Hydraulic-Mechanical-Damage Multifield Coupling Control Mechanism and Gas Production Prediction of Fracturing in Tectonically Deformed Coal Seam Authors: Zhihua Yan, Honggao Xie, Shuxun Sang, Xiaozhi Zhou, Shiqi Liu, Sijie Han, Fansheng Huang, Wenkai Wang, Jie Chen, Bingren Jiang Published: 2026 ABSTRACT Efficient coalbed methane (CBM) development through hydraulic fracturing in tectonically deformed coal seams is governed by the multifield coupling of the hydraulic (H), mechanical (M), and damage (D), and it is urgent to build a data-driven gas production prediction model. Using multifield coupled numerical simulation, this study systematically analyzes the effects of elastic modulus (E), initial permeability (k 0), and fracturing displacement rate (H f) on the fracturing-induced damaged area (D A), maximum postfracturing permeability (k m‑max), and gas production. The applicability of three models of support vector machine (SVM), random forest (RF), and generalized regression neural network (GRNN) in gas production prediction is analyzed. The results show that the initial permeability and fracturing displacement rate have significant positive effects on gas production, increasing cumulative gas production by 341.44% and 48.47%, respectively. The elastic modulus exhibits a negative effect, reducing cumulative gas production by 22.21%. Postfracturing damaged area and permeability also show significant positive effects. The RF model demonstrates the best prediction performance on both the training and test sets, achieving R2 values of 0.9871 and 0.9551, respectively, with error metrics significantly lower than those of the SVM and GRNN models. The RF model is therefore more suitable for gas production prediction in CBM fracturing wells under multifeature parameter conditions. The HMD multifield coupling control mechanism+machine learning drive fusion gas production prediction method realized the technical path from physical mechanism revelation to engineering prediction of hydraulic fracturing coalbed methane development in structurally deformed coal seams. Introduction As a key component of clean energy, the efficient exploration, development, and utilization of coalbed methane (CBM) are of great significance to China’s energy structure adjustment and the achievement of its “dual carbon” goals. − However, over 60% of shallow CBM resources in China occur in tectonically deformed coal seams. As these coal seams are structurally fractured, have low strength and poor permeability, traditional development technologies struggle to achieve economically viable extraction. , As a key means to improve the permeability of coal-bearing gas reservoirs, hydraulic fracturing technology has promoted the rapid growth of unconventional oil and gas resources. − In hydraulic fracturing operations targeting tectonically deformed coal seams, through-layer fracturing via hard roof strata is commonly adopted. − However, when the water pressure is applied to the tectonically deformed coal seams, it faces the engineering dilemma of “unable to open, unable to support, and poor gas production effect”. At the heart of this challenge lies the fracture propagation behavior and permeability response mechanism governed by the multifield coupling of the hydraulic field (H), mechanical field (M), and damage field (D) during hydraulic fracturing, as well as the CBM productivity prediction models and methods under such multifield coupling. Therefore, it is essential to develop machine learning-based gas production prediction approaches that combine physical consistency with data adaptability. Currently, exploration of coal seam fracturing engineering and CBM enhancement methods has evolved from an early focus on a single mechanical field to a stage that considers thermal-hydraulic-mechanical-chemical multifield coupling. − However, existing studies have mostly concentrated on relatively structurally intact high-rank coals, while insufficient attention has been paid to the unique characteristics of tectonically deformed coal seams, such as strong heterogeneity, significant plastic deformation, and complex fracture network generation. , The hydraulic-mechanical-damage (HMD) coupling process during hydraulic fracturing of shallow tectonically deformed coal seam represents a complex system full of uncertainties: the evolution of the seepage (hydraulic) field governs the pressure distribution within fractures and alters the pore structure and permeability of coal through gas adsorption–desorption, mechanical, and damage effects; the response of the mechanical field directly controls the morphology and effectiveness of the fracture network; and the damage field from microdefect initiation to macro-crack penetration is a “dark line” that runs through, reversely coupling and reshaping the state of the other two fields. Through a series of nonlinear constitutive relations, such as stress-permeability coupling and damage-permeability coupling, these three physical fields greatly amplify the small disturbance, resulting in complex behaviors that are difficult to accurately predict by traditional deterministic models. It is this complexity that provides an excellent opportunity and strong demand for the intervention of data-driven methods such as machine learning. For example, an interpretable machine learning model of the causal relationship between explicit parameters and coalbed methane well production during hydraulic fracturing. Furthermore, multimodal data neural networks have been successfully applied to predicting reservoir fracture propagation paths, while convolutional neural