摘要
油气井压裂后要获得好的增产效果,首先要选取最适合进行增产措施改造的井。影响一口井压裂效果的参数包括地质静态、开发动态及压裂施工等多方面参数,从理论上讲,每一个参数对压裂效果都有不同程度的影响。传统的选井方法主要依靠经验或生产需求来选择施工井层,具有一定的主观性、盲目性和风险性。本文充分运用现代数学理论,即模糊聚类、模糊神经分类系统和模糊排序法,筛选了可靠的数据,确定了压裂选井主要影响因素;深入研究了影响压裂成功率的主要因素并应用层次分析法确定各因素的权重,研究了各候选井压裂成功率的大小;同时,充分运用计算机智能技术,提出了改进的自适应遗传算法,将该遗传算法与神经网络结合起来,形成了改进的遗传神经网络,该网络收敛快、泛化能力强,克服了传统数学方法在处理这类问题时的局限性和误差大的缺点,可以用来优选压裂井。应用效果表明,该方法预测结果与现场实施压裂后的增产效果非常吻合。
To obtain expected fracturing treatment results,the most important thing is to select appropriate stimulation candidates. However, there are many parameters influence postfracture response, such as static geology parameters, dynamic development parameters and fracturing design parameters. In theory, every parameter has different effect on postfracture response. The traditional methods of selecting the candidates which depend on experience or production demands,possess definite subjectivity and great risk. The article selects reliable data from the database to determine the primary factors that affect the candidate selection by using modern mathematics theory. The primary factors that affect the success ratio of fracturing have been researched and the weights of every factor have been decided by using the layer analytical method, at the same time, the fracturing success ratio of the candidate wells have been researched. The paper puts forward advanced autoadapted genetic algorithm and combines the genetic algorithm with the neural network to form improved genetic neural network by using the computer intelligent technology. The genetic neural network has high convergent speed and strong generalization ability, and it can overcome the limitations and great declination of traditional mathematical methods. The case study shows that the predicted result of the method is very close to the actual increasing production effect after fracturing.
出处
《钻采工艺》
CAS
北大核心
2006年第6期53-55,共3页
Drilling & Production Technology
基金
国家技术研究发展计划"863计划"(863-306-ZT04-03-3)成果的一部分。
关键词
压裂
选井
遗传算法
人工神经网络
fracturing, well choosing, genetic algorithm, artificial neural networks
