Multi-modal mineral prediction agent coordination method and system based on LLM dynamic perception

Abstract

The application relates to the technical field of mineral exploration, and in particular to a multi-modal mineral prediction agent cooperation method and system based on LLM dynamic perception. The method comprises the following steps: acquiring surface multi-modal data of a permafrost layer mineral, analyzing the interaction between a surface methane-loving bacterial community structure and permafrost respiration gas in the permafrost layer above the mineral based on the surface multi-modal data, and obtaining a permafrost microorganism-gas coupling information set; based on the permafrost microorganism-gas coupling information set, simulating the migration path of the gas in the heterogeneous permafrost through an LLM generator, associating and mapping the migration path with the mineral type, and obtaining a mineral characteristic inference set; based on the mineral characteristic inference set, positioning the methane seepage and the microbial hotspot area through a discriminator, generating a grid sampling-fixed point deep digging cooperation decision, and outputting a multi-modal mineral prediction agent cooperation log of minimum invasion operation. The balance between accurate target positioning and minimum environmental invasion is achieved, and the reliability of mineral prediction is improved.

Description

Technical Field

[0001] This application relates to the field of mineral exploration technology, and in particular to a collaborative method and system for multimodal mineral prediction intelligent agents based on LLM dynamic perception. Background Technology

[0002] Existing mineral exploration methods in permafrost regions mainly rely on single geophysical or surface geochemical measurements. These methods typically analyze data (such as gas concentrations or microbial samples) statically and in isolation, and simply correlate them with minerals. They lack in-depth analysis of the dynamic coupling mechanism between gas migration, microbial activity, and underground minerals in the permafrost environment. In addition, existing exploration deployments (such as uniform grid sampling) are often not targeted enough, have limited exploration accuracy, and are highly invasive to fragile permafrost ecosystems.

[0003] Due to the heterogeneity of permafrost and the complexity of gas-microbe interactions, existing static, single-modal exploration methods are unable to accurately reveal the intrinsic relationship between mineral resources and surface biogeochemical signals, resulting in high uncertainty in prediction results, low exploration efficiency, and potential unnecessary disturbance to sensitive ecological environments. Therefore, there is an urgent need for a new method that can dynamically integrate multi-source data, intelligently simulate coupling processes, and guide precise and minimally invasive operations. Summary of the Invention

[0004] This application provides a collaborative method and system for multimodal mineral prediction agents based on LLM dynamic perception to solve the above problems.

[0005] In a first aspect, this application provides a collaborative method for multimodal mineral prediction agents based on LLM dynamic perception, the method comprising:

[0006] Multimodal surface data of mineral deposits in permafrost is acquired. Based on this data, the interaction between the surface methanophilic bacterial community structure and the respiration gases in the permafrost layer above the mineral deposits is analyzed to obtain a permafrost microbial-gas coupling information set. Based on this information set, an LLM generator is used to simulate the gas migration paths in heterogeneous permafrost, and these migration paths are mapped to mineral types to obtain a mineral feature inference set. Based on this feature inference set, a discriminator is used to locate methane leakage and microbial hotspots, generating a grid sampling-fixed-point deep mining collaborative decision and outputting a multimodal mineral prediction agent collaborative log that minimizes intrusion operations.

[0007] Through the above technical solutions, multimodal data fusion analysis of surface biochemical signals provides a new non-invasive indicator for the detection of hidden minerals in permafrost areas. LLM simulation of gas migration and intelligent inference of mineralization characteristics improve the rationality and reliability of deep prediction. Intelligent discrimination and collaborative exploration decision generation achieve a balance between accurate target area positioning and minimal environmental intrusion. The final output collaborative log ensures full transparency and traceability, forming a complete intelligent, accurate and environmentally friendly mineral prediction technology chain.

