BP神经网络预测最小混相压力

Prediction of minimum miscibility pressure with BP neural network

任双双1, 杨胜来1 ,沈 飞

中国石油大学石油工程教育部重点实验室，北京 102200 江苏石油勘探局石油工程技术研究院，江苏 扬州 225009

MOE Key Laboratory of Petroleum Engineering, China University of Petroleum, Beijing 102200, China Research Institute of Petroleum Engineering and Technology, Jiangsu Petroleum Exploration Bureau, Yangzhou 225009, China

应用BP神经网络预测CO2最小混相压力，选择C5 + 分子量、油藏温度、挥发油（CH4和氮气）的摩尔分数、中间油（C2—C10）的摩尔分数作为参数，用相关文献的实验结果作为样本进行训练，选取网络模型各层函数、隐含层节点数和算法得出适合的BP神经网络，结合实际细管实验的数据及相关参数修改网络输入参数应用于实际油藏，预测最小混相压力并分析相关的影响因素，指导生产和相应理论研究。

This paper presents a BP neural network model for the prediction of CO2 minimum miscibility pressure. We select the molecular weight of C5+ fraction, reservoir temperature, mole fraction of volatile oil(methane and nitrogen gas) and mole fraction of intermediate fractions(C2-C10) as parameters, train the BP neural network by the related experimental data and pick out the functions of network model, number of hidden neurons in hidden layers, and relative algorithm in order to construct the applicable BP neural network. At last the combination of slim tube experimental data with relative parameters can be used to modify and adjust the input parameters for actual reservoir. We predict the minimum miscibility pressure by the BP neural network and analyze the relative effect factors in order to instruct the production and to conduct the relevant theoretical research.