基于ARIMA-Kalman滤波器数据挖掘模型的油井产量预测

1)中国石油大学(华东)石油工程学院,山东青岛 266580; 2)中国石化胜利油田分公司勘探开发研究院, 山东东营 257015; 3)中国石化胜利油田分公司胜利采油厂,山东东营 257015

油气田开发工程; 时间序列; 产量预测; 数据挖掘; ARIMA模型; 卡尔曼滤波器

Oil well production forecasting method based on ARIMA-Kalman filter data mining model
GU Jianwei1, SUI Gulei1, LI Zhitao1, LIU Wei1, WANG Yike1, ZHANG Yigen2, and CUI Wenfu3

1)School of Petroleum Engineering, China University of Petroleum(East China), Qingdao 266580, Shandong Province, P.R.China2)Institute of Petroleum Exploration and Development, Shengli Oilfield Branch Company, SINOPEC, Dongying 257015, Shandong Province, P.R.China3)Shengli Oil Production Plant, SINOPEC, Dongying 257015, Shandong Province, P.R.China

oil and gas field development engineering; time series; production forecast; data mining; autoregressive integrated moving average(ARIMA)model; Kalman filter

DOI: 10.3724/SP.J.1249.2018.06575

备注

影响水驱开发油田产量的因素众多,针对常规产量预测方法无法考虑时序影响因素的非同步性以及滞后性,应用时间序列分析方法,结合卡尔曼滤波器(Kalman filter),建立考虑因素动态关系的产量ARIMA-Kalman滤波器时间序列模型.根据历史产量数据建立时间序列中的产量差分自回归积分移动平均(autoregressive integrated moving average, ARIMA)模型; 再将ARIMA模型与Kalman滤波器相结合,构建产量预测算法; 以实例油田资料开展机器学习和数据挖掘,并采用数据拟合及预测检验评价算法合理性,实现最终产量数据预测.研究结果表明,ARIMA-Kalman滤波器具有高效的时序影响因素的分析能力,能够排除非同步性和滞后性的影响,使识别出的产量时间序列模型具有精准的拟合结果和预测能力.该研究可为油田产量预测提供一种有效方法,为后续的油井开采提供决策和理论依据.

There are many factors affecting oilfield production in water flooding development. The conventional production forecasting method cannot consider the effects of asynchronization and hysteresis of the timing factors. In this paper, the time series analysis is adopted by combining with Kalman filter to establish the ARIMA-Kalman filter time series model of production considering dynamic relationship. Firstly, the ARIMA(autoregressive integrated moving average)model time series model of production is established according to the production history data. Secondly, the ARIMA model is combined with Kalman filter to build an oilfield forecasting algorithm. Finally, we carry out machine learning and data mining for the actual oilfield data, and apply data fitting and predictive testing to evaluate the rationality of the new algorithm, then achieve the ultimate oilfield production forecasting. The research results show that ARIMA-Kalman filter has the ability to analyze time series factors by eliminating the effects of asynchronization and hysteresis. The identified production time series model can deliver accurate fitting results and predictive ability. This research work can provide an effective method for oilfield production forecasting, and deliver reliable decision-making and basis for subsequent oil well production.

·