Improved batch process performance by evolutionary modelling

摘要

A key issue for improving the industrial efficiency in energy and material resources usage is tointegrate the dynamics of the process and its scenario in the actual plant operation decision making. Inthis work, a combination of statistical techniques has been used to build an automatic modelling toolbased on Neural Networks, that overruns the limitations of modelling techniques based on thetheoretical knowledge of the process principles. As the overall system can be run simultaneously withthe process, the tool can be used to continuously readjust its parameters and to follow evolutionaryprocesses. The resulting model can be applied for process forecasting and control, resulting inimprovements in the process performance. In the Neural Network field, a new method is introducedto test the results and a heuristic is proposed to stop the learning process when the best model hasbeen found.