A disturbance rejection based neural network algorithm for control of air pollution emissions

A novel neural network algorithm for training a model of nonlinear systems that is significantly affected by unmeasured disturbances is presented. In this paper, the algorithm is used to develop a model of nitrogen oxides (NO/sub x/) emitted from a coal-fired, power plant. The NO/sub x/ emissions are affected by unmeasured disturbances such as those caused by changes in fuel characteristics and ambient conditions. The resulting NO/sub x/ model is subsequently used in a control system for reduction of NO/sub x/ emissions, therefore, increased accuracy of the model leads to improved verification and validation of the control system. Two examples illustrate that the resulting model provides a better prediction of NO/sub x/ emitted from coal fired, power plants.