Abstract:
Hardware based sensing frameworks such as cooperative fuel research (CFR) engines
are conventionally used to monitor research octane number (RON) in the petroleum
refining industry. In this work, machine learning techniques are employed to predict
the RON of two petroleum refining processes: (1) integrated naphtha reforming and
isomerization process and (2) fluid catalytic cracking process. A dynamic Aspen
HYSYS model was used to generate data by introducing artificial uncertainties in the
range of ± 5\% in process conditions such as temperature, pressures, flow rates, etc.
Generated data was used to train support vector machines, gaussian process regression,
artificial neural networks, regression trees, and ensemble trees. Hyperparameter tuning
was performed to enhance the prediction capabilities of gaussian process regression
(GPR), artificial neural network (ANN), support vector machines (SVM), ensemble
tree (ET) and regression tree (RT) models. Performance analysis of machine learning
models indicates that in case of integrated naphtha reforming and isomerization
process GPR, ANN, and SVM have R
2
values of 0.99, 0.978, and 0.979 and RMSE
values of 0.108, 0.262, and 0.258, respectively. GPR, ANN, and SVM performed
better than the remaining models and had the prediction capability to capture the RON
dependence on predictor variables. ET and RT had an R2
value of 0.94 and 0.89. In
case of fluid catalytic cracking GPR, SVM, and ANN have R
2
values of 0.97, 0.969,
and 0.963 and RMSE values of 0.3908, 0.3934, and 0.4333, respectively. ET and RT
had an R2
value of 0.941 and 0.88. The GPR model was used as a surrogate model for
fitness function evaluations in two optimization frameworks based on the genetic
algorithm and particle swarm method. Particle swarm optimization performed
marginally better than genetic algorithm. The proposed methodology of surrogatebased optimization will provide a platform for plant-level implementation to realize
the concept of industry 4.0 in the refinery.
