Geophysical Log Responses and Predictive Modeling of Coal Quality in the Shanxi Formation, Northern Jiangsu, China

Abstract

Traditional coal quality assessment methods rely exclusively on the laboratory testing of physical samples, which impedes detailed stratigraphic evaluation and limits the integration of intelligent precision mining technologies. To resolve this challenge, this study investigates geophysical logging as an innovative method for coal quality prediction. By integrating scanning electron microscopy (SEM), X-ray analysis, and optical microscopy with interdisciplinary methodologies spanning mathematics, mineralogy, and applied geophysics, this research analyzes the coal quality and mineral composition of the Shanxi Formation coal seams in northern Jiangsu, China. A predictive model linking geophysical logging responses to coal quality parameters was established to delineate relationships between subsurface geophysical data and material properties. The results demonstrate that the Shanxi Formation coals are gas coal (a medium-metamorphic bituminous subclass) characterized by low sulfur content, low ash yield, low fixed carbon, high volatile matter, and high calorific value. Mineralogical analysis identifies calcite, pyrite, and clay minerals as the dominant constituents. Pyrite occurs in diverse microscopic forms, including euhedral and semi-euhedral fine grains, fissure-filling aggregates, irregular blocky structures, framboidal clusters, and disseminated particles. Systematic relationships were observed between logging parameters and coal quality: moisture, ash content, and volatile matter exhibit an initial decrease, followed by an increase with rising apparent resistivity (LLD) and bulk density (DEN). Conversely, fixed carbon and calorific value display an inverse trend, peaking at intermediate LLD/DEN values before declining. Total sulfur increases with density up to a threshold before decreasing, while showing a concave upward relationship with resistivity. Negative correlations exist between moisture, fixed carbon, calorific value lateral resistivity (LLS), natural gamma (GR), short-spaced gamma-gamma (SSGG), and acoustic transit time (AC). In contrast, ash yield, volatile matter, and total sulfur correlate positively with these logging parameters. These trends are governed by coalification processes, lithotype composition, reservoir physical properties, and the types and mass fractions of minerals. Validation through independent two-sample t-tests confirms the feasibility of the neural network model for predicting coal quality parameters from geophysical logging data. The predictive model provides technical and theoretical support for advancing intelligent coal mining practices and optimizing efficiency in coal chemical industries, enabling real-time subsurface characterization to facilitate precision resource extraction.

1. Introduction

China’s resource endowment ensures coal will remain its primary energy source for the foreseeable future [1,2,3]. Coal quality indicators critically influence the economic viability of mining operations and coal chemical industries. Traditional methods for assessing these indicators depend on laboratory tests of physical samples, which are time-intensive, costly, and constrained by limited core availability [4,5,6]. This results in fragmented, discontinuous data that inadequately characterizes coal quality vertically and lacks horizontal quantitative resolution. These limitations hinder coal resource evaluation, development, and utilization, particularly in enabling precise stratification assessment or efficient, targeted mining. Geophysical logging methods offer a promising alternative to address these challenges.

Prior studies demonstrate that bulk density logging effectively identifies coal quality parameters, while natural gamma, compensated neutron, and resistivity logging sensitively detect clay mineral content, moisture, and ash-related mineral types [7,8,9,10,11,12,13,14,15,16,17]. Conventional approaches to interpreting coal quality from logging data involve constructing predictive models tailored to specific coal seams using multiple logging parameters [8]. For example, coal rank is often determined via density-neutron cross plots [9,10], while oil-rich coal is identified by integrating true density with natural gamma [11]. Industrial composition has been predicted using volumetric and probabilistic models based on geophysical logging parameters [12,13,14,15,16,17]. Notably, ash yield correlates positively with density and natural gamma [9,18,19,20], whereas volatile matter and fixed carbon exhibit negative correlations with density and acoustic transit time [12,21,22]. Ghosh et al. applied intersection plots, regression analysis, and machine learning to predict industrial composition using natural gamma, resistivity, and density [13,23]. O’Brien et al. employed neural networks to evaluate moisture, ash, and volatile matter via density, acoustic, and natural gamma data [24]. Shao and Meng utilized multivariate linear fitting and volumetric models, respectively, to predict industrial composition [14,16], while Wu et al. classified sulfur content (>3% vs. <3%) using natural gamma, spontaneous potential, resistivity, density, and acoustic data, establishing multivariate linear models for sulfur prediction in North China’s Shanxi and Taiyuan Formation coals [25].