network models have been used to predict rock parameters such as elastic properties, permeability, and stress and damage. Traditional mechanistic models (based on physical equations) and emerging data science are not substitutes for each other; rather, they are inevitably moving toward integration. Machine learning models, for instance, can extract potential laws and associative features not yet described by explicit physical equations from massive amounts of experimental data, field monitoring data, and numerical simulation data. In the realm of gas production prediction, an increasing number of scholars are considering the influence of reservoir parameters and postfracturing permeability evolution under multifield coupling on production performance. A series of discrete element, finite element, and combined discrete-finite element experimental or numerical simulation studies have been conducted. − Using software such as COMSOL Multiphysics, researchers have investigated the controlling effects of HMD evolution on production during hydraulic fracturing. , With the rapid advancement of artificial intelligence, models such as GA-BP neural networks and GRNN have been employed to predict CBM production under the influence of multiple geological parameters. , Various models have achieved considerable progress in predicting gas production from CBM wells. Furthermore, some scholars have integrated machine learning models such as SVM and MLP to accurately predict gas drainage rates. and have utilized RF and LSTM models to predict gas adsorption–desorption capacity of coal. , Postfracturing gas production efficiency depends not only on fracture geometry but is also governed by the complex fracture network formed under multifield coupling, gas adsorption-diffusion-seepage processes, and competitive adsorption and displacement between water and methane in micronanopores. However, both fracture propagation during hydraulic fracturing of tectonically deformed coal and postfracturing gas production prediction are difficult to accurately describe using traditional models. Therefore, integrating machine learning with HMD-coupled mechanism models holds the promise of unifying the interpretability of multifield coupling mechanisms with high-precision gas production prediction, potentially bringing a paradigm shift to gas production forecasting for hydraulic fracturing in tectonically deformed coal seams. Reviewing previous research, solid foundations have been established in understanding the multifield coupling mechanisms of HMD during hydraulic fracturing and in data-driven CBM production prediction using multiple factors. This paper deepens the understanding of the HMD coupling control mechanisms and key gas production controlling factors in shallow, tectonically deformed coal seams through multifield coupled numerical simulations and machine learning methods. The rest of the paper is as follows: Section constructs a HMD coupling model and a machine learning model for a shallow, tectonically deformed coal seam, and preliminarily defines the numerical simulation scheme as well as the steps of the multifield coupling + machine learning prediction approach. Section presents numerical simulations and engineering gas production validation, generating multifactor coupled data sets for CBM extraction under different simulation cases. The physically constrained simulation results represented by the HMD governing equations (e.g., different Fracturing displacement, damage areas, initial permeability, postfracturing permeability) are embedded into machine learning models. Various machine learning algorithms (e.g., SVM, RF, and GRNN) are explored for their significance in predicting fracturing outcomes and production performance, and the optimal combination of features is selected to develop a prediction model integrating “physical mechanism interpretation + machine learning.” Section discusses and analyzes the engineering application value of multifield coupling theory and the data-driven hydraulic fracturing gas production prediction method. Section presents brief conclusions. Methodology HMD Multifield Coupling Theory and Gas Production Model Construction HMD Multifield Coupling Theory of Broken Soft Coal Seam In the hydraulic fracturing of a shallow broken soft coal seam and the drainage and drainage engineering of CBM, under the interaction between in situ stress and fluid pressure, the coal matrix undergoes elastic-plastic deformation, leading to fracture propagation or closure. This gradually evolves into damaged and undamaged zones, altering the seepage pathways and the space available for fluid migration. Figure illustrates a porous medium fluid flow model based on the damage evolution process in the coal. Based on this, the present study incorporates hydraulic fracture propagation in coal seams, gas–water migration during drainage, and methane adsorption/desorption, and proposes the following model assumptions: (1) The coal is treated as a poroelastic continuum composed of matrix pores and fractures. (2) The coal reservoir fractures are saturated with either water or CH4. (3) Methane in the coal matrix pores is predominantly in an adsorbed state, and the adsorption–desorption process follows the Langmuir eq . CH4 is considered an ideal gas; after desorption, it diffuses from the coal matrix into the fractures, following Fick’s law of diffusion. (5) Under differential pressure, CH4 and water flow through the fracture pathways in the reservoir, following Darcy’s law. (6) The coal reservoir temperature is assumed to be constant. (7) The deformation and