The properties of PP/POE blends were studied by neural network. The effect of neural network was verified by the data of orthogonal test design of temperature

proportion and impact strength. The experimental data of 10℃

20℃ and 30℃ are used for learning and training

and the experimental data of 0℃ are used for testing. At the same time

three kinds of neural network models with different structures are obtained by adjusting different parameters. The results show that neural network has a good effect on the prediction of blending modification. Using neural network method to analyze the data of polymer blending modification can reduce the times of experiments

improve the efficiency of experiments

and quickly complete the product formulation design of blends. The general direction of model parameter adjustment under the network type is obtained

which shows that under the condition of less sample data

less number of hidden layers and neural units can achieve better prediction effect.