基于有限次残差拟合的BP神经网络组合模型

摘要

The optimization of Back Propagation (BP) neural network is always concentrated on improving the prediction accuracy of single predictor. Therefore,a combinedback propagation neural network model based on fitting residuals finitely is first put forward. Results show that in those cases which are suitable for classicalBPneuralnetwork,it is proved that the accuracy of the combined model is better than classical BP neural network. Moreover,its accuracy is superior to genetic BP neural network(GA-BP),and its modelingefficiency is 99.2 percent higher than GA-BP.%关于BP神经网络的优化，鉴于大多数的思路都集中于提高单个预测器的预测精度，首次提出了基于有限次残差拟合的BP神经网络组合模型。结果显示，在经典BP神经网络适用的情境下，组合模型能够有效提高预测精度，此外，进一步证实其预测精度优于遗传神经网络（GA-BP），且建模效率比GA-BP提高99.2%。