摘要
为准确描述致密油藏开发过程中渗透率的动态变化规律.本文首先分析总结致密储层渗透率的3个主要影响——有效应力、温度和含水饱和度,并建立了以有效应力、温度和含水饱和度为输入层,渗透率为输出层,人工蜂群算法优化的BP神经网络渗透率预测模型.利用不同条件下的渗透率数据建立仿真数据样本进行训练,训练结果的最大绝对误差为0.063 47×10-3μm2,最大相对误差为4.382%,平均相对误差1.069%.说明人工蜂群算法优化BP神经网络模型很好的描述致密储层渗透率和各个影响因素之间的内在规律.常规BP神经网络渗透率模型训练结果平均相对误差达10.699%.说明较之常规BP神经网络,改进的BP神经网络预测精度及稳定性均有较大提升.综上,人工蜂群算法优化的BP神经网络模型能较好应用于低渗致密储层渗透率预测.
In the development of the low permeability tight reservoir,the permeability is influenced by so many factors,the common permeability model that based on single variable has trouble in accurately describing the permeability.In this paper,the main factors affecting permeability of the low permeability reservoir are analyzed,including effective stress,temperature and water saturation.And then we establish a BP neural network permeability prediction model optimized by the artificial bee colony algorithm,in which effective stress,temperature and water saturation are input layer nodes,permeability is output layer node.The permeability of different conditions is used to establish learning samples for training and predicting.Result shows that the maximum absolute error of the training results is 0.063 47×10-3μm2,the maximum relative error is 4.382%and the average relative error is 1.069%.It shows that the BP neural network model optimized by artificial bee colony algorithm,in other words,the hybrid neural network can accurately describe the internal relations and laws between the permeability and various influence factors.Whereas,the average relative error of the conventional BP neural network permeability model is 10.699%.Obviously,the hybrid neural network is more accurate and stable than the conventional BP neural network.In one word,the hybrid neural network has a wonderful adaptability to the permeability prediction of low permeability tight reservoirs.
出处
《陕西科技大学学报》
CAS
2018年第1期90-95,共6页
Journal of Shaanxi University of Science & Technology
基金
国家重大科技专项项目(2017ZX05032004-002)
关键词
致密储层
人工蜂群算法
BP神经网络
混合神经网络
渗透率预测模型
tight reservoir
artificial bee colony algorithm
back propagation neural network
hybrid neural network
permeability prediction model