Despite these advances, existing models predominantly target isolated parameters such as industrial composition, coal rank, or sulfur content [10,21,24,25,26]. A systematic approach to invert geophysical logging responses for comprehensive coal quality indicators such as moisture, ash, volatile matter, fixed carbon, total sulfur, and calorific value remains underdeveloped, This critical knowledge gap severely impedes the establishment of robust petrophysical inversion models capable of meeting the operational demands for efficient, high-resolution identification of deep coal seam quality and data-driven optimization of intelligent coal mining systems. The absence of such methodologies constrains our ability to leverage logging data for predictive quality assessment across heterogeneous coal reservoirs, ultimately limiting the implementation of closed-loop, precision mining strategies essential for resource efficiency and automation in complex subsurface environments. To address this gap, this study analyzes the No. 7 and No. 8 coal seams of the Shanxi Formation in the Xuzhuang and Kongzhuang mines of northern Jiangsu, China. Integrating scanning electron microscopy, X-ray analysis, and optical microscopy with interdisciplinary methodologies spanning mathematics, mineralogy, and applied geophysics, we investigate the coal quality and mineral composition of these seams. The intrinsic relationships between geophysical logging responses and coal quality parameters are rigorously analyzed to establish a predictive model, which is subsequently validated through feasibility assessments and error analysis.

This work provides a technical and theoretical foundation for advancing intelligent mining practices and optimizing efficiency in coal chemical industries. By bridging subsurface geophysical data with material properties, the model supports real-time coal quality characterization, enabling precision resource management and utilization.

2. Geological Background

The study area features flat terrain with extensive surface deposits of sandy clay derived from historical Yellow River flooding. The region slopes from higher elevations in the west to lower elevations in the east, reflecting the influence of regional tectonic activity. Structural features are dominated by faults, predominantly northeast-trending tensional-torsional normal faults, with folds being underdeveloped. Early northeast-oriented stress shaped the structural framework, resulting in a monocline structure dipping northwest. Strata dip angles typically range between 14° and 21°. The Datun and Xuzhuang faults divide the area into the Xuzhuang and Kongzhuang coal mines (Figure 1).

The Late Paleozoic Shanxi coal-bearing strata in North China is developed in a transitional environment between oceanic and continental conditions, with high hydrocarbon generation potential [27]. During the Late Carboniferous-Early Permian period, the North China Platform experienced continuous degradation from marine to continental environments [28]. Coal-bearing strata are mainly composed of shale, coal, sandstone, and interbedded sandstone rich in organic matter. These strata were formed in the coastal plains and swamps of northern China [29]. The Shanxi Formation was deposited from bottom to top in three environments: tidal flats, delta front, and delta plains, indicating an overall regression process from nearshore marine to river systems [27,30,31].

The Shanxi Formation comprises coal-bearing strata deposited in nearshore fluvial-lacustrine and peat swamp environments, with thicknesses ranging from 70.00 to 149.32 m (average: 116.40 m). Lithologies include gray to dark-gray sandy mudstone, mudstone, and sandstone, containing three to four coal-bearing layers [32]. In the Kongzhuang Mine, the No. 7 coal seam varies in thickness from 1.40 to 6.52 m (average: 4.41 m), while the No. 8 seam ranges from 0.29 to 4.96 m (average: 2.90 m). In the Xuzhuang Mine, the No. 7 seam measures 1.19–5.45 m (average: 3.89 m), though it is absent in several localities. The Shanxi Formation hosts few but thick, minable coal seams characterized by stable sedimentation, abundant plant fossils, and conformable contact with overlying strata.

3. Materials and Methods

3.1. Sample Collection

Coal samples were collected from the Xuzhuang and Kongzhuang coal mines in northern Jiangsu, comprising underground channel samples and drill core samples (Figure 1). Channel sampling adhered to the Chinese national standard (GB/T 482-2008) with coal seams profiled vertically from top to bottom at intervals of 10 cm and a sampling depth of 5 cm [33]. Samples were wrapped in plastic film, sealed in labeled bags, and transported to the laboratory for preparation and subsequent analysis.

3.2. Logging Data Processing

Geophysical logging responses for coal quality were collected from 1:20-scale coal seam logging curves at 0.2 m intervals. Digital logging data included apparent resistivity (ρ_s), lateral resistivity (LLD), bulk density (DEN), natural gamma (GR), short-spaced gamma-gamma (CGS), and acoustic transit time (AC). Data processing involved three stages: depth correction, logging data extraction, and standardization. Depth correction was applied using logging curve responses from thick sandstone, limestone, and coal seams to align depth references. Logging data extraction calculated weighted averages for ρ_s, DEN, AC, GR, LLD, and CGS curves.

Standardization normalized ρ_s, DEN, AC, and GR values using Formula (1) [14], and CGS and LLD values via Formula (2) [25].

$$N_N={\displaystyle \frac{N - N_{\mathit{min}}}{N_{\mathit{max}} - N_{\mathit{min}}}}$$

(1)

$$M_N={\displaystyle \frac{\lg M - {\lg M}_{\mathit{min}}}{{\lg M}_{\mathit{max}} - {\lg M}_{\mathit{min}}}}$$

(2)

In these formulas, N and M denote raw logging data; N_n and M_n represent normalized data; N_max/M_max and N_min/M_min correspond to the maximum and minimum values of each parameter, respectively.