failure of the coal under stress adhere to damage mechanics theory, obeying both the maximum tensile stress criterion and the Mohr-Coulomb criterion. Fluid seepage in saturated porous media during fluid–solid coupling damage evolution. HMD Multifield Coupling Control Equation Based on the previous theoretical assumptions, the strain control equation of coal during hydraulic fracturing is Gui,jj+G1−2νluj,ji+fji−αpf,i=0 1 {G=E/2(1+ν)α=1−K/KsK=E/[3(1−2ν)]Ks=Es/[3(1−2ν)] 2 where G is the shear modulus, Pa; u j, ji is the component of displacement in the i-direction, m; v is Poisson’s ratio, dimensionless; f ji is the component of body force in the i-direction, Pa; α is the Biot coefficient; p f,i the component of the fracture pressure in the i-direction, Pa; E and E s are the elastic modulus of coal and the elastic modulus of the solid skeleton, Pa; K and K s are the bulk modulus of coal and the bulk modulus of the solid skeleton, respectively, Pa. Fluid seepage in fractures and the matrix follows Darcy’s law. Combined with mass conservation, the governing equation of fluid migration can be expressed as ρw(S∂p∂t+α∂εv∂t)−∇×(ρwkμw∇p)+=Qm 3 where ρ w is the density of the fracturing fluid, kg/m3; k is the permeability (with k f and k m denoting the permeability of fractures and matrix, respectively), m2; Q m is the mass source term; kg/(m3·s); μw is the viscosity of water, Pa·s; S is the storage coefficient (with S f and S m denoting the storage coefficients of fractures and matrix, respectively), which can be expressed as {Sm=φmCw+(1−α)(α−φm)KSf=Cw+1Knbf 4 where C w is the compression coefficient of water. The permeability of natural fractures is related to the fracture aperture. The expression is , kf=bf212 5 When the hydraulic force acts on the fracture, the fracture aperture changes with normal stress and shear stress. The governing equation is as follows: bn=br+(b0−br)exp(−ξσn′) 6 νs={(τs−|σp′|tan⁡φf)Kstan⁡φd,up01−(εc0ε3)2F2=0,dF1>0 11 The elastic modulus E of coal has the following relationship with the damage factor D: E=(1−D)E0 12 Under the action of hydraulic fracturing and in situ stress, the porosity of the coal matrix can be expressed as φm=11+S[(1+A0)φ0+α(A−A0)] 13 where {A=εv+pf/KsA0=εv0+pf0/Ks 14 where εv is the volumetric strain; εv0 is the initial volumetric strain of the coal; p f and p f0 denote the fracture pressure and the initial fracture pressure, respectively, Pa. The matrix permeability can be expressed as km=k0(φmφ0)3eαDD 15 where k 0 is the permeability under initial pressure, m2; D is calculated by eq . α D is the damage-permeability effect coefficient: αD=ln⁡b0312k0h 16 where b 0 is the initial aperture, m; and h is the width of the damaged zone, m. Coalbed Methane Drainage Model Under hydraulic fracturing conditions and in situ stress, the coal mass undergoes strain under the combined effects of stress, fluid pressure, and gas adsorption/desorption. The governing equation for the coal stress field can be expressed as Gui,jj+G1−2νuj,ji+fji=αmpm,i+αfpf,i+Kεa,i 17 where {αm=1−K/Ksαf=1−K/Knεa=αsgVsgVsg=VLpm/(pL+pm) 18 where αm and αf are the Biot effective stress coefficients for the matrix and fractures, respectively; p m and p f are the matrix pressure and fracture pressure, respectively, Pa; ε a is the matrix shrinkage strain induced by CH4 adsorption/desorption; αsg is the adsorption strain coefficient, kg/m3; V sg is the gas desorption volume, m3/kg; V L is the Langmuir volume, m3/kg; pL is the Langmuir pressure, MPa. CH4 is an autogenous and self-storing gas in coal seams, primarily occurring in an adsorbed state within the coal matrix. After desorption, it serves as a mass source of CH4 within the fractures. Therefore, the CH4 migration equation can be expressed as ∂∂t{VsgρsMgRTsps+φmMgRTpm}=−MgτRT(pm−pf) 19 where ρs is the density of the rock skeleton, kg/m3; M g is the molar mass of CH4, kg/mol; R is the universal gas constant; T s is the temperature under standard conditions; p s is the standard atmospheric pressure; φm is the matrix porosity, %; T is the reservoir temperature, K; and τ is the CH4 desorption time, s. The gas–water two-phase flow equation within coal fractures can be expressed as {∂∂t(SgφMgpfgRT)+∇[−Mg(pfg+bk)RTkrgkμg∇pfg−Df∇ρgSg]=(1−φf)MgτRT(pm−pf)+QCH4∂∂t(Swφfρw)+∇(−ρwkrwkμwpfw)=QH2O 20 where S g and S w are the gas-phase and water-phase saturations, with S w + S g = 1; φf is the fracture porosity, %; b k is the Klinkenberg factor, Pa; k rg and k rw are the relative permeability of the gas and water phases; D f is the diffusion coefficient, m2/s; k is the absolute permeability of fractures, m2; μg and μw are the dynamic viscosities of the gas and water phases, Pa·s; p fg and p fw are the gas-phase and water-phase pressures within fractures, Pa; and c is the temperature coefficient of water, kg/(m3·K), respectively. The gas and water relative permeability are expressed as krg={krg0(1−Sw−Swr1−Swr−Sgr)2[1−(Sw−Swr1−Swr)2]0,0≤Sg≤Sgr,Sgr E > k 0. All three parameters have positive effects on D A, with fracturing displacement rate having the strongest effect in promoting damage formation. The correlation of the three parameters with k m–max, from highest to lowest, is k 0> H f > E. k 0 exhibits a very strong positive correlation with k m–max, indicating that initial permeability is the most important factor determining the postfracturing maximum permeability. H f shows a weak positive correlation with k m–max, indicating a limited effect. E shows a negative correlation with k m–max; the higher the elastic modulus, the lower the k m–max. Correlation heat map of different factors with (a) D A and (b) k