Prediction of n-Alkanes Using Artificial Neural Networks to Enhance Waste Cooking Oil Hydrodeoxygenation over a Tire Rubber-Derived Carbon-Supported Nickel Catalyst

Waste cooking oil is a valuable feedstock for renewable biofuels via hydrodeoxygenation reactions. This study presents a novel strategy with nickel catalysts supported by carbon derived from waste tire rubber (Ni/C). The findings reveal the synthesis of renewable biofuel with diesel-like properties. Under the investigated conditions, optimal yields of 70.12% and 96.5% are achieved for n-C17 and the total C10–C18 alkane fraction, respectively. To achieve them, we conducted 246 hydrodeoxygenation reactions in a stainless-steel batch autoclave with a vertical four-blade agitator; the impeller was operated in alternating rotation at 120 rpm to ensure comprehensive radial and axial mixing. For each one, 0.9 g of the Ni/CTR catalyst was used, varying reaction inputs: system pressure (20–25 bar), active metal loading of the catalyst (5–10 wt.% Ni), isothermal reaction temperature (320–340 °C), and reaction time (4–5 h), obtaining output variables as the molar yields of catalytic n-C17 (dynamic range: 36.78% to 63.03%) and the global yield of the C10–C18 alkane series (50.99% to 88.98%). After that, we trained and evaluated 2500 distinct neural network configurations to identify the optimal architecture. The best model achieved R2 values of 0.9525 for the C10–C18 alkane series and 0.9544 for n-C17, with error metrics of MSE, MAE, and MAPE at 0.0019, 0.0388, and 9.54%, respectively. Finally, we conducted 1000 simulations, increasing input ranges and step sizes, varying reaction parameters, and predicting yields using the neural network to identify the maximum alkane production. This led to improvements in n-C17 and total alkane (C10–C18) yields of 11.25% and 8.45%, respectively.