摘要
考虑到开发过程中影响聚合物驱油效果的动态因素,如聚合物注入浓度、注入速度和采油速度等,提出了一种基于三层前向神经网络的聚合物驱含水率预测动态模型。该模型在训练神经网络时采用了改进的Levenberg-Marquardt算法,从而避免了矩阵求逆运算。对孤东油田七区所建模型的研究结果表明,应用该方法可以获得与实际结果拟合均方差为0.0460的聚合物驱含水率预测模型。应用该方法与应用数值模拟预测含水率的拟合均方差仅为0.3608,表明该方法可用于与具有相似地质特征区块的聚合物驱含水率建模预测。
After the dynamic factors of polymer flooding are considered, such as injection concentration, oil production rate, injection rate, a dynamic predicting model of water cut during the polymer flooding is proposed based on three-layer forward neural network. This model uses modified Levenberg-Marquardt algorithm when neural network training,avoiding converse operation of matrix. Comparing with the prediction result of numerical simulation for Gudong Oilfield, the quadratic mean deviation of the water cut in the polymer flooding exploitation between observed and predicted data is 0. 0460. The quadratic mean deviation of predicted water cut that calculated by this method and numerical simulation is 0. 3068. It indicates that the modeling method can be used to predict water cut of polymer flooding of other unit with similar geological characteristics.
出处
《油气地质与采收率》
CAS
CSCD
北大核心
2007年第5期56-58,65,共4页
Petroleum Geology and Recovery Efficiency
关键词
聚合物驱
含水率
预测
神经网络
polymer flooding, water cut, prediction, neural network