3.3. Analytical Methods

Approximate analysis, involving the quantification of total sulfur, ash yield, and volatile matter, moisture of the samples, was measured with reference to the ASTM Standard D2492-02 (2012), D3174-12 (2018), D3175-20 (2020), and D3173-17a (2017) [34,35,36,37]. The maximum vitrinite reflectance (R_o,max),= and calorific value (Q_gr,ad) were tested following the ASTM standards D2798-21 (2021) and Chinese standards GB/T 213-2008 [38,39]. Meanwhile, optical microscopes with ultraviolet light and white light reflections, X-ray analyzers, and scanning electron microscopes were used to observe the mineral morphology in coal samples.

3.4. Neural Network Model

The backpropagation (BP) neural network model integrates principles from mathematics, neuroscience, and computer science to emulate aspects of human brain functionality [38]. It establishes nonlinear mappings between input and output layers, enabling machine learning and the approximation of complex continuous functions [39]. For the coal quality geophysical logging prediction model, the BP network comprises three core layers: input, hidden, and output. The hidden layer, positioned between input and output layers, processes data without relying on predefined functional relationships to describe input-output mappings (Figure 2). Input variables (X: x₁, x₂, …, x_n) include lateral resistivity (LLD), natural gamma (GR), density (DEN), apparent resistivity (ρ_s), short-spaced gamma-gamma (CGS), and acoustic transit time (AC). Output variables (Y: y₁, y₂, …, y_j) predict coal quality parameters: moisture (M_ad), ash yield (A_d), volatile matter (V_daf), fixed carbon (FC), total sulfur (S_t,d), and calorific value (Q_gr,ad).

The hidden layer (S) mediates information transfer to the output layer (Y). The neural network’s output is mathematically expressed as [40]

$$Y = g\left[f\left(X{h}_1\right)h_2\right]$$

(3)

The functions g and f represent activation functions within the BP neural network model. To enhance the algorithm’s capacity for nonlinear learning, tanh and sigmoid are typically selected for g and f, respectively [40].

A two-sample independent t-test evaluates whether the difference between the mean measured and predicted values of coal quality parameters is statistically significant relative to their populations. The t-test statistic is calculated as [40]

$$t={\displaystyle \frac{Y_1-Y_2}{\sqrt{\frac{\left(n_1-1\right)S_1^2+(n_2-1)S_2^2}{n_1+n_2-2}\left(\frac{1}{n_1}+\frac{1}{n_2}\right)}}}$$

(4)

In the formula, Y₁ and Y₂, S₁² and S₂², n₁ and n₂ are the average values, variances, and sample sizes of measured and predicted coal quality parameters, respectively.

4. Results and Discussion

4.1. Characteristics of Coal Quality

The No. 7 coal seam in the study area exhibits a vitrinite maximum reflectance (R_o,max) averaging 0.69%, classifying it as medium-metamorphic bituminous coal. Key coal quality parameters include an average moisture content (M_ad) of 1.28%, ash yield (A_d) of 12.49% (low ash), volatile matter (V_daf) of 38.48% (high volatile), fixed carbon (FC) of 53.59% (low fixed carbon), total sulfur (S_t,d) of 0.70% (low sulfur), and an air-dried high calorific value (Q_gr,ad) of 30.13 MJ/kg (high calorific value) (

Table 1. Statistical table of coal quality indicators of Shanxi Formation in Northern Jiangsu region.

Coal Seam	Ro,max	Mad	Ad/%	Vdaf	FC	St,d	Qgr,ad
No. 7	0.67~0.85	0.52~1.90	7.15~20.58	35.51~43.38	47.61~59.06	0.17~2.36	26.79~32.40
	0.69	1.28	12.49	38.48	53.59	0.70	30.13
No. 8	0.66~0.87	0.18~1.76	7.19~16.57	35.16~40.60	51.46~58.43	0.10~2.03	28.48~31.80
	0.72	1.13	11.06	37.93	54.98	0.93	30.44

R_o,max, Maximum reflectivity of vitrinite (%); M_ad, Moisture content (%); A_d, Ash yield (%); V_daf, Volatile matter yield (%); FC, Fixed carbon content (%); S_t,d, Total sulfur content (%); Q_gr,ad, Calorific value (MJ/kg); ad, Air drying base; d, desiccant; daf, Dry ash free base; The data above the horizontal line in the table represents the minimum to maximum values, and below the horizontal line represents the average value.

).

Similarly, the No. 8 coal seam records a vitrinite R_o,max of 0.72%, also indicative of medium-metamorphic bituminous coal. Its average M_ad, A_d, V_daf, FC, S_t,d, and Q_gr,ad values are 1.13%, 11.06%, 37.93%, 54.98%, 0.93%, and 30.44 MJ/kg, respectively, categorizing it as low ash, low fixed carbon, high volatile, low sulfur, and high calorific value coal (Table 1).

In summary, both the No. 7 and No. 8 coal seams of the Shanxi Formation are characterized as medium-metamorphic bituminous coals with low sulfur, low ash, low fixed carbon, high volatile matter, and high calorific value.

