储层敏感性预测模型中BP网络隐层数的优选及应用

储层敏感性预测是多变量的非线性系统,而神经网络解决非线性问题有其独特的优势,是目前用于储层敏感性预测是较好的方法。在建立预测方法过程中,BP网络隐层结点数的确定直接影响到网络的学习效率。通过对目前四种隐层结点数确定方法进行探讨,优选出储层敏感性预测中BP神经网络合理的隐层结点数,并在实际预测中进行应用,从而使预测结果更客观和符合实际。应用情况表明,该方法可大大缩短网络学习时间,从而提高学习效率,使网络以最快的速率达到收敛。

OPTIMIZATION AND APPLICATION OF BP NETWORK LATENT LAYER NUMBER IN PREDICTING RESERVOIR SENSIBILITY

Reservoir sensibility prediction is a multivariable nonlinear system.Neural network,whcih is a good method for reservoir sensibility prediction,has unique predominance in solving nonlinear problems.In establishing the prediction method,the latent layer node numbers of BP network have direct influence on the learning efficiency of network.Through discussion on the four methods used at present,the reasonable latent layer node numbers of BP neural network are optimized to predict reservoir sensibility.They have been applied to practical prediction,so the result is more objective.Application shows that the method can shorten network learning time and improve learning efficiency to convergence at quickest rate.