Implementing a neuro fuzzy expert system for optimising the performance of chemical recovery boiler

Abstract

In chemical recovery boilers of paper mills, main steam outlet temperature control cannot be solved by straight forward automation control. As prior knowledge of the mechanism to maximise steam generation without affecting steam main temperature is unknown, a backpropogation supervisory neural network has been designed which exhibits a good degree of reinforcement learning. Various parameters considered encompassing concentration, composition and firing load of black liquor solids may not have ideal fixed values. Hence, a type 2 fuzzy logic model has been designed which in turn monitors the parameters and predicts the results. Errors are fed back iteratively through the backpropogation network, until the network learns the model. Fuzzy C-means clustering technique has been used to find coherent clusters. Then sensitivity analysis has been done to identify the parameters playing a significant role in obtaining the results. As it can be observed that the behaviour is stochastic, particle swarm optimisation has been implemented to optimise the combined effect of all parameters. Through this tool connecting steam attemperation control and smart soot blowing, clean heating surface is ensured resulting in enhanced green energy output and availability.