偏最小二乘回归神经网络模型在爆破振动峰值速度预测中的应用  

Application of neural network model with partial least-square regression in prediction of peak velocity of blasting vibration

作　　者：史秀志[1] 武永猛[1] 唐礼忠[1] 黄宣东[1] 

机构地区：[1]中南大学资源与安全工程学院,长沙410083

出　　处：《振动与冲击》2013年第12期45-49,共5页Journal of Vibration and Shock

摘　　要：神经网络方法是处理非线性问题的有力工具,但当输入变量较多,输入变量间存在的多重共线性性会使得网络的建模效率下降。偏最小二乘回归方法通过提取对因变量解释性较强的成分,能较好地克服变量间的多重共线性。将两种方法相结合,建立了爆破振动峰值速度的偏最小二乘回归BP神经网络预测模型。利用偏最小二乘法对影响爆破振动的因素进行分析,提取出3个新综合变量,使BP网络的输入层节点数目由9个减少到3个,简化了网络结构,提高了计算速度,增强了网络稳定性。分析结果表明,耦合模型的平均预测误差为7.62%,相较于传统的萨氏公式及标准的BP神经网络模型其预测精度有了明显提高。The neural network method is a powerful tool to deal with problems of nonlinearity, but when input variables are so many, the muhicollinearity among variables can lead to a lower modeling efficiency. The partial least- square regression（PLSR） method can extract components with better interpretation to the dependent variables, thus it can overcome the multicollinearity among variables. Here, by combining the two methods, a BP neural network prediction model for peak velocity of blasting vibration based on PLSR was established. The affecting factors on blasting vibration were analyzed by means of PLSR, and three new synthesis variables were extracted. Since the input layer nodes of the BP neural network decreased from nine to three, the network structure was simplified and the netweork became, more efficient and more stable. The results showed that the average prediction error of the combined model is 7.62%, the new model is more accurate than Sadaovsk formula and a normal BP neural network modelbe.

关 键 词：爆破振速 多重共线性 偏最小二乘回归 BP神经网络 

分 类 号：TD235[矿业工程—矿井建设]