An eXtreme Gradient Boosting Algorithm Combining Artificial Bee Colony Parameters Optimized Technique for Single Sand Body Identification
Due to the problems of traditional artificial single sand body identification methods such as strong subjectivity, heavy workload and low efficiency, we propose a fast and objective ABC-XGBoost. The algorithm consists of two parts: eXtreme gradient boosting (XGBoost) and artificial bee colony algori...
Guardado en:
| Autores principales: | , , , , , |
|---|---|
| Formato: | article |
| Lenguaje: | EN |
| Publicado: |
IEEE
2021
|
| Materias: | |
| Acceso en línea: | https://doaj.org/article/3101e991a6bd4b168f74f5c264897432 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Sumario: | Due to the problems of traditional artificial single sand body identification methods such as strong subjectivity, heavy workload and low efficiency, we propose a fast and objective ABC-XGBoost. The algorithm consists of two parts: eXtreme gradient boosting (XGBoost) and artificial bee colony algorithm (ABC). XGBoost introduces a regular term, which can effectively prevent overfitting, and uses the second derivative to make the identification result more accurate. However, a large number of parameters in XGBoost need to be adjusted manually, which affects the efficiency of the algorithm. In this regard, ABC is used to optimize the parameters based on XGBoost, and then the single sand body can be identified quickly and effectively. We take the <italic>C6</italic> <inline-formula> <tex-math notation="LaTeX">$_{1}^{2}$ </tex-math></inline-formula> oil-bearing layer in the second area of Dalugou, Jing’an Oilfield as the research object, and use the ABC-XGBoost to identify the single sand body in the research area. Based on the reasonable selection of physical parameter data and logging data, the partition and interlayer data should be eliminated first to avoid data redundancy. The results indicate that ABC-XGBoost is more efficient and accurate than the existing mainstream machine algorithms, such as support vector machines (SVM), random forests (RF), and XGBoost using trial and error tuning under the same logging data and computer hardware conditions. The accuracy can reach 90.6%, which has certain practical application value in the middle and late development of oil and gas fields. |
|---|