Correlation analysis of moisture, ash content, volatile matter, fixed carbon, total sulfur, and calorific value in the Shanxi Formation coals from the Kongzhuang and Xuzhuang mines of northern Jiangsu reveals distinct relationships. Ash yield demonstrates a significant negative correlation with fixed carbon content and calorific value. Fixed carbon content, in turn, shows strong correlations with volatile matter yield and calorific value. Total sulfur content exhibits a significant positive correlation with volatile matter yield but only weak associations with ash yield and volatile matter (Figure 3). Meanwhile, the fixed carbon, calorific value, and volatile matter in coal quality can further predict coal grades [41].

4.2. Mineralogical Characteristics

The coal mineral composition in the study area is dominated by calcite, pyrite, and clay minerals. Optical microscopy reveals diverse pyrite morphologies, including euhedral and subhedral fine-grained crystals, framboidal aggregates, fracture infillings, irregular blocky forms, and disseminated particles. Euhedral and subhedral pyrite grains (1–20 μm) occur alongside calcite within fractures (Figure 4a). Fracture-filling pyrite, forming strips 0.1–2.0 μm wide, lines coarse-grained mineral surfaces (Figure 4b), while oxidized filamentous and semi-filamentous pyrite infills fractures (Figure 4c,d), a morphology linked to late-stage coalification processes [42]. Irregular blocky pyrite (2–30 μm), characterized by smooth surfaces with minor pitting, commonly infills pores and fractures in association with calcite (Figure 4e). Disseminated pyrite (1–5 μm) is predominantly hosted within calcite (Figure 4f), with localized nodular clusters observed in matrix vitrinite (Figure 4g), reflecting the selective affinity between gel components and pyrite [43,44].

Calcite occurs as veins or blocky masses (5–60 μm wide), infilling homogeneous and matrix vitrinite (Figure 4a,e–h). Clay minerals appear as granular or striated inclusions within vitrinite, with some exhibiting a roe-like texture (1–6 μm) embedded in vitrinite matrices (Figure 4i). Meanwhile, coal quality parameters are closely related to mineralogy; for example, the thermal decomposition of calcite will lead to an increase in the yield of volatile matter in coal quality [45].

Scanning electron microscopy with energy-dispersive spectroscopy reveals stacking and infilling of 3 μm-diameter pores within massive iron dolomite (Figure 5a). Fractures in the coal seam are occupied by band-like and irregular blocky calcite (Figure 5b,f), while cylindrical calcite fills pores measuring 3–15 μm in diameter (Figure 5h). Long columnar gypsum is observed within organic matter (Figure 5c). Granular pyrite (2–20 μm) is embedded within vitrinite, exhibiting octahedral, tetrahedral, and transitional crystal habits (Figure 5d,g). Fine-grained aggregated pyrite (1–10 μm) is hosted in kaolinite, with irregular blocky pyrite infilling coal fractures (Figure 5e). Agglomerated and disseminated kaolinite occur within vitrinite or pore spaces (Figure 5d,f,g,i), while columnar dolomite occupies 10–20 μm pores (Figure 5i).

Banded pyrite, formed during alkaline peat swamp evolution, reflects gel-enhanced pyrite infilling processes [46]. Sulfide mineralization is linked to seawater-derived sulfur sources, promoting framboidal pyrite development [47,48]. Irregular blocky pyrite forms under iron-deficient aqueous conditions [49]. Post-depositional seawater alteration triggers calcite dolomitization via Ca²⁺-Mg²⁺ ion exchange [50].

4.3. Geophysical Logging Response Characteristics of Coal Quality

4.3.1. Logging Response Characteristics of Moisture Content

Moisture in coal exists as both free water, which readily evaporates, and endogenous moisture adsorbed onto clay particle surfaces. The moisture content exhibits an initial decrease followed by an increase with rising apparent resistivity and density (Figure 6a,c), a trend governed by moisture occurrence states and coal reservoir physical properties [25]. In the study area, deep burial depths promote water molecule adsorption onto inorganic mineral surfaces, enhancing ion conductivity and reducing apparent resistivity [51,52]. However, in middle-to-low rank coals, increasing moisture and humic acid content, combined with active tectonic activity fostering pore-fracture systems, elevate apparent resistivity [14]. Density logging values correlate with coal rank, mineral content, lithotype composition, and reservoir properties [53]. For low-rank bituminous coals, advancing metamorphism reduces moisture, humic acid, and inorganic minerals, lowering seam density. Conversely, higher vitrinite content enhances water molecule adsorption capacity, increasing density [54,55].

Moisture content negatively correlates with lateral resistivity, natural gamma (GR), short-spaced gamma-gamma (CGS), and acoustic transit time (AC) (Figure 6b,d–f). Water molecules enhance coal conductivity, reducing lateral resistivity [14]. Pore and fracture development in coal slows longitudinal acoustic wave propagation, increasing AC. As moisture rises, wave velocity accelerates, decreasing AC [56,57,58]. Radioactive elements concentrate in inorganic minerals; however, clustered or striated mineral infillings in pores/fractures reduce water retention, causing moisture to decline with increasing GR [59,60].

