The Application Of BP Networks Based On Genetic Simulated Annealing Algorithm For The Prediction

In recent years, polymer materials, especially plastics industry achieved rapid development, because of the stability of the national economy. Plastics industry became an important pillar industry, and got the third place in light industry.With advantages as excellent mechanical properties, light weight, corrosion resistance, electrical insulation performance, and so on, plastics have a wide range of applications in various fields, such as auto parts, household appliances, health care, package, building materials and aerospace. With the use of injection molding products, people's demands of appearance and quality are becoming increasingly. Among the defects of products, warpage is the important one which we put much focus on. The control of warpage become our concern object.With the maturity of the finite element theory and numerical optimization methods, we can simulate the injection molding processing using different mathematical model. Numerical simulation technology provides a tool and a new way to study how to control the warpage defects.Under the background of the above-mentioned technology, this paper has done research on the controlling method on wrapage. With the warpage of a plastic part as the target, we set up the numerical simulation experiment. Simultaneously, we select mold time, injection temperature, mold temperature, packing pressure, packing time, injection speed as the experimental parameters. According to the arrangements of experimental program, we use Moldflow as a tool to simulate the injection process.Then we identify impact factor of every parameters use the signal to noise ratio analysis method. We also find the best combination of process parameters. By regression analysis, SPSS tools can predict warpage for each group. We compare the warpage of simulation experiments and the predictions.For the low prediction accuracy of regression analysis, we creat a BP neural network model which can predict the warpage. In this model, we treat the six process parameters as network input, and warpage as output. At last, we use a hidden layer in BP neural network.The defects of unstable initial value and the trap in local lead to the low low prediction accuracy of BP algorithm. In this paper, we design a BP network which is optimized by genetic simulated annealing algorithm. This model combines parallel search algorithm, probability jump features and generalization performance.We compares the prediction accuracy of BP network and the optimized BP network. Results show that the prediction accuracy of optimized BP network is higher than that of BP network. At the same time, the optimized BP network can not only speed up convergence rate,but also enhance the global searching ability.This paper provide a new method to study on the prediction of warpage. It has important applications, and can be the basis for further research.