加速BP网络在单桩沉降预估中的动态适应

提出一种学习率自适应加速算法，即采用动态的网络学习率代替在整个学习过程中根据经验选取的传统的学习率，使ＢＰ神经网络的收敛速度得到显著的提高。应用这种神经网络，对单桩沉降预估过程以及信息扩散原理所取得的样本－教师模式匹配，可进行动态的学习。数值模拟表明，加速的学习算法使这种神经网络对问题的解具有动态适应性。

Dynamic Adaptation of Accelerated BP Network in Preestimating the Settlement of Single Pile

For enhancing convergence speed of BP neural network,an adaptive accelerated algorithm of learning rate is advanced.That is to say,the conventional learning rate which is chosen according to experience during the entire process of learning will be replaced by the dynamic learning of the network.By applying this neural network,the settlement of single pile is preestimated;and dynamic learning of sample-teacher pattern matching based on information diffusion principle can be carried out.As indicated by numerical simulation,the accelerated learning algorithm makes the problem settlement os this neural network to have dynamic adaptability.When it is adopted to preestimat the settlement of single pile,the learning of network can be shortened,moreover,its preestimate is better than standard BP network in accuracy after similar round or time of learning.