Modelagem da queima do carbono na caldeira de recuperação 2 da Klabin Monte Alegre por rede neural artificial

Abstract

The brazilian paper and wood pulp sector have been growing in a substantial way, reaching a prominent position in the global economy. Tough our country is a spotlight in this sector, it also faces challenges on the process optimization and the reduction of environmental damage. The quest for problem solutions and improvement is a continuous challenge. For doing that, process simulation is a widely used tool. The simulation using artificial neural networks have been object of growing researches and becoming a useful tool for several applications, including Paper and pulp processes. The purpose of this paper is to create a mathematical model through an artificial neural network able to describe the black liquor burning process, a pulp process subproduct, used for energy production and chemical recovery. The neural network building was made by using the Property Predictor software by Pavilion Techology. The entrance variables selection was based on the operation knowledge, by an opinion questionnaire. The model created by neural network used pre-processing data, training, sensitivity analysis and validation. The artificial neural network was formulated with 21 inputs and the neurons in the hidden layer were calculated automatically by the software. The training was used backpropagation.The proposed model didn’t present a meaningful result. The small data quantity, and poor data quality have resulted on the overfitting of the net, which only allows to represent the training data, resulting in a poor predictive performance. For better results, more data is required. With the model improvement it is possible to use the virtual sensor technique, which will allow quick information to the operational area, contributing on process parameters changing decisions.