Recurrent neural network prediction of steam production in a Kraft recovery boiler

摘要

In this paper, neural networks approaches are compared for predicting the high pressure (HP) steam flow rate from a Kraft recovery boiler. We apply two types of neural networks: a static multilayer perceptron and a dynamic Elman's recurrent neural network. Starting from a one-day database of raw process data related to the boiler, the goal is to model and predict the next 12-hours of HP steam flow production from the boiler to the steam turbine. The results illustrate the potential of the dynamic approach in this task.