Dual channel modeling of SCR denitrification system for coke oven flue gas

Establishing an accurate mathematical model for the Selective Catalytic Reduction (SCR) denitrification system of coke oven flue gas is pivotal for utilizing advanced process control algorithms. In this paper, a modeling approach grounded in field data is proposed. To ensure the model completeness and reduce modeling intricacies, the system is divided into process and disturbance channels, each yielding a Single Input Single Output (SISO) model. In response to the periodic oscillations caused by coke oven reversals, Controlled Auto Regression Integrated Moving Average (CARIMA) models are compared with Controlled Auto Regression Moving Average (CARMA), pushing the boundaries of conventional methodologies. The structure and parameters of the models are identified utilizing the batch least squares method initially. To make full use of the continuous stream of site data, Recursive Least Squares with Forgetting Factor (FFRLS) is also utilized to identify the parameters. Parameter convergence is ensured and the matrix P is monitored through the introduction of an intelligent supervision shell. In the frequency domain, an analysis of model parameters is conducted, leveraging physical properties to refine model fidelity. Through rigorous cross-validation, the efficacy of the established models is validated. Notably, the process channel model achieves an MSE of approximately 8 and an MRE of 2.7%, while the disturbance channel model exhibits an MSE of around 2.8 and an MRE of 1.3%.