关于人工智能方法用于钻井机械钻速预测的探讨

Discussion on the application of artificial intelligence method to the prediction of drilling machinery ROP

石祥超 王宇鸣 刘越豪 陈雁

SHI Xiangchao WANG Yuming LIU Yuehao CHEN Yan

油气藏地质及开发工程国家重点实验室·西南石油大学 西南石油大学计算机科学学院

State Key Laboratory of Oil and Gas Reservoir Geology and Exploitation, Southwest Petroleum University, Chengdu 610500, Sichuan, China School of Computer Science, Southwest Petroleum University, Chengdu 610500, Sichuan, China

人工智能方法被广泛地用于预测钻井过程中的机械钻速，虽预测精度都能超过80%，但以前的算法大多仅选取一口井或一个开次的数据进行预测和检验，缺乏对邻井或整个区块推广和预测的研究，泛化能力有待检验。针对上述问题，讨论了相关性分析在机械钻速预测中对钻井参数选取的影响以及训练数据选取对人工智能模型泛用能力问题。引入地层参数、钻头参数及钻井参数作为输入参数，选取四川盆地某区块的实际钻井数据进行训练，评价了随机森林、支持向量机、梯度提升树、人工神经网络4种人工智能算法对整个区块机械钻速预测的精度。结果显示，随机森林算法对区块内各单井数据的预测精度能达到90%，对整个区块数据预测的准确度能达到88%，且使用区块数据训练的随机森林模型具有较好的泛化能力，认为该方法能够推广至整个区块，有利于指导该区块的钻井工程技术优化。

Artificial intelligence methods are widely used to predict the rate of penetration in the drilling process. Although the prediction accuracy can exceed 80%, most of the previous algorithms only select the data from one well or one drilling section for prediction and inspection, lacking the research on the generalization and prediction of adjacent wells or the whole block. Therefore, the generalization ability of these algorithms needs to be tested. In view of the above problems, the influence of correlation analysis on drilling parameter selection during prediction the rate of penetration and the influence of training data selection on the generalization ability of artificial intelligence models were discussed. The formation parameters, drill bit parameters and drilling parameters were introduced as input parameters, and the actual drilling data of a block in the Sichuan Basin was selected for training. The accuracy of 4 artificial intelligence algorithms, that is random forest, support vector machine, gradient boosting tree and artificial neural network, were evaluated for predicting the penetration rate of the entire block. The results show that the prediction accuracy of the random forest algorithm for the data of each single well in the block can reach 90%, and the prediction accuracy for the data of the whole block can reach 88%. The random forest model trained with the block data has better generalization ability, it is believed that this method can be extended to the whole block, which is beneficial to guide the optimization of drilling engineering technology in this block.

10.13639/j.odpt.2022.01.016