顾及不确定因素的GA-BP神经网络在路基沉降预测中的应用

Application of GA-BP neural network with uncertainties to the subgrade settlement prediction

BP神经网络初始权值和阈值输入不同,将导致BP神经网络预测不稳定,精度也不是很高。通过遗传算法(GA)对BP神经网络的初始权值和阈值进行优化,能很大程度上提高预测的精度,但是,由于输入层不可能将影响输出的所有因素都包含在内,而这些没有考虑到的因素势必影响预测结果。文中将这些无法得知的不确定因素当做一个综合影响因素,定义为X因素,在建立模型时加以考虑。实验结果表明,这种顾及不确定因素的GA-BP神经网络模型能进一步提高预测精度。

Initial weights and the threshold of BP neural network input will result in unstable BP neural network,and very low precision.A genetic algorithm（GA） can largely improve the accuracy of the prediction.The input layer can not affect the output of all factors included,so some of the factors are bound to affect the prediction results.These uncertainties as a combined effect that can not be learned,are defined as the X factor to be taken into account in the modeling.The experimental results show that the uncertainties of GA-BP neural network model can further improve the prediction accuracy.