摘要
为克服单项预测方法产生的误差,利用灰色模型GM(1,N)、多元线性回归、BP神经网络等3种单项预测方法建立组合预测模型,并采用熵值法确定加权系数。通过对PHC管桩承载力进行比较预测,结果显示GM(1,N)法平均绝对百分比误差(MAPE)值为5.4%,多元线性回归法的MAPE为3.0%,BP神经网络法的MAPE为2.8%,组合预测法的MAPE为2.3%。因此组合预测法精度较高,实用性更强。
The combination forecasting model was building to overcome the potential errors generated by single forecast model on the basis of the grey system GM(1,N),multiple linear regression and back-propagation neural network,and the weighting coefficients were determined by the entropy method.The contrast test was conducted to predict the bearing capacity of PHC pile,and the results show that the method means absolute percentage error(MAPE) of GM(1,N) is 5.4%,the MAPE of multiple linear regression is 3.0%,the MAPE of BP neural network method is 2.8%,and the MAPE of the combined forecasting method is 2.3 %.Therefore the combined forecasting has high precision and practicability.
出处
《河北工程大学学报(自然科学版)》
CAS
2011年第1期64-67,共4页
Journal of Hebei University of Engineering:Natural Science Edition
关键词
PHC管桩
熵值法
组合预测
BP神经网络
PHC pipe
entropy method
combination forecasting
BP neural network