基于 BP 神经网络的叠前流体识别方法

探讨塔里木盆地桑塔木地区三叠系辫状河三角洲沉积储层流体识别方法。该地区储层横向变化大，流体类型复杂。本文提出利用 BP 神经网络的信息整合模式识别功能来进行储层流体识别的方法，通过叠前反演得到对流体敏感的弹性参数数据体和电测解释结果标定建模样本，采取随机抽样形成建模样本集与测试样本集，选取26口井的700个样本作为学习样本，62个作为测试样本，建立 BP 神经网络模型。预测结果和实钻结果吻合程度高，正确率达90％以上。该方法可以很好地对桑塔木地区储层中所含流体进行识别。

Study of pre-stack fluid identification method based on BP neural network

The delta deposits of Triassic reservoirs in Sangtamu area of Tarim oilfield vary laterally,so it is very difficult to identify the fluid type in the reservoir.Discrimination of reservoirs is vital to oil and gas exploration in this formation.Therefore,a method of fluid identification in reservoir by application of distinguishing oil-bearing layers from water-bearing or dry layers by BP neural network is proposed.Pre-stack sensitive elastic parameters inversion data and logging interpretation results are used to generate training samples,and to divide the training samples into subsets of modeling building and verification by adopting random sampling.Accordingly,700 training samples and 62 test samples from 26 wells are used to build BP neural network.It shows that the success rate is more than 90%and the model is used to predict the whole sand formation successfully.The practice indicates that the method is suitable for fluid identification in the study area.