酸化期间基于BP神经网络法的井底压力计算

Calculation of bottomhole pressure during acidizing based on BP neural network method

为解决酸化期间井底压力直接获取难度大、常规计算方法计算精度低的难题,建立了一种基于BP神经网络法的井底压力计算方法 :根据井底压力计算多参数输入、单参数输出的特点,搭建出基于误差反向传播、正向反馈的三层结构模式的BP神经网络;利用历史井施工数据(井口压力、井底压力、流量、管柱尺寸、流体性质)在神经网络上开展训练,获得井筒摩阻系数计算的网络矩阵;后期把地面酸化施工数据输入该网络矩阵即可计算出井底压力数据。分析表明利用神经网络法计算井底压力最主要的影响因素是液体排量和液体性质,另外历史训练数据覆盖范围和数据量决定神经网络的适应性和计算结果精度。利用川渝气田灯影组15口酸化井历史施工数据开展训练,获得了适应于该区域气井井底压力计算的BP神经网络,并开展了4井次现场应用,与实测井底压力对比,最大误差7.8%,相比于常规井底压力计算模型(误差20%左右),精度得到大幅提升。

It is difficult to directly obtain bottomhole pressure during acidizing,and the calculation accuracy of conventional method is low.So,another method to calculate bottomhole pressure based on BP neural network was developed.First,according to calculation characteristics including multi-parameter input and single-parameter output,the three-layer BP neural network based on error back propagation and forward feedback is established.Second,the construction data of historical wells(wellhead pressure,bottomhole pressure,flow rate,pipe string size,and fluid properties)are used in the neural network for training,and the network matrix to calculate wellbore friction coefficient is created.And third,the ground construction data of acidizing is introduced into the matrix to calculate bottomhole pressure.Results show that fluid discharge and liquid properties are the most important influential factors when the neural network method is used to calculate the bottomhole pressure,and the adaptability and accuracy of the neural network are dominated by the coverage and volume of historical training data.Finally,the BP neural network suitable for calculating the bottomhole pressure in some gasfields of Sichuan-Chongqing areas has been established after training was carried out based on the historical construction data of 15 acidizing wells of Dengying Formation.And it has been applied to four wells,and the maximum error between the calculated bottomhole pressure and the measured one is 7.8%.Compared with some conventional calculation models(error about 20%),its calculation accuracy is improved greatly.