Application of Neural Networks Optimized by Genetic Algorithm in Forecasting Electric Field Aging Technics

In the study, back-propagation neural networks (BP-NN) theory and genetic algorithm (GA) were used to build a nonlinear prediction model reflecting the relationship between technics parameters of electric field aging and mechanical properties of LY12 aluminum alloy. In this model, electric field intensity, aging temperature and time were as input parameters. Tensile strength, yield strength and micro-yield strength were as output parameters. The result shows that BP-NN model has good training ability whose error was less than 0.1%. The maximal error of BP-NN model for forecasting the mechanical properties under selected technics was close to 10%. Using genetic algorithm to optimize BP-NN (GA-BP) can not increase the training ability which had a higher training error in the condition of less experiment datas, but GA-BP model can improve the prediction ability of BP-NN model and the maximal prediction error was less than 4% which lied at rational range. GA-BP model can be used to optimize technics parameters and decrease experimental work and cost which is a new method for studying electric field aging technics.