传统BP神经网络在训练完之后,其权重是固定不变的,加上神经网络的样本的标准化处理,将使得网络不易描绘样本峰值.因此,本文考虑变权的方法,以调节训练后的BP网络权重,基于变权次数,建立不同网络模型,并利用不同网络输出值与相应实测值进行比较.结果表明:变权BP网络预报效果有较大提升,同时,降低了对因子相关性的要求.%As the weight of BP neural network is fixed after completing the training, with the normalized samples of neural networks, the network has difficulty with the description of the peak of sample. Therefore, this paper considers the method to change the weight of BP network after training, based on the number of variable weights, to establish different network model, and use different network output compared with the corresponding measured values. The results show that the BP neural network based on weight adjustment has high performance in forecast, and reduced the requirement for the factor correlation.
展开▼