基于改进神经网络的渗透率预测方法

Permeability prediction method based on improved BP neural network

由于传统BP算法具有收敛速度慢、易陷入局部极小值等不足,文中对其进行了改进。在Kozeny-Carman方程和杨正明研究的基础上,借助于MATLAB神经网络工具箱,建立了预测岩石渗透率的3层前馈型BP神经网络模型。对改进的神经网络模型进行的仿真训练结果表明:改进模型具有更快的收敛速度和更高的精度,模型预测值与实验室测试值的一致性比较好,其相对误差小于10%,完全能够满足现场精度要求。

The traditional BP algorithm has slow convergence rate,and is easy to fall into local minimum.It is improved based on Kozeny-Carman equation and the study of Yang Zhengming,and a three-layer feedforward BP neural network model for permeability prediction is established by means of MATLAB neural network toolbox.The simulation training of the improved neural network model is carried out.The result shows that the improved model has faster convergence rate and higher accuracy.The values predicted by the model are consistent with the laboratory test values,and the relative error is less than 10%,so it can completely meet the accuracy demand of well site.