Abstract: AIM: To set up the non-linear corresponding experimental data model between the ingredients compatibility of proportioning of pinellisa decoction for purging stomach-fire and gastric protease by BP artificial neural network.METHODS: 24 groups were formed based on 8 kinds of medical herbs including ginger rind and herbal ingredients in proportioning of pinellisa decoction for purging stomach-fire according to homogeneous design.The influence of the activity of gastric protease in normal rats was measured by ansongshifa.The research was programmed by MATLAB 6.5, and the experimental data was conformed by BP artificial neural network.In the 24 groups, 21 groups as samples were studied to form the model, and the other 3 groups as the unstudied samples were test and verify the forecast ability of data.RESULTS: BP artificial neural network that had 8-10-1 structural topology was formed according to the study of the experiment data in 21 groups.It was a kind of good method to uniform the studied samples and to forecast the unstudied samples, and the interrelated coefficient in the forecast value and reality value was 0.9433.CONCLUSION: The BP artificial neural network shows the excellent ability of self-organization and self-acclimation, and it conforms to the complicated non-linear relation during the proportioning of compound prescriptions.

Key words: BP artificial neural network, proportioning of pinellisa decoction for purging stomach-fire, compound prescription, homogeneous design