m‑max. Gas Production Analysis of Coal Reservoir under HMD Multifield Coupling To further investigate the effects of different fracturing displacement rates, permeabilities, and elastic moduli on gas production, fracturing simulations were conducted according to schemes 1–4, and 18 sets of 185-day cumulative gas production data were obtained, as shown in Figure . The stimulation effect of fracturing displacement rate shows a significant positive correlation (Figure a,b): as the fracturing displacement rate increases from 0 to 10 m3/min, the 185-day cumulative gas production increases from 1.09 × 104 to 10.23 × 104 m3, representing an increase of 834.66%. From a low displacement rate (5 m3/min) to a high displacement rate (10 m3/min), the cumulative gas production increase is 48.47%. An appropriate high fracturing displacement rate significantly enhances reservoir conductivity by intensifying fracture propagation and damage stimulation, representing a universally applicable and efficient stimulation method. The effect of elastic modulus shows a negative correlation (Figure c,d): as the elastic modulus increases from 2 to 6 GPa, the cumulative gas production decreases from 10.36 × 104 to 8.06 × 104 m3, a reduction of 22.21%. Under the same fracturing conditions, low-elastic-modulus coal is more prone to plastic yielding, exhibits greater volumetric strain in the near-wellbore region, and achieves better fracturing stimulation within the fractured damaged zone, resulting in a well-developed matrix-fracture network and stronger gas production capacity. High-elastic-modulus reservoirs have greater stiffness, higher resistance to fracture propagation, and higher fracture initiation pressure requirements, leading to limited volume fracturing stimulation and significantly reduced gas production. The amplification effect of initial permeability on gas production is the most prominent (Figure e,f): as the initial permeability increases from 0.001 mD to 0.2 mD, the cumulative gas production increases from 6.12 × 104 to 27.00 × 104 m3, an increase of 341.44%. High-permeability reservoirs inherently possess a higher conductivity baseline, and the synergistic effect between the fracture network and matrix pores further amplifies the gas production potential after fracturing, exhibiting a significant permeability-sensitive characteristic. Cumulative gas production and gas production growth rate under (a, b) different fracturing displacements, (c, d) different elastic moduli, and (e, f) different initial permeabilities. To further investigate the effects of permeability and elastic modulus on gas production under different fracturing displacement rates, 180 sets of 185-day cumulative gas production data were obtained from simulations based on scheme 5, as shown in Figure . Overall, under all working conditions, the cumulative gas production shows a significant linear increasing trend with increasing fracturing displacement rate. High flow rates enhance fracture propagation and matrix damage, forming a more complex flow conductivity network, and exhibit stimulation effects on reservoirs with different permeabilities and elastic moduli. The amplification effect of initial permeability is very significant: as k 0 increases from 0.001 to 0.2 mD, the overall gas production level increases substantially, and the stimulation effect of flow rate is significantly amplified. In low-permeability reservoirs, the baseline gas production is low, and the increase brought by higher flow rates is limited. In high-permeability reservoirs, which inherently possess strong flow conductivity, the synergistic effect between the fracture network and matrix pores after fracturing fully activates the gas production potential, making the stimulation effect of increasing the fracturing displacement rate more prominent. The effect of elastic modulus exhibits a clear permeability dependence. In low-permeability reservoirs, the cumulative gas production of low-elastic-modulus reservoirs is significantly higher, consistent with the mechanical characteristic that low-elastic-modulus coal is prone to plastic yielding and forms a uniform damaged zone. As permeability increases, the influence of elastic modulus on gas production gradually weakens; the gap in gas production between different elastic modulus conditions narrows significantly, and a trend even emerges where high-elastic-modulus reservoirs surpass low-elastic-modulus ones in gas production (Figure d–f). This indicates that under high-permeability conditions, the reservoir’s own flow conductivity is sufficient to support efficient gas production, and the influence of the fracture propagation mode is diminished. In summary, an appropriate fracturing displacement rate is a key factor for stimulation, while the coupling effect between initial permeability and elastic modulus determines the gas production potential of the reservoir. Influence of elastic modulus and permeability on gas production under different fracturing displacements (185 d): (a) 0.001; (b) 0.0057; (c) 0.01; (d) 0.05; (e) 0.1; and (f) 0.2 mD. In addition to the effects of reservoir elastic modulus, permeability, and fracturing displacement rate on postfracturing gas production, the influence of the postfracturing damaged area (D A) and postfracturing permeability (interpreted in this study as the maximum postfracturing permeability, k m–max) on gas production is also an important aspect. Using the pre- and postfracturing gas production results from 181 simulations obtained from schemes 1 and 5, the Pearson correlation coefficients and SHAP values were calculated and