利用RBF神经网络自适应调整算法预测储层产能

传统径向基(RBF)神经网络的基函数宽度值都由经验公式确定或人为选取,没有考虑误差分布情况,所以在应用中常常效果不够理想。为此本文以最近邻距离算法为基础提出一种改进的RBF神经网络算法,能够自适应地调整基函数中心点宽度值以达到高精度、快速逼近样本的目的。在石油储层产能的预测中,首先对样本信息进行属性约简、预处理;然后在网络学习训练中通过计算输入样本的RBF神经网络的中心点值,再确定RBF神经网络基函数的宽度值,直到满足系统精度为止;最后用训练好的网络来进行储层产能预测。仿真结果表明,改进的RBF神经网络算法应用效果显著,不仅比传统RBF神经网络算法拟合精度高,而且收敛速度快。

Using RBF neural networks adaptive adjustment algorithm to predict production capacity of reservoir

The width value of basic function of ordinary RBF neural networks is determined from experimental formula or selected by human and without considering the errors distribution, so that the effects are often not ideal during application. For that reason,the paper presented an improved RBF neural networks algorithm based on the nearest adjacent distance that can adaptively adjust width value of central points in basic function in order to attaint the goal approaching samples with high precision and high speed. In prediction of production capacity of oil reservoir, first of all, reduction and preprocessing of samples information were implemented ; then, determination of the width value of basic function of RBF neural networks was conducted by input of the value at central point of RBF neural networks of samples during studying training of networks until to meet precision of system ; finally, using trained networks to predict the production capacity of reservoir. The results of simulation showed that improved RBF neural net-works algorithm had good application effects and both in high fitting precision and in high speed in comparison with ordinary RBF neural networks algorithm.