基于BP神经网络的气井配产方法

Gas well production allocation method based on BP neural network

气井合理配产是保障气田高效、稳定、持续生产的关键前提,针对延安气田目前生产井数多且投产时间差异大、气井配产困难的问题,优选原始地层压力、井控储量、地层系数、表皮系数、稳产时间作为输入神经元,运用反向传播(back propagation,BP)神经网络对91口气井的资料进行网络训练和学习,建立合理产量预测模型,并对5个参数进行影响程度分析和敏感性分析。结果表明:模型预测91口气井的产量平均误差为1.77%,预测结果真实可靠;所选择的5个神经元对合理产量均有显著影响,且影响程度从大到小依次为稳产时间、井控储量、地层系数、表皮系数、原始地层压力;原始地层压力越大、井控储量越大、地层系数越大、表皮系数越小,一定稳产时间内气井合理产量就越大。

Reasonable gas well production allocation is the key prerequisite to ensure efficient,stable and continuous production of gas fields.Currently,Yan’an gas field has a large number of producing wells with remarkably distinct production time,which brings difficulty in production allocation of producing gas wells.The original formation pressure,well-controlled reserves,formation coefficient,skin coefficient,and stable production time are selected as input neurons,and the back propagation(BP)neural network is used to conduct network training and learning on the data of 91 gas wells.Then,a reasonable yield prediction model is established,and the influence degree and sensitivity analysis on 5 parameters is conducted.The results show that:the average error of the production of 91 gas wells predicted by the model is 1.77%,and the prediction results are true and reliable;the selected five neurons have a significant impact on reasonable production,and the influence degree decreases in the order of the stable production time,well-controlled reserves,formation coefficient,skin coefficient,and original formation pressure;the greater the original formation pressure,the greater the well-controlled reserves,the greater the formation coefficient,and the smaller the skin coefficient,the greater the reasonable production of gas wells within a certain stable production period.