Monitoring and control of complex energy-intensive industrial systems is an urgent need for minimizing their harmful greenhouse gas (GHG) emissions. Fortunately, these systems are equipped with enormous number of sensors, resulting in data that express the behavior of their whole process. This presentation proposes a novel data representation method that maximally exploits these collected data through converting the observations into informative polygons. These polygons represent all interrelationships between the system input variables and outputs through Hamiltonian cycles. These generated polygons are fed into sophisticated deep learning (DL) architectures such as convolutional neural network (CNN) for accurate classification or conditional generative adversarial network (cGAN) for accurate multi-output regression. This proposed prediction method was validated using different benchmark and real industrial case studies for fault diagnosis and key performance indicators’ (KPI) prediction; Tennessee Eastman process (TEP), a reboiler system of heat recovery network in thermomechanical pulp mill and black liquor recovery boiler (BLRB) in Kraft pulp and paper mill. The proposed prediction approach outperformed other classical and state-of-art machine learning (ML) and DL predictors. The results obtained demonstrate the effectiveness of our proposed method in terms of accurate modeling of complex highly non-linear industrial systems. Accordingly, this helps the stakeholders move towards better decision-making process to minimize their environmental footprint.
Share this page on your timeline