Acer truncatum seed oil (ATO) is a unique woody oil resource, but its high acid value resulting from traditional pressing methods falls short of meeting the necessary food standards. In this research, molecular distillation (MD) was employed for the deacidification of ATO and compared with conventional techniques such as alkali refining and solvent extraction. Predictive models for MD deacidification were established through single-factor experiments, orthogonal experimental design (OED), response surface methodology (RSM), and genetic algorithm - back propagation artificial neural network (GA-BP-ANN). Using RSM, a fitting model was obtained with R = 0.999 and Rpre = 0.990 through multiple linear regression. The constructed BP-ANN model (3-9-1) exhibited R = 0.998 and Rpre = 0.998, indicating slightly higher prediction accuracy. With GA-BP-ANN, the optimal process conditions are an evaporator temperature of 190 °C, condensation temperature of 40 °C, scraper speed of 324 rpm, yielding an acid value of refined ATO at 0.240 ± 0.001 mg g−1. Therefore, under appropriate conditions, the MD deacidification method can produce high-quality refined ATO. The developed orthogonal, RSM, and GA-BP-ANN models demonstrated excellent capabilities in optimizing and predicting the MD deacidification process, meeting even the strictest food regulations.
Read full abstract