[0008] Optionally, the surface multimodal data includes surface methane concentration data and methanotrophic microbial distribution data. Based on the surface methane concentration data, the spatial diffusion patterns and temporal fluctuation rhythms of surface methane are analyzed to identify continuously evolving methane anomaly characteristics, thus obtaining surface methane dynamic anomaly information. Based on the methanotrophic microbial distribution data, the synergistic changes in methanotrophic microbial population abundance and metabolic potential are analyzed to identify methanotrophic microbial community characteristics that are sensitive to gas changes and have active metabolic functions, thus obtaining surface methanotrophic bacterial community response information. Based on the surface methane dynamic anomaly information, combined with the surface methanotrophic bacterial community response information, the bidirectional interaction between methane anomalies and methanotrophic microbial responses in terms of spatial location and change trends is analyzed to obtain the permafrost microbial-gas coupling information set.

[0009] Optionally, based on the surface methane concentration data, spatial trend surface analysis is used to analyze the gradual and abrupt changes in methane concentration in continuous geographic space to obtain spatial context information of methane diffusion; based on the surface methane concentration data, time series decomposition and sliding window statistics are used to analyze the rate of change characteristics of methane concentration in multiple continuous time series to obtain temporal rhythm information of methane fluctuations; based on the spatial context information of methane diffusion, combined with the temporal rhythm information of methane fluctuations, the coupling relationship between spatial diffusion trends and temporal fluctuation patterns in the same geographical location is analyzed to identify areas with stable source orientation in space and regular enhancement or weakening characteristics in time, thereby obtaining the dynamic anomaly information of surface methane.

[0010] Optionally, based on the distribution data of the methophile microorganisms, the overlap patterns in the spatial distribution of different methophile microbial populations and the abundance relationship of the methane oxidation functional genes they carry are analyzed through functional gene abundance statistics to obtain information on the synergistic combination of microbial functions. Based on the information on the synergistic combination of microbial functions, node screening is used to analyze the contribution of different methophile microbial populations to maintaining the overall methane oxidation metabolic function of the community, to obtain information on the contribution of microbial populations. Based on the information on the contribution of microbial populations, dynamic simulation of metabolic pathways is used to analyze the correlation between the abundance changes of methophile microbial populations with prominent buffering and maintenance functions and the changes in the overall methane oxidation metabolic potential of the methophile microbial community, to obtain the response information of the surface methophile bacterial community.

[0011] Optionally, based on the methane anomaly characteristics and the methophile microbial community characteristics, spatial overlay analysis is performed to analyze the degree of spatial overlap between methane anomaly regions with stable source orientation and methophile microbial hotspot regions with prominent metabolic maintenance functions, obtaining spatial matching information; based on the methane fluctuation temporal rhythm information and the surface methophile bacterial community response information, phase coupling analysis is performed to analyze the degree of temporal synergy between the phase of regular methane concentration fluctuations and the phase of changes in the metabolic activity of methophile microbial communities, obtaining trend synergy information; based on the spatial matching information and the trend synergy information, a mutual feedback mechanism is used to identify and analyze the spatial diffusion patterns of methane anomalies modulated by the metabolic activities of methophile microorganisms at corresponding locations, and the metabolic response trends of methophile microbial communities driven by the corresponding methane concentration fluctuation rhythms, thereby identifying stable mutual indicative and reinforcing coupling units between gas migration paths and methophile microbial metabolic hotspots, obtaining the permafrost microbial-gas coupling information set.

[0012] Optionally, based on the coupling units, the spatial arrangement orientation and intensity gradient of different coupling units are analyzed to generate an initial migration trend network tracing back from the methanotrophic microbial hotspot to the potential gas source. Based on the initial migration trend network, according to prior knowledge of the geological structure of heterogeneous permafrost, the influence of geological interfaces and fracture networks on gas migration trends is analyzed. Advantageous channels that conform to geological constraints are screened and strengthened in the initial migration trend network to form a modified gas migration path. Based on the gas migration path, according to the characteristic gas fingerprint knowledge associated with the formation or alteration of different types of minerals, the correspondence between the changing patterns of gas-microbe coupling intensity on the path and the mineral type is analyzed to obtain the mineral feature inference set.

