MPC with a Disturbance Model Using Online Extreme Learning Machine with Kernels for SCR Denitrification System

摘要

Due to big delay, nonlinearity and unknown disturbance, selective catalytic reduction denitrification system cannot always achieve satisfactory control performance using PI/PID based controllers. To increase the control performance, this paper uses extreme learning machine with kernels to develop a disturbance increment model for MPC. A novel online learning algorithm with an adaptive training set for extreme learning machine with kernels is proposed to make it more suitable for applications in real time. The online algorithm is employed to model and predict the disturbance increments. Then, the MPC controller with an adaptive disturbance increment model is constructed. Simulation study indicates that this controller can increase performance of SCR control system, especially for periodic and structured disturbances.