Abstract: The varying characteristics of composite water cut in the area of polymer flooding and multi factors affecting the results of polymer flooding were analyzed.Taking these factors as inputs and the varying characteristics of the composite water cut as outputs,and taking the known inputs and the outputs of the early production areas as learning samples,a modified model for the three-layer cumulative back propagation (CBP) neural network was established.Some variables of the output layers may be unknowable in the learning samples in this model.Therefore,the all areas with the different production histories can be put into the learning samples,and the unknown points can be predicted.This method can avoid the deviation of the composite water cut in the industrial areas predicted by mode chart method and human correction and used to quantitatively analyze the various factors influencing the dynamic performances of polymer flooding.This model has been successively applied to predict composite water cut,liquid production and oil production of some subsequent production areas.The prediction result may be the good bases for making oilfield development plan.

Key words: polymer flooding, influencing factors, water cut, prediction, neural network model

摘要： 分析了工业化聚合物驱区块综合含水率的变化特征及其多种影响因素,把影响因素作为输入参数,把综合含水率的变化特征作为输出参数,以早期投产区块的已知输入和输出参数作为学习样本,建立了改进的三层CBP神经网络模型.在模型的训练样本中允许一些未知元素作为输出层变量,这样油田开发中时间不同的区块可以同时放在训练样本中,未知的点能通过预测而得到.该方法解决了以往的模式图方法预测工业化区块综合含水的偏差和人为的修正问题,并能够定量地分析各因素对聚合物驱动态特征的影响程度.利用该模型预测了新投产区块的综合含水率、产液量和产油量等指标,为油田开发规划的编制及计划安排提供了较为合理的依据.

关键词: 聚合物驱, 影响因素, 含水率, 预测, 神经网络模型