本文提出一个输入层为分支延迟线形式的反向传播神经网络,可以有效地利用一条或多条测井曲线进行训练,从而识别地质标志层。接下来,应用训练好的神经网络预测出研究区其它井上该标志层的位置。其后,应用与第一个神经网格结构类似的另一网络完成第二个标志层的训练和识别。在此期间,除利用测井曲线外,还要以第一个标志层的深度参考函数作为辅助输入。研究结果表明,该方法较标准剖面相关性对比技术具有更好的性能和识别能力。若在工作站上由地质家进行交互处理则其应用效果会更佳。 In this paper, we propose a backpropagation neural network whose input layer is in the form of a branch delay line, which can be effectively trained by one or more well logs to identify the geological marker layer. Next, the well-trained neural network is used to predict the location of the marker layer on other wells in the study area. Thereafter, another network similar to the first neural grid structure was used to train and identify the second landmark layer. During this period, besides taking advantage of the well logging curve, the depth reference function of the first marker layer should also be used as the auxiliary input. The results show that this method has better performance and recognition ability than the standard cross-correlation technique. If the workstations by the geologists to interact with its application will be better.