一种改进的BP神经网络深基坑变形预测方法

Improved BP Neural Network Method for Deformation Predication of Deep Excavation

针对深基坑变形因素复杂的特点和现有理论解决变形计算的不足,提出将BP神经网络应用于基坑变形的非线性和不确定性问题的求解,BP模型构建过程中既考虑基坑变形的主要自变量,又兼顾模型在时间上的延展性;同时,通过遗传算法对模型权重初值进行优选,避免模型陷入局部极小值,以达到预测误差全局最小的目的。应用结果显示:改进的预测方法无论是在模型的收敛速度,还是泛化能力方面均表现出较好的性能,对未来2~5d的预测与实际监测值基本一致,完全能满足工程施工要求,该方法对深基坑变形安全监测具有实用指导价值。

In view of the complicated deformation factors of deep excavation and the shortage of existing theory to solve the deformation calculation,the BP neural network is proposed to solve the nonlinear and uncertain problems.In the process of building the BP model,the main variable of the deep excavation’s deformation is considered and the delay of the model is taken into account as well.The genetic algorithm optimizes the initial weight value of the model to avoid the local minimum value,so as to achieve the goal of minimizing the overall prediction error.The application results show that the improved prediction method shows good performance in both the convergence speed and generalization ability.The prediction of the next 2 to 5 days is basically consistent with the actual monitoring value,and it can fully meet the engineering construction requirements.This method has practical guiding value for the deformation safety monitoring of deep excavation.