plotted (see Figure ). These reveal the linear correlations between multiple pre- and postfracturing reservoir parameters under HMD multifield coupling control and gas production, as well as their contribution magnitudes in machine learning models. The two analyses show a high degree of consistency regarding the core influencing factors, while also demonstrating the complementarity of linear and nonlinear analyses. Correlation analysis indicates that initial reservoir permeability, damaged area, and maximum postfracturing permeability exhibit significant positive correlations with gas production, with correlation coefficients of 0.8888, 0.6817, and 0.8743, respectively, making them key linear influencing factors that govern gas production performance. Elastic modulus shows a weak negative correlation with gas production, with a relatively weak linear influence. The ranking of linear correlations with gas production is k 0 > k m‑max > D A > H f > E. SHAP interpretability analysis further quantifies the nonlinear contributions of each parameter in the SVM, RF, and GRNN models. The order of contribution of each parameter, as shown by the SHAP results for the SVM, RF, and GRNN models, is D A> k m‑max > E > H f > k 0, k m‑max>k 0>D A>E > H f, and k m‑max> D A > k 0> H f > E; The differing SHAP values across the three models validate their key driving roles within nonlinear models. The discrepancy between the aforementioned Pearson correlations and the SHAP ranking stems from the fact that correlation coefficients reflect only the linear relationship between two variables, whereas SHAP reflects the marginal predictive contribution of each feature within a specific nonlinear model (such as RF), while also accounting for interactions, multicollinearity, and nonlinear effects between features. However, parameters such as k m‑max and D A remain significant factors influencing gas production. The discrepancy between the Pearson coefficients and SHAP values highlights the nonlinear characteristics of the fracturing process, indicating that parameters such as fracturing volume and elastic modulus primarily exert a linear influence on the fracturing damage area and postfracturing permeability; however, their impact on gas production cannot be fully characterized by linear correlation alone and requires a comprehensive assessment incorporating the nonlinear interpretations provided by machine learning models. The combined analysis of Pearson correlation coefficients and SHAP values provides a comprehensive quantitative basis for understanding the influence mechanisms of fracturing parameters and further validates that permeability, damaged area, and fracturing displacement rate are the core factors affecting gas production. Analysis of correlation between multiparameters and gas production in fracturing process and SHAP interpretability: (a) correlation coefficient and (b) SHAP value. Machine Learning Training and Error Analysis of Prediction Results Figure compares the training effect and error index of SVM, RF, and GRNN and reveals the internal relationship between model performance and feature importance by combining with SHAP interpretable analysis. The training set fitting results of the three machine learning models show that all three can capture the overall distribution trend of the gas production data reasonably well. Among them, the RF model achieves the highest goodness of fit (R 2 = 0.9871). Its data points closely align with the fitted line, and the prediction errors are significantly lower than those of the other models: its MAE, MAPE, and RMSE values are 0.5547, 0.0428, and 0.8476, respectively, representing the best error metrics among the three. This demonstrates the RF model’s strong nonlinear fitting capability and indicates that it possesses good interpretability regarding the causal relationships between parameters and gas production in production prediction, which is highly consistent with previous studies. In contrast, the SVM and GRNN models achieve R 2 values of 0.9485 and 0.9463, respectively, with significantly higher error levels, indicating a weaker ability to capture complex nonlinear relationships compared to the RF model. Machine learning training results and error analysis. The superior performance of the RF model is directly related to its ability to accurately capture key postfracturing features. It assigns higher contribution weights to strongly influential features such as initial permeability, postfracturing damaged area, and postfracturing permeability, enabling it to effectively learn the nonlinear coupling effects of pre- and postfracturing reservoir properties and operational parameters on gas production. In contrast, the SVM and GRNN models have a relatively limited ability to characterize these core features, resulting in larger prediction errors. Therefore, the RF model, leveraging its effective utilization of key features and strong nonlinear fitting capability, demonstrates the best performance in the fracturing gas production prediction. Its error levels are significantly lower than those of the SVM and GRNN models, providing a reliable model foundation for subsequent gas production prediction and parameter optimization. Figure shows the gas production prediction effect and error index of SVM, RF, and GRNN on the test set, which is in contrast with the performance of the training set, further verifying the generalization ability and performance difference of the models. In the test set predictions, the RF model continues to perform best, achieving the highest coefficient of determination among the three, with