基于PCA-BP神经网络模型的充填体强度预测

充填体强度预测对矿山充填设计具有重要意义。选取胶砂比、水泥、石灰、石膏及矿渣含量作为充填体强度影响因素,借助主成分分析(PCA)消除自变量间相关性,降低数据维数,再结合BP神经网络具有较好预测性的特点,建立了PCA-BP模型以预测充填体强度。对18组充填体试验数据进行主成分分析,5个影响因子降维为3个主成分,将其作为BP神经网络的输入因子,进而探讨了隐含层神经元个数对充填体强度训练和预测精度的影响,并比较了PCA-BP神经网络、标准BP神经网络和二次线性回归效果。结果表明:PCA-BP模型最佳预测结构为3-7-1;PCA-BP神经网络结果优于BP神经网络和二次线性回归;PCA-BP神经网络训练和预测的最大相对误差仅为3.65%,实现了充填体强度的准确预测。PCA-BP模型为充填体强度预测提供了一种高精度的分析方法。

Strength Prediction of Filling Body Based on PCA and BP Neural Networks

The strength prediction of filling body is of significant importance in the design of mine filling.The cement to sand ration and the contents of cement,lime,gypsum and slag are selected as influence indexes of the strength of filling body.In addition,the PCA-BP model to predict strength of filling body was constructed by taking advantages that the principal component analysis(PCA)can eliminate the correlation between independent variables and reduce the input data and the BP neural networks has a good predictability.The PCA has been tested on 18 pieces of experimental data,and the five influencing factors that were eventually reduced into three main factors.Then these three main factors were used as BP neural networks input data.Furthermore,the influence of hidden layer neuron on model training process and prediction accuracy was discussed,and results of PCA-BP method,standard BP method and quadratic polynomial regression method were compared.The results show that the optimal structure of PCA-BP method is 3-7-1 and the PCA-BP method is better than the standard BP method and the quadratic polynomial regression method with an error ratio within 3.65%,achieving an accurate prediction for the strength of filling body.In a conclusion,the PCA-BP model provides a high accuracy means for the prediction of the strength of filling body.