Research on Improved Neural Network Forecast Basing on Genetic Algorithm

The problem of forecast belongs to an input-output nonlinear system in nature. And most of problems which need to be forecasted have a large number of predictors which are relatively correlated. Therefore neural network has unique superiority in dealing with such problems. But when traditional BP (Back-Propagation Network) neural network is used to predict, there are many inadequacies in predictive results because of the inherent defects of BP network. For this reason, genetic algorithm (GA) is used to optimize the weights and thresholds of the BP network and to optimize the number of hidden layer neurons in order to get an optimal network model which can be applied to forecast practices. Several cases show that in many aspects, such as forecast accuracy, convergence rate, forecast error, the number of successful training, the new model is superior to the traditional model. Thus the new forecast model has broad application prospects.