4.3.2. Logging Response Characteristics of Ash Yield

The sedimentary environment exerts a controlling influence on coal ash yield, which reflects the degree of mineral development [61,62]. Ash yield exhibits an initial decrease followed by an increase with rising apparent resistivity (Figure 7a,c). Elevated ash yield and inorganic mineral content enhance coal conductivity, reducing seam resistivity [63,64]. In the study area, blocky and banded calcite, pyrite, and clay minerals infill pores, fractures, or organic matter, further lowering apparent resistivity [25,65,66]. Minerals such as pyrite within ash enhance coal rock conductivity [67], increasing brittleness and promoting cleavage, pores, and fractures that amplify conductivity variations [47,51,68].

Ash yield similarly trends with density, decreasing initially before rising (Figure 7c). This relationship is governed by material composition, coal rank, clay mineral content, and drilling-induced factors such as borehole diameter changes and mud invasion [69]. Syngenetic minerals predominantly influence coal seam pore structure through density, composition, and deformation, with density being the most critical factor [70]. Sulfide minerals, denser and more conductive than silicates, increase seam density and reduce resistivity when enriched [71].

Ash yield correlates positively with lateral resistivity, natural gamma (GR), short-spaced gamma-gamma (CGS), and acoustic transit time (AC) (Figure 7b,d–f). This arises because organic matter in coal has higher resistivity than hydrated minerals, and GR intensity increases with ash yield [14]. Clay minerals, with their high surface area, strongly adsorb radioactive elements and magmatic minerals during diagenesis, leading to higher GR values in ash-rich coals [14]. Late-stage sedimentation fills coal reservoir pores and fractures with banded minerals, yet pervasive pore-fracture systems elevate ash yield and AC [47,72,73].

4.3.3. Logging Response Characteristics of Volatile Matter

Volatile matter in coal represents the fraction released as gases during high-temperature heating. Its yield displays a parabolic relationship with apparent resistivity and density (Figure 8a,c), a pattern influenced by the content, properties, and structure of organic and inorganic components, as well as coalification processes [14,74]. Increased sulfide mineral content reduces coal seam apparent resistivity [75], while advancing coalification decreases inorganic mineral content, elevating resistivity [64,76]. Density logging values correlate with coal lithotype, rank, ash composition, and external factors such as borehole diameter variations and fracture development [72].

Volatile matter yield positively correlates with lateral resistivity, natural gamma (GR), short-spaced gamma-gamma (CGS), and acoustic transit time (AC) (Figure 8b,d–f). This relationship stems from reduced mineral content and higher organic matter resistivity in bituminous coals with increasing metamorphism [14,77]. GR intensity rises proportionally with radioactive particle abundance in coal [14], predominantly hosted in pyrite and sulfur-rich organic matter [78,79,80]. Consequently, volatile matter correlates positively with lateral resistivity and GR. Concurrently, pore-fracture systems in medium- to low-rank bituminous coals elevate CGS and AC values [73].

4.3.4. Logging Response Characteristics of Fixed Carbon

Fixed carbon content in coal initially increases before decreasing with rising apparent resistivity (Figure 9a), a trend governed by coal metamorphism. Advancing coalification in mid- to low-rank coals reduces inorganic minerals, moisture, and humic acid, sharply lowering ion conductivity and increasing fixed carbon content alongside resistivity [51,77]. However, further coalification enhances electronic conductivity in higher-rank coals, reducing resistivity [75].

Fixed carbon and density similarly exhibit an initial rise followed by decline (Figure 9c), influenced by coal rank, mineral content, lithotype composition, and reservoir properties [47,53]. Increasing metamorphism in mid- to low-rank coals reduces moisture, elevating seam density, while continued coalification decreases inorganic minerals, lowering density [55,81].

Fixed carbon content negatively correlates with lateral resistivity, natural gamma (GR), short-spaced gamma-gamma (CGS), and acoustic transit time (AC) (Figure 9b,d–f). Progressive coalification diminishes inorganic minerals and adsorbed radioactive elements, reducing GR [25]. Medium- to low-rank bituminous coals, with loose structures and developed pore-fracture systems, exhibit elevated lateral resistivity and AC [14,68,82]. As metamorphism advances, denser coal structures and enhanced electronic conductivity reduce lateral resistivity and AC [81,83].

4.3.5. Logging Response Characteristics of Total Sulfur

Total sulfur content in coal exhibits an initial decrease followed by an increase with rising apparent resistivity (Figure 10a), a trend linked to coal’s micro-pore structure and the type and abundance of conductive mineral elements [83]. Pyrite’s polarizability and thermoelectric properties induce pronounced resistivity anomalies in coal seams [68,71,82,84]. In the study area, deep burial depths and active tectonics amplify crystal defects like pores and fractures, extending ion and electron migration distances while reducing micro-pore proportions and ion-conductive pathways, thereby elevating resistivity [4,70,82]. Total sulfur and density similarly show an initial increase before declining (Figure 10c), reflecting the dominance of lower-density organic sulfur in medium- to high-sulfur coals compared to pyritic sulfur in low-sulfur varieties [17].