[0013] Optionally, based on the gas migration path, the gas-microbe coupling strength characterized by the permafrost microbe-gas coupling information set at different spatial nodes along the path is analyzed to obtain a coupling strength variation sequence distributed along the migration path; based on the coupling strength variation sequence, according to the measured gas-microbe coupling pattern library of historically proven mineral samples, the similarity between the spatial distribution pattern of the coupling strength variation sequence and the characteristic coupling patterns corresponding to known types of minerals is analyzed to obtain the pattern matching degree; based on the pattern matching degree, the correlation between gas accumulation trend and path geometric features is analyzed to infer the possible types and spatial distribution characteristics of minerals, and the mineral feature inference set is obtained.

[0014] Optionally, based on the possible mineral types and spatial distribution characteristics, the spatial distribution of the pattern matching degree is analyzed to identify continuous areas where the pattern matching degree is higher than a set threshold, thus obtaining methane leakage areas and microbial hotspots as priority exploration targets. Based on the methane leakage areas and the microbial hotspots, the requirements for different sampling methods to achieve efficient coverage and core verification are analyzed, and a collaborative operation plan is generated to implement regular grid sampling around the hotspots to determine anomaly boundaries and to implement intensive fixed-point deep mining inside the hotspots to verify the existence of minerals, resulting in a grid sampling-fixed-point deep mining collaborative decision. Based on the grid sampling-fixed-point deep mining collaborative decision, the geographical environment and permafrost ecological sensitivity of the methane leakage and microbial hotspots are analyzed, the operation path is adjusted to avoid ecologically vulnerable points, and structured records are made to output the collaborative log of the multimodal mineral prediction agent.

[0015] Optionally, based on the methane leakage area and the microbial hotspot area, the spatial distribution information and aggregation pattern of signal points are analyzed to obtain the spatial distribution pattern information of the target area; based on the spatial distribution pattern information, the grid sampling requirement information is obtained by analyzing the peripheral area where the methane leakage and microbial hotspot signals are discretely distributed but point to an undefined range; based on the spatial distribution pattern information, the fixed-point deep mining requirement information is obtained by analyzing the internal area where the methane leakage and microbial hotspot signals converge with high intensity and the convergence position coincides with the target point position indicated by the mineral feature inference set; based on the grid sampling requirement information and the fixed-point deep mining requirement information, the sequential relationship in the operational logic and the complementary relationship in space between the two types of requirements are analyzed to obtain the requirements for different sampling methods used to generate the collaborative operation scheme.

[0016] Secondly, this application provides a multimodal mineral prediction intelligent agent collaborative system based on LLM dynamic perception, the system comprising:

[0017] The coupling analysis module is used to acquire surface multimodal data of permafrost minerals. Based on the surface multimodal data, it analyzes the interaction between the surface methanophilic bacterial community structure and the permafrost respiration gases in the permafrost layer above the minerals, obtaining a permafrost microbial-gas coupling information set. The feature inference module is used to simulate the gas migration path in heterogeneous permafrost using an LLM generator based on the permafrost microbial-gas coupling information set, and associate the migration path with the mineral type to obtain a mineral feature inference set. The collaborative prediction module is used to locate methane leakage and microbial hotspots using a discriminator based on the mineral feature inference set, generate grid sampling-fixed-point deep mining collaborative decision, and output a multimodal mineral prediction agent collaborative log that minimizes intrusion operations. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram illustrating an application scenario provided in one embodiment of this application;

[0020] Figure 2 A flowchart of a multimodal mineral prediction agent collaborative method based on LLM dynamic perception provided in an embodiment of this application;

[0021] Figure 3 This is a schematic diagram of the structure of a multimodal mineral prediction intelligent agent collaborative system based on LLM dynamic perception, provided in an embodiment of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0023] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0024] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.

[0025] Because permafrost is heterogeneous and the interaction mechanism between gases and microorganisms is complex, current static and single-mode exploration methods often fail to accurately establish the correlation between underground mineral resources and surface biochemical signals. This leads to significant uncertainty in prediction, low exploration efficiency, and may cause unnecessary disturbance to the fragile ecological environment.

