Prediction of Physicochemical Properties of Biodiesel/n-Butanol Using Backpropagation Neural Networks

With the growing global demand for renewable energy, biodiesel has garnered widespread attention due to its advantages such as renewability. However, owing to its high viscosity and poor low-temperature performance, n-butanol is introduced to improve its kinematic viscosity and low-temperature properties. This study determined the kinematic viscosity, density, freezing point, cold filter plugging point, and calorific value of Jatropha curcas biodiesel/n-butanol blended fuel, and employed GA-BP, PSO-BP, and BP models for prediction. Results indicate that at the ratios set in this experiment, increasing n-butanol content reduces the kinematic viscosity and density of the blended fuel. When the n-butanol ratio reaches 60%, the maximum decreases in freezing point, cold filter plugging point, and calorific value reach 18%, 21%, and 10%, respectively. The GA-BP model achieved training set determination coefficients close to 1 for all indicators in the prediction task, with the lowest correlation coefficient reaching 95.26%. Its prediction accuracy outperformed both the PSO-BP and BP models. After reformulating the blended fuel, the GA-BP model yielded predicted values with a minimum relative error of 0.11% compared to experimental measurements, while the maximum relative error remained below 9.94%. Overall, the GA-BP model demonstrated higher predictive accuracy for both linear and nonlinear data, providing reliable data support for blended fuel design and application.