地质科技通报 (Sep 2021)

Stratigraphic lithology identification based on no-dig Logging While Drilling system and random forest

  • Han Xu,
  • Kongxuan Yao,
  • Danyi Cheng,
  • Qiangyin Song,
  • Zhiming Ma,
  • Xuming Zhu,
  • Xiaoming Wu,
  • Guanhui Zhao,
  • Xiaochun Cai

DOI
https://doi.org/10.19509/j.cnki.dzkq.2021.0039
Journal volume & issue
Vol. 40, no. 5
pp. 272 – 280

Abstract

Read online

Through the self-developed and designed no-dig Logging While Drilling(LWD) system, it can collect the parameters of no-dig drilling, identify the real-time formation lithology of no-dig drilling, and provide safety information guarantee for no-dig construction.In view of the lack of prospecting data in no-dig engineering, it is difficult to determine the lithology of the excavation stratum.A real-time data acquisition system based on the no-dig LWD system is proposed.The random forest algorithm is used to establish the stratum identification model, and identify the unknown strata.The identification results are displayed visually.Through the practical application of the detection while drilling system in the engineering field, the drilling sensitive parameters including Rate of Penetration(ROP), torque, rotation speed, pulling force, pump pressure and pump volume are obtained as training samples.The random forest algorithm is used to train the collected drilling parameters, and the decision tree and random forest are constructed to classify the drilling parameters.A classification model aiming at the classification of typical no-dig strata is established, and the classification labels of miscellaneous fill, clay, silty fine sand, gravel and silt are determined respectively.Furthermore, based on the classification results of machine learning, PCA principal component analysis is used to reduce the dimension of strata recognition features to three-dimensional, and realize the three-dimensional display of formation lithology identification results.The prediction model is applied to practical engineering to verify its effectiveness.The results show that the method can quickly identify the drilling formation under the condition of no-dig real-time drilling, and the recognition accuracy is as high as 92%.The research results provide important information for the selection of no-dig mud and no-dig reaming stage.

Keywords