Application of Random Forest Algorithm with Bayesian Optimization to Reservoir Prediction: Application of Random Forest Algorithm with Bayesian Optimization to Reservoir Prediction

Reservoir prediction in the actual work area is usually based on the characteristics of logging data, combined with the experience of experts in the work area to carry out manual judgment, which is not only time-consuming but also has a strong subjective factor. The parameter information of conventional logging data is enough, but when reservoir prediction is extended to three-dimensional seismic data, only elastic parameters can be used. Therefore, it is particularly important to predict reservoirs based on a limited number of elastic parameters. The random forest algorithm has achieved good results in reservoir prediction of well logging data. How to extend the random forest algorithm to seismic data for reservoir prediction, improve its performance, and avoid the phenomenon of local optimal solutions in the algorithm is still a problem that needs to be solved. In this paper, we construct a variety of sensitive parameters from elastic parameters such as longitudinal wave velocity, transverse wave velocity and longitudinal and transverse wave velocity ratio, and establish a set of reservoir prediction method flow with Bayesian optimization of Random Forest. For the sample imbalance problem in the training samples, adaptive sample weighting is introduced to improve the attention to the less class samples, and the results show that the method is more accurate than the traditional random forest in reservoir prediction on the logging-seismic scale. The final reservoir prediction accuracy on seismic scale reaches 86.4%.