an R 2 = 0.9551. Its MAE, MAPE, and RMSE values are 0.7726, 0.1794, and 1.4327, respectively, which are significantly lower than those of the SVM and GRNN models. This indicates that the RF model can still accurately capture the variation trend of gas production on unseen data, demonstrating the strongest generalization capabilities. Combined with the training set performance presented earlier and the SHAP interpretability analysis of the parameters, the excellent performance of the RF model stems from its effective utilization of key features: it fully learns the nonlinear coupling effects of core parameters such as initial permeability, postfracturing damaged area, and postfracturing permeability, achieving stable predictions with low error and high goodness of fit on both the training and test sets. The SVM model shows intermediate generalization performance, while the GRNN model exhibits a significant increase in error on the test set, indicating a certain tendency toward overfitting and an insufficient capability to capture complex nonlinear relationships. In summary, the RF model demonstrates the best prediction performance and generalization ability on both the training and test sets, making it the optimal model for fracturing gas production prediction in this scheme. It can provide reliable model support for subsequent reservoir gas production predictions and parameter optimization. Machine learning gas production prediction results and error analysis. Discussion Multifield Coupling Mechanism and Data-Driven Gas Production Prediction System In this study, a framework of “multifield coupling mechanism-data-driven prediction” integrated fracturing gas production analysis method was constructed (Figure ). This framework reveals the gas production control mechanism of HMD multifield coupling and elaborates on the mechanism-data fusion prediction methodology. At the multifield coupling mechanism level, hydraulic fracturing increases pore pressure, which, on one hand, directly drives fracture propagation and alters coal structure to form a damage field; on the other hand, it reduces the effective stress through fluid–solid coupling, inducing coal deformation. The evolution of the damage field, in turn, counteracts the mechanical parameters and fluid migration pathways, further intensifying the coupling effects among the multiple fields, collectively determining reservoir stimulation effectiveness and gas production potential. This forms a regulatory interaction among the hydraulic, mechanical, and damage fields. At the methodological level, this framework achieves a deep integration of the multifield coupling mechanism model and machine learning. First, through HMD multifield coupling numerical simulation, gas production data sets under various operating conditions are constructed, providing high-quality samples for machine learning. Then, through data set partitioning, model training, and optimization, a high-precision gas production prediction model considering the influence of multiple parameters is established, ultimately realizing a complete pathway from mechanism revelation to engineering prediction. This approach retains the accurate physical characterization of the multifield coupling mechanism while leveraging the advantages of machine learning in nonlinear mapping and rapid prediction, providing theoretical and methodological support for efficient gas production prediction and parameter optimization in fracturing. HMD multifield coupling gas production control mechanism and data-driven prediction method system diagram. Analysis of Engineering Application Value Machine learning has already been applied to parameter optimization and gas production prediction in coal seam hydraulic fracturing. , The “multifield coupling mechanism + data-driven” gas production prediction methodology established in this study is applied to gas production prediction for hydraulic fracturing in a tectonically deformed coal seam. It considers reservoir geological parameters as well as the characteristic factors that influence gas production predictionspecifically, the postfracturing damaged area and postfracturing permeability. The methodological framework accounts for the correlations, interpretability, and causal relationships between physical mechanisms and machine learning parameters in relation to gas production.(1)Tectonically deformed coal seams are characterized by low coal strength and strong permeability heterogeneity. Conventional empirical designs often lead to poor fracturing stimulation effectiveness and high variability in gas production. This method, through HMD multifield coupling simulation, accurately characterizes the coupled evolution of the hydraulic-mechanical-damage fields and reveals the mechanisms of fracture propagation and gas production under different fracturing displacement rates, elastic moduli, and initial permeability conditions, thereby providing a physical basis for optimizing fracturing parameters.