煤层气高产井和低产井预判方法研究

A Pre-Judge Method for High Production Wells and Low Production Wells of Coalbed Methane

摘要: 煤层气单井产量低一直是制约煤层气开发利用的主要技术瓶颈。为实现煤层气效益开发和可持续发展，急需对煤层气低产井、高产井进行准确预测，从而进行科学决策和技术革新。根据单井产气量，将煤层气井划分为高产井和低产井，对于影响煤层产气量的地质工程参数，先利用灰色关联分析法进行重要性排序，再通过交叉验证法(CV)进行优选；以典型的DJ区块为例，影响煤层产井气量的主控因素为声波时差、体积密度、自然伽马、深侧向电阻率、总液量、总砂量和施工排量。在此基础上，优选并引入AdaBoost算法，提出高产井和低产井的预判方法，预测符合率达到93.33%。最后，对不同方法、不同样本个数、不同划分标准、不同区块进行讨论和对比分析，结果表明AdaBoost算法效果明显优于BP神经网络等传统方法，提出的预判方法适用于不同区块，具有一般性，对实现煤层气效益开发和可持续发展具有指导意义。

Abstract: The low production of single well of coalbed methane (CBM) has always been the main limitation. Propose methods for predicting CBM high production wells and low production wells to achieve beneficial development and sustainable development of CBM, so as to carry out scientific decision-making and technological innovation. First of all, the article divided all wells into high production wells and low-production wells according to the average daily CBM production of each well. For the geological and engineering parameters affecting CBM production, the importance ranking was carried out by grey correlation analysis method, and then the main control factors were optimized by cross validation (CV). Taking the typical DJ block as an example, the main control factors affecting CBM production were AC, DEN, GA, RD, total fracturing fluid volume, total sand volume and construction displacement. On this basis, AdaBoost algorithm was optimized and introduced to put forward the pre-judge method of high-production wells and low-production wells, and t he pre-judge coincidence rate reached 93.33%. Finally, different methods, different sample numbers, different division standards of production and different blocks were discussed and compared. The results showed that the effect of AdaBoost algorithm was obviously better than traditional methods such as BP neural network. The proposed prediction method is applicable to different blocks and has generality. It has guiding significance for realizing the benefit development and sustainable development of CBM.