Total sulfur content correlates positively with lateral resistivity, short-spaced gamma-gamma (CGS), and acoustic transit time (AC) (Figure 10b,e,f). CGS values directly reflect inorganic mineral content [14,85], with sulfide-rich seams exhibiting lower resistivity than organic sulfur-dominated coals [4]. Increased inorganic minerals reduce micro-pore abundance [70], prolonging free electron presence in fractures and elevating lateral resistivity and AC [81,83].

A positive correlation between total sulfur and natural gamma (GR) (Figure 10d) arises from Th and U enrichment during peat accumulation or early diagenesis [86,87]. Carboniferous-Permian coal seams, formed in alkaline oxidation-reduction transition environments, further promote U, K, and Th migration alongside pyrite and kaolinite formation [78,79,80,88,89].

4.3.6. Logging Response Characteristics of Calorific Value

The calorific value exhibits a non-monotonic relationship with apparent resistivity, initially increasing before reaching a peak and subsequently decreasing. A similar unimodal response pattern is observed with bulk density, where calorific value rises to an optimum before declining. Conversely, calorific value demonstrates significant negative correlations with laterolog resistivity, short-spaced density log (gamma-gamma) readings, natural gamma ray intensity, and sonic transit time.

Coal calorific value declines with increasing ash and moisture content, where ash exerts a growing influence as metamorphism advances, while moisture’s impact diminishes [90]. Calorific value initially rises before decreasing with apparent resistivity (Figure 11a). In low- to medium-rank bituminous coals, reduced inorganic minerals and moisture elevate resistivity [63,91], compounded by deep burial depths and well-developed pore-fracture systems in the study area, further increasing resistivity [47,51,68].

Calorific value similarly peaks and then declines with density (Figure 11c), influenced by coal rank, mineral content, lithotype, and reservoir properties [53,74]. Progressive coalification reduces moisture, enhancing calorific value alongside density in bituminous coals [14,64]. However, as bituminous coal transitions to higher ranks, density rises while calorific value decreases [91,92].

Calorific value negatively correlates with lateral resistivity, short-spaced gamma-gamma (CGS), natural gamma (GR), and acoustic transit time (AC) (Figure 11b,d–f). Heat distribution across macerals follows liptinite > vitrinite > inertinite, driven by liptinite’s higher oxygen content [55]. Resistivity decreases as coal lithotype transitions from dull to bright coal. Concurrently, declining inorganic minerals and adsorbed radioactive elements reduce calorific value as GR increases [14,25]. Coal conductivity arises from ionic conduction in mineralized water and electronic conduction via free radicals [93]. Advancing metamorphism in mid- to low-rank coals reduces moisture, densifies coal structures, and lowers lateral resistivity and AC [77,94].

4.4. Prediction Model for Geophysical Logging Response of Coal Quality

Based on the correlation coefficient between coal quality parameters and geophysical logging response values, the geophysical logging response parameters of each coal quality parameter are optimized. Among them, the correlation of M_ad logging parameters from high to low is ρ_s, GR, LLD, and DEN. A_d has a good correlation with LLD, DEN, GR, CGS, and AC. V_daf is LLD from high to low AC, DEN, GR; FC from high to low is GR, LLD, ρ_s, AC, S_t,d from high to low is ρ_s, GR, AC, and DEN; Q_gr,ad have good correlations with LLD, DEN, GR, and CGS (Figure 12). We first constructed a neural network model for coal quality indicators, and then evaluated and verified the effectiveness of the prediction model.

4.4.1. Evaluation of Prediction Model for Coal Quality Logging Response

Two-sample independent t-tests comparing predicted and measured values for moisture content yielded a p-value of 0.430 and a two-tailed significance value (Sig) of 0.901, both exceeding the 0.050 threshold. Similar tests for ash yield, volatile matter yield, fixed carbon content, and total sulfur content produced p-values of 0.870, 0.098, 0.354, and 0.888, respectively, with corresponding two-tailed Sig values of 0.685, 0.336, 0.878, and 0.905, all above 0.050. For calorific value, the p-value (0.014) fell below 0.050, but the two-tailed Sig value (0.501) remained nonsignificant, indicating no statistically meaningful variance between predicted and measured values. These results confirm statistically insignificant differences across all coal quality parameters, validating the feasibility of the neural network model for predicting coal quality indicators using geophysical logging responses (

Table 2. Two-sample independent t-test table for predicted and measured coal quality parameters.

Parameters	AV	AVP	p	t	Sig (Two Tailed)
Mad	MV	1.685	0.430	0.126	0.901
	PV	1.677		0.126	0.901
Ad	MV	11.520	0.870	−0.413	0.685
	PV	12.061		−0.413	0.685
Vdaf	MV	38.578	0.098	0.978	0.336
	PV	38.048		0.978	0.340
FC	MV	54.105	0.354	0.156	0.878
	PV	53.893		0.156	0.878
St,d	MV	0.787	0.888	0.121	0.905
	PV	0.762		0.121	0.905
Qgr,ad	MV	30.297	0.014	0.688	0.498
	PV	30.089		0.688	0.501

AV, average value; MV, measured value; PV, predicted value; AVP, average value of parameters.

