Total saponins retention rate prediction model of ultrafiltration for fibrous rhizome herbs based on BP neural network

The objective of this study was to establish the total saponins retention rate prediction model of ultrafiltration for fibrous rhizome herbs. It could avoid optimizing ultrafiltration technology of similar herbs repeatedly.The model was established based on the ultrafiltration data of Hedysari Radix by Levenberg- Marquardt( LM) arithmetic combining BP neural network.The performance and applicability of the improved LM- BP neural network model were evaluated.Then the relative importance of input variables were assessed using the connection weights of the model.Results showed that the model had the better accuracy and applicability.The mean absolute error for Hedysari Radix,Astragali Radix and Glycyrrhiza uralensis was 1.10%,1.28% and 1.52%,respectively.Mean error rate was 1.48%,1.95% and 2.20%,respectively. The relative contribution of input variables to retention rate presented the order of membrane pore size > pressure > temperature. The study could provide a useful reference for the application of ultrafiltration and intelligence algorithm in Chinese medicine industry.