延迟神经网络测井资料评价水淹层的方法及应用

A METHOD OF TIME-DELAY NEURAL NETWORKS FOR WELL LOG INTERPRETATION OF WATER DRIVE

以取心分析的孔隙度和水饱和度为基础，建立测井信息与这些地居参数之间的非线性计算关系。用这种方法进 行参数解释，测井信息与油层参数之间的复杂关系不需要具体的数学物理模型描述，而只需要合适的样本集对网络进行 训练来获得解释模型，避开了油层水淹后，混合水电阻率求不准的问题。不同于已有的点对点的建模方法，本方法采用延迟神经网络模型，在建模和计算过程中自动考虑了测井响应上下围岩的影响。从而较好地解决了测井资料解释中地层厚度自适应校正和地层参数计算同时进行的问题。对一个油田注水开发后期 ５０口井的测井资料计算表明，这种方法具有良好的效果。

A method of time-delay neural network (TDNN) to calculate the parameters of flooding reservoir from well logging data is presented. This method is based on porosity,water saturations (Swi,Sw) of core to build the non-linear models of well logging data and the porosity,water saturations,which avoid the problem of conventional log interpretation to work out the resistivity of formation water. It is different from common BP neural network. The structure of TDNN enables the network to discover the depth varying feature of multiple well logging information in depths. When the TDNN has been trained using the core porosity,water satura- tions and well logging data,it built the non-linear relationship between multiple wireline logs that contain information of formation in adjacent depths and parameters of the reservoir. Examples of fifty wells of Henan Oilfield showed that the method is effective.