Optimization process of coffee pulp wines combined with the artificial neural network and response surface methodology

Abstract

Abstract Coffee pulp wine was made from coffee pulp. The level range of fermentation factors was determined by one-factor experiment. The significance factors affecting fermentation were screened by Plackett-Burman and steepest climbing experiments, which were material-liquid ratio, initial pH, initial sugar and yeast amount, respectively. The screened factors were then subjected to a central combination design, and the results were optimized using RSM and ANN-GA. The ANN-GA shows a more accurate optimization effect compared with RSM and a higher degree of model fitting. The coefficient of determination (R2) of the ANN-GA predicted value was 0.9140, while the RMSE was 0.0896. The best results of optimization process showed that the material-liquid ratio was 4.25 : 95.75, the initial pH value was 6.92, the initial sugar concentration was 22.248%, the yeast addition was 1.98%, and the final predicted value was 10.255 mg/L. The research results provided a technical reference for the production of coffee pulp wines.