1. PetroChina Huabei Oilfield Company,Renqiu 062550,China; 2. School of Mechanics & Civil Engineering,China University of Mining and Technology(Beijing),Beijing 100083,China

The evaluation and prediction of coal bed methane ( CBM) is the critical basis of CBM exploitation schemes. Random Forest algorithm performs well in the evaluation and prediction of CBM,which has the advantages of small computation and high accuracy. The production of CBM is controlled by the geological parameters,engineering measures and extraction process parameters. The geological parameters of coal reservoir are divided into dynamic parameters and static parameters. The static geological parameters,such as the buried depth of the coal seam,the thickness of the coal seam,and the ground stress,are determined by the essential properties of the coal seam. Dynamic geological parameters,such as reservoir pressure and permeability,change dynamically in the process of pump and production. Mainly controlled by human operation,the pumping process parameters are mostly dynamic parameters,including bottomhole pressure,casing pressure,dynamic liquid level,pumping speed and stroke,which play an important role on CBM production,when the coalbed methane well enters the production stage after site selection,drilling and hydraulic fracturing stages. According to random forest algorithm,we analyzed how geological parameters and drainage parameters affect the gas production and ranked the parameters impacting on the gas production of coalbed methane well:flow pressure>casing pressure>dynamic liquid level>stroke>buried depth. The production data of coalbed methane wells in the latest 60 days were used as the test sample of gas production prediction,and the historical production data were used as the learning sample. The learning sample data were input into the R Programming Language after processing the missing values and abnormal production data,and Random Forest algorithm was used to fit and analyze the CBM production data. The production model of CBM wells was established by considering the effects of the process parameters and the dynamic changes of historical gas production on the subsequent daily gas production of CBM. Based on the branching goodness criterion of Random Forest algorithm,the model predicted the daily gas production of coalbed methane wells under different pumping schemes. After comparing the predicted value with the test sample,we found that the error between more than 95% of the predicted daily gas volume and the actual production data ( test samples) is less than 5% ,which means the production model of CBM vertical wells based on Random Forest algorithm has high fitting and prediction accuracy,providing a new method for CBM well productivity evaluation and prediction.

国家科技重大专项资助项目(2017ZX05064)

ZHU Qingzhong,HU Qiujia,DU Haiwei,et al. A gas production model of vertical coalbed methane well based on random forest algorithm[J]. Journal of China Coal Society,2020,45(8):2846-2855.