JIN Y, ZHENG C H, SONG H B, et al. Prediction of coalbed methane productivity based on neural network models[J]. Journal of Henan Polytechnic University(Natural Science) , 2025, 44(1): 46-56.

doi: 10.16186/j.cnki.1673-9787.2023030083

Received: 2023/03/31

Revised: 2023/05/20

Published: 2025/01/02

Prediction of coalbed methane productivity based on neural network models

JIN Yi1,2,3, ZHENG Chenhui1, SONG Huibo1, MA Jiaheng1, YANG Yunhang1, LIU Shunxi1, ZHANG Kun1, NI Xiaoming4

1. School of Resources and Environment， Henan Polytechnic University， Jiaozuo  454000， Henan， China; 2. Collaborative Innovation Center of Coal Work Safety and Clean High Efficiency Utilization， Jiaozuo  454000， Henan， China; 3. The Collaborative Innovation Center of Coalbed Methane（Shale Gas） of Central Plains Economic Region， Jiaozuo  454000， Henan， China; 4. School of Energy Science and Engineering， Henan Polytechnic University， Jiaozuo  454000， Henan， China

Abstract: Objectives The productivity of coalbed methane is mainly affected by geological and engineering factors. Clarifying the influence mechanism of these factors on the productivity of coalbed methane wells is the basis for achieving fine reservoir reconstruction and increasing production of coalbed methane wells. Methods Therefore， this paper takes Shizhuang South Block in Qinshui Basin as the research object， and comprehensively considers the geological background， reservoir physical properties and dynamic drainage data， uses neural network algorithm to carry out CBM productivity prediction. Firstly， 10 geological parameters were selected as the main controlling factors for CBM productivity prediction by grey correlation analysis. On this basis， the fuzzy mathematics method was used to realize the division of 34 coalbed methane wells in the study area. Finally， according to the classification results， combined with the actual drainage data， the BP and LSTM neural network algorithms were used to predict the daily gas production of CBM wells. Results The results show that： （1） Based on the grey correlation method model analysis， 10 parameters such as permeability， gas saturation and reservoir pressure gradient in the study area are the key factors affecting the gas production performance of coalbed methane； （2） Using fuzzy mathematics evaluation method to evaluate the enrichment of coalbed methane， the gas production effects of 34 wells in the study area is divided into three categories： favorable area， relatively  favorable area and unfavorable area. （3）A coal reservoir daily gas production prediction model was established based on the LSTM algorithm， with a prediction error value between 4.06% and 14.79%， and the average error value of 11.09%. The prediction accuracy is significantly higher than the BP model. Conclusions The model has good stability and high prediction accuracy. It can be used as an effective means for long-term prediction of coal reservoir producti-vity， and then provide scientific basis for deployment of coalbed methane development processes and the formulation of procurement plans. the formulation of coalbed methane development plan and the scientific deployment of drainage technology.

Key words: LSTM neural network; BP neural network; grey correlation analysis; productivity prediction