基于ARIMA-RTS和LSTM的致密气藏产能预测  

Productivity Prediction of Tight Gas Reservoir Based on ARIMA-RTS-RTS and LSTM

作　　者：付豪 方全堂[2] 杜一鹤 Fu Hao;Fang Quantang;Du Yihe(Well Completion Test Management Center of Sinopec Northwest Oilfield Company,Luntai 841600;School of Petroleum and Natural Gas Engineering,Southwest Petroleum University,Chengdu 610500)

机构地区：[1]中国石化西北油田分公司完井测试管理中心,新疆轮台841600 [2]西南石油大学石油与天然气工程学院,四川成都610500

出　　处：《石化技术》2023年第11期120-122,共3页Petrochemical Industry Technology

摘　　要：建立长短时记忆神经网络(LSTM)和基于卡尔曼平滑算法的自回归差分移动平均模型(ARIMA-RTS)模型,通过气井实际数据测试和拟合,调整参数后对比,LSTM模型表现较好,MSE误差为0.89%,但耗时较长。ARIMA-RTS模型MSE误差为2.6%,整体精度满足要求,耗时较短。In this paper,LSTM and ARIMA-RTS models are established.Through training andfitting of actual gas well data,the parameters are adjusted and compared.The actual data results show that the LSTM model performs well,with an MSE error of 0.89%,but it is time-consuming and complex.The ARIMA-RTS model has an MSE error of 2.6%,which meets the overall accuracy requirements and is time-consuming and simple.

关 键 词：致密气藏 时间序列 数据挖掘 产能预测 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]