摘要
针对研究区目标储层典型页岩油藏物性较差导致全区及中高初产油井产能递减的问题,笔者提出一种长短期记忆神经网络(LSTM),选取研究区生产时间大于48个月全区油井以及中高初产递减型井的平均月生产数据并对其未来4个月的产能进行预测。结果表明,长短期记忆神经网络(LSTM)产能预测相比于传统的指数递减规律对产量进行拟合的误差更小,预测结果更为可靠。从而为后期油田生产以及后续开发采取的工程措施提供理论依据。
With the development of petroleum industry,unconventional oil and gas have been playing an important role in the whole petroleum market.The target reservoir in the study area is a typical shale reservoir with poor physical properties,therefore,the formation of the main reservoir by volume fracturing means.However,in the process of development in recent years,it is difficult to determine the main factors affecting the production capacity of the study area,it is a typical non-linear characteristic prediction,and the traditional productivity prediction is not ideal in applicability and precision,which leads to great differences in development effect,and brings great difficulties to the stable and increasing production of oil field,the real-time prediction of productivity for the whole area and the middle-high initial production decline wells in the study area is of great significance to the evaluation of late fracturing effect and the optimization design.In this paper,we propose a new Long Short-Term Memory(LSTM),the productivity of the oil wells in the study area in the next four months is forecasted by using the average monthly production data of the oil wells in the study area whose production time is more than 48 months and the declining initial production of the medium and high initial production wells,compared with the traditional law of exponential decline,the LSTM capacity forecast has less error in fitting the Long Short-Term Memory,and the forecast result is more reliable.Therefore,it can provide theoretical basis for later oil field production and subsequent development engineering measures more efficiently.
作者
苏存娃
王子龙
鲁鹏
SU Cunwa;WANG Zilong;LU Peng(Xiasiwan Oil Production Plant of Yanchang Oil Field Co.Ltd.,Yanan 716100,China;Exploration and Development Technology Research Center of Yanchang Oilfield Co.,Ltd.,Yanan 750021,China)
出处
《北京石油化工学院学报》
2024年第1期50-55,共6页
Journal of Beijing Institute of Petrochemical Technology
关键词
产能预测
长短期记忆神经网络
时序产能预测
中高初产递减型井
productivity prediction
long short-term memory
time series productivity prediction
medium and high primary production decline type wells
作者简介
苏存娃(1977-),男,本科,工程师,研究方向为油田开发,E-mail:1762301870@qq.com;通信联系人:王子龙(1998-),男,硕士,助理工程师,研究方向为油田勘探,E-mail:135334936@qq.com。
