in training speed and accuracy.