Fuzzy neural network integrated with PCA and its application in raw meal grinding process

A fuzzy neural network model has been proposed and successfully applied to an annual clinker production capacity of 0.73 million ton of Jiuganghongda Cement Plant in China. Because the measurement values from raw meal grinding process are not independent, data sets with higher dimension increased model structure. Thus, a novel method based on fuzzy neural network(FNN) and principal component analysis (PCA) is discussed in detail. In this method, the PCA was applied to the model, which not only solved the linear correlation of the input variables, but also simplified the fuzzy neural network(FNN) structure and improved the training speed. Industrial application results show that the fuzzy neural network model has high accuracy and guidance to calciner temperature setting.

1 INTRODUCTION

In a raw meal grinding process, the particle size of the raw meal is directly related to quality of final products. However, it is difficult to be measured by sensors. In practical production process, the particle size of the raw meal is measured off-line once an hour, then an operator adjusts the separator frequency and materials ratio to control particle size of raw meal. However, there is a large lag for off-line detection as well as subjective analysis error. Therefore, according to the measurement off-line, it is hard to control efficiently the particle size of raw meal. The measurable auxiliary variables are used to develop a fuzzy neural network(FNN) model.