综合运用BP网络和优化算法建立储层敏感性预测模型

Establishing formation sensitjVity model with BP networks and oPtimized calculations

通过对近年来储层敏感性预测方法的分析研究,认为神经网络方法是一种较理想的预测储层敏感性的新方法.但是常规的BP算法存在收敛速度慢、局部根小值等缺点。为此，采用了动量自适应学习率调整方法和L-M优化算法，效果明显改善，其中L-M优化算法效果最好，收敛速度快，误差最小。利用L-M优化方法建立的储层速敏网络模型，预测渗透率损害程度的准确率达93%，基本上满足了油气层敏感性预测的需要。

This paper reviews the predicting methods of forma-tion damage appeared in recent years，and it con-cludes that the BP(back propagation) Model is amore practical one，though there are disadvantages over slow collection speed and local minimums.New methods are presented and evaluated to improve this model，e.g.，momentum method，self一adaptation，and Levenberg一Marquardt(L一M)method.In which，theL一M method 15 a better one and the Pre-d icting accuracy witll this nletllod for BP niodel reaches as high as 93%.