Paper title

“Soft Sensor for NOx Emission using Dynamical Neural Network”

Authors: M. Shakil1, M. Elshafei1, M. A. Habib1 and F. Al-Maleki2
Affiliation
: 1. King Fahd University of Petroleum and Minerals,

2. Saudi Petrochemical Co. (Sadaf)
, Saudi Arabia

Abstract—In this paper we propose a soft sensor for prediction of NOx emission from the combustion unit of industrial boilers. The soft sensor is based on a dynamical neural network model. A simplified structure of the dynamical neural network model is achieved by grouping the input variables using basic knowledge of the system. Neural network model is trained using real data logs of an industrial boiler. Principal Component Analysis (PCA) is used to reduce number of input variables. Lag space for the model is found by using genetic algorithm to find the best time delayed model. Lag space obtained from the linear model is then used for constriction of the dynamical neural network. The proposed model is validated using different data from the same boiler and its ability to accurately predict NOx emission from the boiler is demonstrated.