).

4.4.2. Validation of Coal Quality Logging Response Prediction Model

The neural network model’s reliability in predicting coal quality parameters was further validated through application to the No. 7 and No. 8 coal seams of the Shanxi Formation in the Xuzhuang and Kongzhuang mines, northern Jiangsu.

For moisture content (M_ad), processed ρ_s, GR, and DEN logging values served as input layers. The absolute error between predicted and measured values ranged from −0.48% to 0.47% (average: 0.03%), with relative errors between −21.05% and 35.88% (average: 6.41%) (

Table 3. Error analysis of predicted moisture content in coal.

DHN	NCS	Nρs	NDEN	NGR	Mad-PV (%)	Mad-MV (%)	AE (%)	RE (%)
KZ-2	8	0.69	0.26	0.14	1.82	1.78	0.04	2.25
KZ-5	8	0.52	0.58	0.44	1.79	1.36	0.43	31.62
KZ-8	7	0.51	0.52	0.10	1.8	0.28	−0.48	−21.05
XZ-1	8	0.30	0.85	0.42	1.78	1.31	0.47	35.88
XZ-3	8	0.03	0.82	0.72	1.79	0.22	−0.43	−19.37
XZ-7	7	0.67	0.98	1.00	1.79	1.64	0.15	9.15

DHN, Drill hole number; NCS, number of coal seams; N_ρs, normalization value of apparent resistivity; N_DEN, normalization value of density; N_GR, normalization value of natural gamma; M_ad-MV, measured value of moisture content; M_ad-PV, predicted value of moisture content; AE, absolute error; RE, relative error.

). Ash yield (A_d), modeled using DEN, AC, and GR inputs, showed absolute errors of −5.12% to 1.23% (average: −1.64%) and relative errors of −37.65% to 8.77% (average: −10.88%) (

Table 4. Error analysis of ash yield in coal.

DHN	NCS	NDEN	NGR	NAC	Ad-PV (%)	Ad-MV (%)	AE (%)	RE (%)
KZ-2	8	0.26	0.14	0.08	10.41	13.08	−2.67	−20.41
KZ-5	7	0.58	0.44	0.04	14.43	13.32	1.11	8.33
KZ-8	7	0.52	0.10	0.05	8.48	13.60	−5.12	−37.65
XZ-1	8	0.85	0.42	0.06	15.25	14.02	1.23	8.77
XZ-3	8	0.82	0.72	0.09	14.56	17.00	−2.44	−14.35
XZ-7	7	0.98	1.00	0.07	17.54	19.48	−1.94	−9.96

DHN, drill hole number; NCS, number of coal seams; N_DEN, normalization value of density; N_GR, normalization value of natural gamma; N_AC, normalization value of acoustic; A_d-MV, measured value of ash yield; A_d-PV, predicted value of ash yield; AE, absolute error; RE, relative error.

). Volatile matter (V_daf), predicted via LLD, GR, and AC inputs, exhibited absolute errors of −8.65% to 7.16% (average: −0.64%) and relative errors of −18.33% to 22.75% (average: −0.10%) (

Table 5. Error analysis of volatile matter yield in coal.

DHN	NCS	NLLD	NGR	NAC	Vdaf-PV (%)	Vdaf-MV (%)	AE (%)	RE (%)
KZ-1	8	0.32	0.40	0.05	38.53	47.18	−8.65	−18.33
KZ-4	8	0.32	0.39	0.04	38.50	39.11	−0.61	−1.56
KZ-5	7	0.00	0.72	0.24	38.62	39.48	−0.86	−2.18
KZ-8	8	1.00	0.11	1.00	38.63	36.44	2.19	6.01
XZ-1	8	0.46	0.57	0.99	38.63	31.47	7.16	22.75
XZ-3	7	0.34	0.40	0.14	38.94	42.01	−3.07	−7.31

DHN, drill hole number; NCS, number of coal seams; N_LLD, natural logarithm of lateral resistivity value and then normalizing; N_GR, normalization value of natural gamma; N_AC, normalization value of acoustic; V_daf-MV, measured value of volatile matter yield; V_daf-PV, predicted value of volatile matter yield; AE, absolute error; RE, relative error.

).

Fixed carbon (FC), optimized with LLD, GR, and AC inputs, yielded absolute errors of −1.67% to 2.52% (average: 0.78%) and relative errors of −3.07% to 5.29% (average: 1.63%) (

Table 6. Error analysis of fixed carbon content in coal.