(2)The machine learning RF model trained on the gas production data set constructed from numerical simulations enables rapid and high-precision prediction of long-term gas production after fracturing, addressing the time-consuming and low-efficiency limitations of traditional methods and providing support for rapid evaluation and dynamic adjustment of fracturing treatment designs. The data set currently generated in this study is based on numerical simulations; at the engineering application level, field data from coalbed methane fracturing operations are required to calibrate the methodological framework. (1)Tectonically deformed coal seams are characterized by low coal strength and strong permeability heterogeneity. Conventional empirical designs often lead to poor fracturing stimulation effectiveness and high variability in gas production. This method, through HMD multifield coupling simulation, accurately characterizes the coupled evolution of the hydraulic-mechanical-damage fields and reveals the mechanisms of fracture propagation and gas production under different fracturing displacement rates, elastic moduli, and initial permeability conditions, thereby providing a physical basis for optimizing fracturing parameters. Tectonically deformed coal seams are characterized by low coal strength and strong permeability heterogeneity. Conventional empirical designs often lead to poor fracturing stimulation effectiveness and high variability in gas production. This method, through HMD multifield coupling simulation, accurately characterizes the coupled evolution of the hydraulic-mechanical-damage fields and reveals the mechanisms of fracture propagation and gas production under different fracturing displacement rates, elastic moduli, and initial permeability conditions, thereby providing a physical basis for optimizing fracturing parameters. (2)The machine learning RF model trained on the gas production data set constructed from numerical simulations enables rapid and high-precision prediction of long-term gas production after fracturing, addressing the time-consuming and low-efficiency limitations of traditional methods and providing support for rapid evaluation and dynamic adjustment of fracturing treatment designs. The data set currently generated in this study is based on numerical simulations; at the engineering application level, field data from coalbed methane fracturing operations are required to calibrate the methodological framework. The machine learning RF model trained on the gas production data set constructed from numerical simulations enables rapid and high-precision prediction of long-term gas production after fracturing, addressing the time-consuming and low-efficiency limitations of traditional methods and providing support for rapid evaluation and dynamic adjustment of fracturing treatment designs. The data set currently generated in this study is based on numerical simulations; at the engineering application level, field data from coalbed methane fracturing operations are required to calibrate the methodological framework. This methodological system integrates mechanism- and data-driven approaches. It can effectively guide the optimization of fracturing treatment parameters for tectonically deformed coal seam while improving the efficiency of gas production prediction, offering a scientifically sound and practically viable technical pathway for the efficient development of tectonically deformed coal seam. Future research should fully consider the effects of reservoir temperature, coal seam thickness, and other factors on fracturing gas production. Additionally, an extensive collection of fracturing parameters, geological parameters, and gas production data from gas-producing wells in tectonically deformed coal seams could further improve the multifield coupling mechanism + data fusion gas production prediction methodology. Conclusions This study established a theory of HMD multifield coupling controlling gas production during hydraulic fracturing of tectonically deformed coal reservoirs. Using COMSOL Multiphysics software, the gas production performance under multiparameter conditions in the tectonically deformed coal seam was systematically investigated. An innovative HMD multifield coupling + data-driven gas production prediction methodology was developed, leading to the following main conclusions:(1)The initial permeability and fracturing displacement rate have positive effects. Increasing the initial permeability from 0.001 mD to 0.2 mD results in a cumulative gas production increase of up to 341.44%. Increasing the fracturing displacement rate from 5 m3/min to 10 m3/min yields a 48.47% increase in the cumulative gas production. Elastic modulus has a negative effect on gas production; as the elastic modulus increases from 2 to 6 GPa, the cumulative gas production decreases by 22.21%. Both the postfracturing damaged area and postfracturing permeability also have significant positive effects on gas production.(2)HMD multifield coupling governs reservoir gas production performance. The fracturing displacement rate and initial permeability exhibit significantly positive driving effects on gas production. The influence of the elastic modulus exhibits a clear permeability dependence. In low-permeability reservoirs, a low elastic modulus is more favorable for fracture propagation and gas production, whereas this effect is weakened in high-permeability reservoirs. In engineering practice, differentiated fracturing parameters should be matched to reservoirs with different permeabilities to optimize the fracture propagation morphology and stimulated volume, thereby achieving efficient stimulation under various reservoir conditions while balancing the stimulation extent and operational costs.(3)The RF model significantly outperforms the SVM and GRNN models in terms of goodness of fit and error metrics on both the training and test sets, demonstrating a strong ability to capture the nonlinear relationships between pre- and postfracturing reservoir parameters and gas production, as well as good generalization capability. This provides a model foundation for gas production prediction under multifeature parameter conditions in CBM fracturing wells.