Prediction of coalbed methane content based on composite logging parameters and PCA-BP neural network

Abstract

The coalbed methane content (CBM) is a key parameter for the evaluation and efficient exploration and development of coalbed methane reservoirs. The traditional gas content experiment methods are time-consuming, costly, weak in generalization ability and large in calculation error. Therefore, accurate, efficient and low-cost calculation of CBM content is of great significance in CBM development. In this paper, the coalbed methane prediction model is constructed by exploring the hidden geological information between coalbed methane content and logging parameters. Firstly, principal component analysis and person method are used to analyze the correlation between each logging parameter, and then compound parameters are constructed to improve the correlation between each parameter. Finally, BP neural network model is used to build a CBM content prediction model based on compound logging parameters. On this basis, the prediction results of BP neural network model are compared with KNN, Ridge regression, random forest, XGBoost and other machine learning models, and the determination coefficient, root-mean-square error and relative error are used to evaluate the model. The results show that BP neural network is more suitable for constructing CBM prediction model with complex logging parameters, and the prediction effect is good, the relative error is 4.5 %, and the prediction accuracy is improved by about 61 % compared with other models. This model has potential application in the field CBM reservoir development, can predict the gas content of coal seam quickly and accurately, speed up the CBM reservoir development process, and provide a new method for coal seam exploration and reservoir logging evaluation.