Research on gas extraction volume prediction and adaptive control model based on GA-SVR

To address the challenges of reduced gas extraction volume due to declining gas content in the later stages of extraction, which result in unpredictable extraction outcomes and low efficiency, this study leverages the nonlinear coupling characteristics of various factors during the extraction process. Support Vector Regression（SVR） and Random Forest（RF） algorithms are employed to develop a predictive model for gas extraction volume, with hyperparameters optimized using a Genetic Algorithm（GA）. The results indicate that the GA-SVR model achieved an average absolute error of 303.62, a root mean square error of 565.42, and an average absolute percentage error of 0.023, outperforming comparable models. Moreover, during stratified borehole extraction, the influence of extraction negative pressure on extraction concentration and volume diminishes after multiple extraction cycles, reducing its control effect on extraction concentration but still showing a correlation with daily cumulative extraction volume. In contrast, under cross-layer borehole extraction conditions, both extraction volume and concentration are more sensitive to changes in negative pressure. Based on the GA-SVR prediction model, an adaptive control model for negative pressure regulation is developed to maximize the gas extraction volume of the pipeline network. Field comparison tests demonstrate that the adaptive control model, with a 30-day regulation cycle, increased the average gas extraction volume by 5.08% compared to the control group in the later stages of extraction, with a 7.15% increase observed in the first 15 days. These findings provide valuable guidance for predicting gas extraction volume and regulating negative pressure in the later stages of gas extraction.