(4)The “multifield coupling mechanism + data-driven” gas production prediction method developed in this paper combines the advantages of physical mechanism characterization and efficient prediction, providing reliable support for hydraulic fracturing parameter optimization and gas production prediction in a tectonically deformed coal seam. In light of the shortcomings of existing research, future studies could explore the influence of reservoir geological and engineering parameters on fracturing operations and further investigate the factors affecting gas production under a wider range of characteristic parameter conditions. (1)The initial permeability and fracturing displacement rate have positive effects. Increasing the initial permeability from 0.001 mD to 0.2 mD results in a cumulative gas production increase of up to 341.44%. Increasing the fracturing displacement rate from 5 m3/min to 10 m3/min yields a 48.47% increase in the cumulative gas production. Elastic modulus has a negative effect on gas production; as the elastic modulus increases from 2 to 6 GPa, the cumulative gas production decreases by 22.21%. Both the postfracturing damaged area and postfracturing permeability also have significant positive effects on gas production. The initial permeability and fracturing displacement rate have positive effects. Increasing the initial permeability from 0.001 mD to 0.2 mD results in a cumulative gas production increase of up to 341.44%. Increasing the fracturing displacement rate from 5 m3/min to 10 m3/min yields a 48.47% increase in the cumulative gas production. Elastic modulus has a negative effect on gas production; as the elastic modulus increases from 2 to 6 GPa, the cumulative gas production decreases by 22.21%. Both the postfracturing damaged area and postfracturing permeability also have significant positive effects on gas production. (2)HMD multifield coupling governs reservoir gas production performance. The fracturing displacement rate and initial permeability exhibit significantly positive driving effects on gas production. The influence of the elastic modulus exhibits a clear permeability dependence. In low-permeability reservoirs, a low elastic modulus is more favorable for fracture propagation and gas production, whereas this effect is weakened in high-permeability reservoirs. In engineering practice, differentiated fracturing parameters should be matched to reservoirs with different permeabilities to optimize the fracture propagation morphology and stimulated volume, thereby achieving efficient stimulation under various reservoir conditions while balancing the stimulation extent and operational costs. HMD multifield coupling governs reservoir gas production performance. The fracturing displacement rate and initial permeability exhibit significantly positive driving effects on gas production. The influence of the elastic modulus exhibits a clear permeability dependence. In low-permeability reservoirs, a low elastic modulus is more favorable for fracture propagation and gas production, whereas this effect is weakened in high-permeability reservoirs. In engineering practice, differentiated fracturing parameters should be matched to reservoirs with different permeabilities to optimize the fracture propagation morphology and stimulated volume, thereby achieving efficient stimulation under various reservoir conditions while balancing the stimulation extent and operational costs. (3)The RF model significantly outperforms the SVM and GRNN models in terms of goodness of fit and error metrics on both the training and test sets, demonstrating a strong ability to capture the nonlinear relationships between pre- and postfracturing reservoir parameters and gas production, as well as good generalization capability. This provides a model foundation for gas production prediction under multifeature parameter conditions in CBM fracturing wells. The RF model significantly outperforms the SVM and GRNN models in terms of goodness of fit and error metrics on both the training and test sets, demonstrating a strong ability to capture the nonlinear relationships between pre- and postfracturing reservoir parameters and gas production, as well as good generalization capability. This provides a model foundation for gas production prediction under multifeature parameter conditions in CBM fracturing wells. (4)The “multifield coupling mechanism + data-driven” gas production prediction method developed in this paper combines the advantages of physical mechanism characterization and efficient prediction, providing reliable support for hydraulic fracturing parameter optimization and gas production prediction in a tectonically deformed coal seam. In light of the shortcomings of existing research, future studies could explore the influence of reservoir geological and engineering parameters on fracturing operations and further investigate the factors affecting gas production under a wider range of characteristic parameter conditions. The “multifield coupling mechanism + data-driven” gas production prediction method developed in this paper combines the advantages of physical mechanism characterization and efficient prediction, providing reliable support for hydraulic fracturing parameter optimization and gas production prediction in a tectonically deformed coal seam. In light of the shortcomings of existing research, future studies could explore the influence of reservoir geological and engineering parameters on fracturing operations and further investigate the factors affecting gas production under a wider range of characteristic parameter conditions.