Fuel blending plays a very important role in petroleum refineries, because it directly affects the quality of the end products, as well as the overall profitability of the refinery. This process of blending involves a combination of various hydrocarbon streams to make fuels that meet specific performance standards and comply with regulatory guidelines. For many decades, most refineries have been dependent on linear programming (LP) models for developing recipes for blending optimization. However, most LP models normally fail to capture the complex nonlinear interaction of blend components with fuel properties, leading to off-specification products that may necessitate re-blending. This work discusses a case study of a hybrid artificial intelligence (AI)-based method for gasoline blending based on a genetic algorithm (GA) combined with an artificial neural network (ANN). AI-based blending systems are more flexible and will enable the refineries to meet the product specifications regularly and result in cost reduction owing to the fall in quality giveaways. The AI-powered process discussed can predict, with much better accuracy, critical combustion properties of gasoline such as the Research Octane Number (RON), Motor Octane Number (MON), and Antiknock Index (AKI), compared to the classical LP models, with the added advantage of optimization of the blend ratio in real time. The results showed that the AI-integrated fuel blending system was able to produce fuel recipes with a mean absolute error (MAE) of 1.4 for the AKI. The obtained MAE is close to the experimental uncertainty of 0.5 octane. A high coefficient of determination (R2) of 0.99 was also obtained when the system was validated with a new set of 57 fuels comprising primary reference fuels and real gasoline blends. The study highlights the potential of AI-integrated systems in transforming traditional fuel blending practices towards sustainable and economically viable refinery operations.
