Hybrid machine learning approach towards olefins plant optimization

Abstract

The present disclosure describes systems, methods, and computer readable media that provide a hybrid approach that uses machine learning techniques and phenomenological reactor models for optimization of steam cracker units. While the phenomenological model allows capturing the physics of a steam cracker using molecular kinetics, the machine learning methods fill the gap between the phenomenological models and more detailed radical kinetics based steam cracker models. Also, machine learning based models can capture actual plant information and provide insight into the variation between the models and plant running conditions. The proposed methodology shows better interpolation and extrapolation capabilities as compared to stand-alone machine learning methods. Also, compared to detailed radical kinetics based models, the approach utilized in embodiments requires much less computational time in order to carry out whole plant-wide optimization or can be used for planning/scheduling purposes.