基于BP神经网络预测的附加质量法密度反演

针对附加质量法堆石体密度反演存在参数获取困难、计算过程复杂等问题，以堆石体刚度和参振质量为影响因素，构建基于BP神经网络的密度预测模型，通过分析基于梨园堆石坝实测资料训练下的网络的训练性能，选取最优预测模型实现密度的反演输出。结果表明，BP神经网络对于附加质量法的密度反演预测具有较高的适应性和精度；与相关法相比，BP神经网络法可避免复杂密度函数的推导，且其反演结果的相对误差保持在局部稳定区间。研究成果为快速解决非线性密度反演问题提供了参考。

Rockfill Density Inversion Prediction in Additive Mass Method Based on BP Neural Network

Based on BP neural network, the prediction model of density was constructed by taking rockfill rigidity and vibrating mass as influencing factors in condition that the density inversion of additive mass method has difficulty to obtain the parameters and the data calculation supposed to be complex. As a result of the performance analysis of the network trained by measured data of Liyuan rock-fill dam, predictive output of density was accomplished by picking the optimal prediction model. The results show that the BP neural network was proved to be with good adaptability and accuracy on prediction of the density inversion in additive mass method. Compared with correlation method, BP neural network method could avoid the complicated derivations of density function, and its relative errors of the inversion results are kept within a stability interval. The research results provide reference for solving nonlinear density inversion problem quickly.