Abstract: In order to explore the effect of material components on material properties and the optimal ratio of paste-like filling material，the BP network model was established based on the 16 groups results of ratio test.The structure of the model is the components and 5-10-3 type of layered degree，slump and long-term strength.By means of entering the refined ratio parameters into prediction model，the optimal ratio were obtained.The results showed that，the maximum relative error of the network model is 6.08%; The amount of cementing material and aggregate determine the long-term strength of materials together.High proportion of coal gangue is beneficial for forming strong structure filling body with the accurate ratio of other components.When the proportion of fly ash is 3 to 4 times that of cement and the proportion of sandstorm is almost equal to coal gangue，the slurry has the best overall performance.The gob-side entry retaining was backfilled according to the optimal material ratio in experimental mine，resulting that the convergence of surrounding rock was light and stable.

Key words: Back-propagation neural network, Paste-like material, Slurry ratio, Optimization