一种用于地应力分析的改进型BP神经网络方法

地应力测量及分析研究,对地应力活动方式、构造体系的研究以及岩土工程与结构的设计和稳定性,都具有重大的理论意义和实用价值.BP神经网络算法比较成熟,已被广泛应用,但一般BP神经网络算法存在训练学习速度较慢、样本泛化能力差的问题,通过引入动态学习因子和惯性因子以及模拟辅助样本,对神经网络进行了改进.通过调节动态学习因子和惯性因子以及对样本集数据处理等手段,对样本的学习、训练进行优化处理,通过实例验证,将优化好的网络样本训练结果与一般结果进行比较,结果表明对三层BP神经网络进行的优化,在提高计算精度的同时也提高了网络的收敛速率,证明改进的算法能够很好用于地应力分析.

An Improved BP Neural Network Method for Geostress Analysis

The research of geostress measurement and analysis is of important theoretical significance and practical value.So are the studies of geostress activity patterns,tectonic system and the design and stability of geotechnical engineering and structure.The BP neural network algorithm is widely used as it is relatively mature,but in general there are some problems about the usual BP algorithm.For example,the training speed is slow,and the ability of sample generalization is poor.The dynamic learning factor and inertial factor,as well as the aid of simulating assistant samples of neural network,shoule be introduced to improve the neural network algorithm.By adjusting the dynamic learning factor and inertial factor and by optimizing the sample set,the learning and training of the sample can be optimized.These optimizations are verified by examples and the results are compared.The results show that the optimizations of three-layer BP neural network can improve the accuracy of calculation and enhance the network convergence rate.Experimental results have proved that this algorithm can be used well in stress analysis.