FOA Application in the GRNN Optimization for CBM Productivity Prediction

With the development of computer technology, the relationship between machine learning and the application of Coalbed Methane (CBM) gas has become closer, i.e., using machine learning to predict the CBM production capacity can not only improve the prediction accuracy but also significantly lower the prediction cost. This paper proposes an improved fruit fly optimization algorithm (FOA)-GRNN model for CBM production prediction based on the CBM production capacity classification. The proposed model can predict different production capacity types by integrating logging, fracturing, and drainage data of CBM wells. It employs the FOA to optimize the GRNN network smoothing factor to improve algorithm accuracy by improving the optimization target. The proposed model is verified by comparison experiments with linear regression, local weighted regression, RBF regression, traditional GRNN regression, and other algorithms on data obtained from high-, middle, and low-yield wells. The algorithms are compared regarding the regression evaluation indexes, and the results indicate that the improved FOA-GRNN model has better prediction accuracy and prediction effect than the other models. This study represents an interdisciplinary application and research of deep learning and CBM development. The main objective of this study is related to the CBM production prediction and the CBM productivity analysis and evaluation system design. The results presented in this study could help to improve the single-well gas production of CBM and also have high application value.