- 1School of Geosciences, China University Of Petroleum(East China), Qingdao, China (1172511787@qq.com)
- 2School of Geosciences, China University Of Petroleum(East China), Qingdao, China (zhangsaimeng2001@outlook.com)
- 3School of Geosciences, China University Of Petroleum(East China), Qingdao, China (zjh@upc.edu.cn)
- 4School of Geosciences, China University Of Petroleum(East China), Qingdao, China (gzp_upc@163.com)
One of the key points of oil and gas exploration is the accurate description of reservoir thickness. However, due to the complexity of sand overlapping structure, the actual well earthquake relationship is poor, and the correlation between single seismic attribute and sand body thickness is weak, so the sand body thickness cannot be accurately predicted. In this paper, 8 kinds of seismic attribute information are extracted and selected, and the LightGBM model optimized by Newton-raphson-based optimizer (NRBO) is used to predict reservoir thickness with multiple attribute combination. It is found that the arc length, average amplitude, bandwidth, energy half an hour and other attributes of the selected working area are strongly correlated with the thickness. Meanwhile, the influence of the ratio of validation machine on the prediction results is studied. When the ratio of verification set is 20%, the best prediction effect is obtained, and the effect of the optimized model is significantly improved compared with the traditional machine learning methods such as LightGBM. The study of NRBO-LightGBM model in the prediction of sand body thickness has great popularization value and reference significance.
Newton-raphson-based optimizer (NRBO) is a new meta-heuristic optimization method, which is inspired by two key principles: Newton-Raphson search rule (NRSR) and trap avoidance operator (TAO). NRSR uses Newton-Raphson method to improve the exploration capability of NRBO and increase the convergence rate to achieve improved search space position. TAO helps NRBO avoid the local optimal trap. NRBO has the characteristics of strong evolutionary ability, fast search speed and strong optimization ability. This algorithm was proposed by Sowmya et al in 2024.
A large sample of machine learning reservoir thickness prediction research is carried out. Part of the data samples selected in this paper are actual samples, and part are thickness information predicted by SVM model. Fig.1 shows the prediction results when the proportion of test sets is 20%.
Fig.1 Comparison of well point thickness prediction between NRBO-LightGBM and LightGBM
How to cite: Zhang, W., Zhang, S., Zhang, J., and Gui, Z.: Research on reservoir thickness prediction of river channel sandstone based on ensemble learning model, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2756, https://doi.org/10.5194/egusphere-egu25-2756, 2025.
