BO-RF Based Shale Gas Recovery Predicting Method

China has abundant shale gas reserves, shale gas exploration and exploitation incur high costs, therefore predicting the shale gas recovery is of great significance for efficient exploitation. Random forest (RF) method is commonly used to predict shale gas recovery. However, the parameter adjustment of random forest algorithm is complex and inefficient. To solve this problem, this paper proposes a random forest shale gas recovery prediction method based on Bayesian optimization. Firstly, 12 characteristic variables such as well spacing, porosity, and water saturation were selected as input parameters through Spearman correlation coefficients, recovery of shale gas as output parameter. Secondly, Bayesian optimization is used to optimize the maximum depth of decision trees, the number of decision trees, the maximum number of features and other parameters in the random forest regression algorithm. Finally, a random forest measurement model based on Bayesian optimization (BO) is established using this parameter. The experimental results indicate that compared with three prediction models of SVR, RF and grid search optimized RF. The decision coefficient (R2) of the prediction model proposed in this paper reached 0.898, an average increase of 10%. It shows that the BO-RF method is effective in predicting shale gas recovery.