基于神经网络的油气预测方法

以濮阳凹陷白庙构造为研究区，以沙二下亚段第1砂层组为目的层，根据神经网络原理，选取能够反映储层参数及油气分布特征的“S”型函数，构造出可以产生任意复杂判断的三层感知器，利用误差反传播算法，使估价函数最小化。提取20余种地震波反射特征参数，将地震道旁的特征参数输入到神经网络中，使其学习、训练和记忆，给定神经网络各个节点间的连接权值和节点内部的阅值，通过神经网络所记忆的知识，用钻井、测井、地质、试油等资料进行约束，对未知区域的储层特征参数和含油气分布进行预测，取得了较好的效果。

Neural Network Prediction Method and its Application to Oil and Gas Forecast

Neural network was used to forecast the oil and gas distribution of Baimiao geological structure in Puyang, P.R.China . To get optimal forecast results, we took a certain layer as the destination layer and used Sigmoid function to reflect both the reservoir parameter and the character of oil and gas distribution. Thus, we applied neural network method to construct the three layer perception and we minimized the cost function with error back propagation algorithm. We seleced more than 20 parameters of seismic wave echo characteristic and we considered seismic path side characteristic parameters as the inputs of the neural network. By means of gradient optimal method, we determined invisible layer parameters and thresholds of neural networks. Our neural network prediction method gave satisfactory forecast results for the Baimiao geological structure in Puyang, P.R.China as explained in section 5 and shown in Figs.2, 3, and 4.