基于改进BP神经网络的地下工程围岩分类

摘要

Traditional BP network has its shortcomings of local minimum,the slow convergence speed of learning algorithm and the difficulty defi-ning the structure. Therefore,it analyzed the stability of surrounding rock of underground engineering by using improved BP neural network. Using single-axle saturated rock's compression strength,rock quality designation,integrity coefficient of rock mass,strength coefficient of structural plane and seepage discharge of groundwater as classification index of surrounding rock,using improved BP neural network with momentum factor,L-M optimization algorithm,Nguyen-Widrow method to establish a forecasting model of the stability of surrounding rock,classifying the learning sample by using MATLAB program,confirming the grade of rock mass. Determining main factors of affecting the stability of the surrounding rock by com-bined MIV algorithm selection,the results show that the leading roles of rock stability are rock mass quality,strength and the state. Structural plane occurrence and characters are secondary factors.%传统BP神经网络存在局部极小值、学习算法收敛速度慢、网络结构难以确定等缺点，因此采用改进的BP神经网络分析地下工程围岩的稳定性。将岩石单轴饱和抗压强度、岩石质量指标、岩体的完整性系数、结构面的强度系数、地下水渗流量作为围岩的分类指标，利用附加动量因子、L-M优化算法与Nguyen-Widrow初始化算法相结合的改进BP神经网络建立围岩稳定性的分类预测模型，并运用MATLAB对学习样本进行分类识别，确定地下工程围岩的岩体等级，并结合平均影响值特征筛选法确定影响围岩稳定性的主导因素。实例分析结果表明：岩体的质量、强度以及所处的状态对围岩的稳定起主导作用，结构面产状和性状则为次级影响因素。