DHN	NCS	LLD	GR	AC	FC-PV (%)	FC-MV (%)	AE (%)	RE (%)
KZ-1	8	0.32	0.40	0.05	52.66	51.23	1.43	2.79
KZ-4	8	0.32	0.39	0.04	52.74	54.41	−1.67	−3.07
KZ-5	7	0.00	0.72	0.24	51.41	49.72	1.69	3.40
KZ-8	8	1.00	0.11	1.00	51.56	52.08	−0.52	−1.00
XZ-1	8	0.46	0.57	0.99	51.56	50.36	1.20	2.38
XZ-3	7	0.34	0.40	0.14	50.13	47.61	2.52	5.29

DHN, drill hole number; NCS, number of coal seams; N_LLD, natural logarithm of lateral resistivity value and then normalizing; N_GR, normalization value of natural gamma; N_AC, normalization value of acoustic; FC-MV, measured value of fixed carbon; FC-PV, predicted value of fixed carbon; AE, absolute error; RE, relative error.

). For total sulfur (S_t,d), ρ_s, AC, and GR inputs produced absolute errors of −0.14% to 0.04% (average: 0.03%) and relative errors of −15.12% to 6.78% (average: −2.82%) (

Table 7. Error analysis of total sulfur content in coal.

DHN	NCS	ρs	GR	AC	St,d-PV (%)	St,d-MV (%)	AE (%)	RE (%)
KZ-2	8	0.69	0.14	0.08	0.73	0.86	−0.13	−15.12
KZ-5	7	0.52	0.44	0.04	0.65	0.65	0.00	0.00
KZ-8	7	0.51	0.10	0.05	0.55	0.54	0.01	1.85
XZ-1	8	0.30	0.42	0.06	0.65	0.63	0.02	3.17
XZ-3	8	0.03	0.72	0.09	0.89	1.03	−0.14	−13.59
XZ-7	7	0.67	1.00	0.07	0.63	0.59	0.04	6.78

DHN, drill hole number; NCS, number of coal seams; N_LLD, natural logarithm of lateral resistivity value and then normalizing; N_GR, normalization value of natural gamma; N_AC, normalization value of acoustic; S_t,d-MV, measured value of total sulfur content; S_t,d-PV, predicted value of total sulfur content; AE, absolute error; RE, relative error.

). Calorific value (Q_gr,ad), modeled with LLD, GR, and CGS inputs, showed absolute errors of −6.18% to 0.91% (average: −1.66%) and relative errors of −16.9% to 3.14% (average: −4.72%) (

Table 8. Error analysis of calorific value in coal.

DHN	NCS	LLD	GR	CGS	Qgr,ad-PV (MJ/kg)	Qgr,ad-MV (MJ/kg)	AE (%)	RE (%)
KZ-1	8	0.32	0.40	0.07	30.38	36.56	−6.18	−16.90
KZ-4	8	0.32	0.39	0.08	30.37	32.98	−2.61	−7.91
KZ-5	7	0.00	0.72	0.00	29.93	29.02	0.91	3.14
KZ-8	8	1.00	0.11	0.24	29.58	29.64	−0.06	−0.20
XZ-1	8	0.46	0.57	0.12	29.93	30.53	−0.60	−1.97
XZ-3	7	0.34	0.40	0.08	30.36	31.79	−1.43	−4.50

DHN, drill hole number; NCS, number of coal seams; N_LLD, natural logarithm of lateral resistivity value and then normalizing; N_GR, normalization value of natural gamma; N_CGS, natural logarithm of short-range gamma-gamma value and then normalizing; Q_gr,ad-MV, measured value of calorific; Q_gr,ad-PV, predicted value of calorific; AE, absolute error; RE, relative error.

).

Analysis of absolute and relative errors confirms the neural network model’s accuracy, with minimal deviations supporting its utility for precise coal quality stratification and intelligent mining operations.

5. Conclusions

An approximate analysis of No. 7 and No. 8 coal samples from the Shanxi Formation within the study area indicates these are gas coals characterized by low sulfur, low ash, low fixed-carbon, high volatile matter, and high calorific value. Coal quality correlation analysis reveals a significant negative correlation between ash yield and both fixed-carbon content and calorific value. Significant correlations exist between fixed-carbon and volatile matter yield, and between fixed-carbon and calorific value. A significant positive correlation is observed between total sulfur content and volatile matter yield.

Analysis of the response relationship between coal quality parameters (moisture, ash, volatile matter, fixed carbon, total sulfur, calorific value) and geophysical logging parameters indicates that a quantifiable relationship exists. The geophysical logging response characteristics of coal quality parameters are controlled by factors including coalification degree, maceral composition, reservoir properties, and mineral types and mass fractions. Furthermore, it has been demonstrated that BP neural network models can effectively establish predictive models for the geophysical logging response of coal quality parameters. This modeling approach provides technical and theoretical support for the accurate identification of deep coal seam quality and intelligent mine mining.

Identifying coal quality indicators through logging parameters exhibits a degree of innovation. However, the complexity and diversity inherent in coal seam mineral composition, sedimentary environment, and coalification processes necessitate systematic exploration of the logging response mechanism for coal quality parameters. Concurrently, the availability of high-resolution logging parameters requires further investigation, as this will determine the feasibility of establishing accurate coal quality prediction models via geophysical logging.