[0026] Based on this, this application provides a collaborative method and system for multimodal mineral prediction intelligent agents based on LLM dynamic perception. By relying on multimodal data fusion and analysis of surface biochemical signals, it provides a new non-invasive indicator for the exploration of concealed minerals in permafrost areas. By combining large language models to simulate gas migration processes and intelligently infer mineralization characteristics, it effectively enhances the rationality and reliability of deep prediction. Through intelligent discrimination to generate collaborative exploration schemes, it achieves the collaborative optimization of accurate target area positioning and minimal environmental disturbance. Finally, the output of the full-process collaborative log ensures process transparency and traceability, thereby constructing a complete technical system for mineral prediction that combines intelligence, accuracy, and green sustainability.

[0027] Figure 1 This application provides an illustration of an application scenario. In the prediction of mineral resources in permafrost, the method provided in this application is used to construct a non-invasive intelligent prediction system for hidden mineral resources in permafrost areas based on multimodal data and large model simulation. The system achieves a balance between exploration efficiency and minimal environmental disturbance through precise target area positioning, and its fully traceable collaborative logs ensure the reliability and transparency of the method.

[0028] Specifically, the method provided in this application can be applied to any server, which interacts with ground monitoring points to obtain multimodal surface data provided by the ground monitoring points. This provides a new non-invasive indicator for the detection of hidden minerals in permafrost areas and outputs a multimodal mineral prediction intelligent agent collaborative log to mineral exploration personnel, forming a complete intelligent, accurate and environmentally friendly mineral prediction technology chain.

[0029] For specific implementation details, please refer to the following examples.

[0030] Figure 2 This is a flowchart illustrating a multimodal mineral prediction agent collaborative method based on LLM dynamic perception, provided in one embodiment of this application. The method of this embodiment can be applied to servers in the above scenarios. For example... Figure 2 As shown, the method includes:

[0031] S201. Obtain surface multimodal data of permafrost mineral deposits. Based on the surface multimodal data, analyze the interaction between the surface methanophilic bacterial community structure and the permafrost respiratory gases in the permafrost layer above the mineral deposits to obtain a permafrost microbial-gas coupling information set.

[0032] Surface multimodal data can be a comprehensive dataset reflecting the physical, chemical, and biological characteristics of the permafrost surface, with ground monitoring stations as the data source. Surface methanophilic bacterial community structure can be the composition, abundance, and spatial distribution of microbial populations existing on the permafrost surface that use methane as their primary carbon and energy source. Permafrost respiration gases can be the mixture of gases released from the permafrost layer under the influence of microbial activity and physicochemical processes in the permafrost environment. The permafrost microbial-gas coupling information set can be a dataset of data on the spatial distribution characteristics and activity levels of methanophilic bacterial communities and the correlation between the concentration and release flux of permafrost respiration gases (especially methane).

[0033] Specifically, mineral exploration in permafrost regions faces significant challenges because thick permafrost covers deep ore bodies, existing physical methods suffer from severe signal attenuation, and large-scale drilling is costly and environmentally invasive. However, the formation and occurrence of deep mineral deposits are often accompanied by specific fluid migration and microbial geochemical effects, which may leave detectable "fingerprints" on the surface. Recent studies have shown that certain metal deposits (such as volcanic uranium deposits and certain sulfide deposits) produce fluids rich in gases such as methane during their formation or later alteration. These gases can migrate upwards along tectonic fissures to the permafrost layer and nourish specific methanophilic bacterial communities on the surface, thereby altering the surface gas emission patterns and microbial ecological structure.

[0034] S202. Based on the frozen soil microbial-gas coupling information set, the gas migration path in heterogeneous frozen soil is simulated by an LLM generator, and the migration path is associated with mineral type to obtain a mineral feature inference set.

[0035] An LLM generator can be a large-scale language model trained with domain-specific knowledge (including literature and reports on permafrost physics, gas transport mechanics, and mineral deposit geology). Heterogeneous permafrost can be a permafrost layer with complex internal structure and significant spatial variations in ice-water-rock components. The mineral feature inference set can be a set of probabilistic inference results output by the LLM generator, which correlates features such as potential mineral type, size, and burial depth.

[0036] Specifically, while permafrost microbial-gas coupling information indicates surface anomalies, the path of gas migration from deep mineral sources to the surface is strongly controlled by the heterogeneity of permafrost. For example, high-ice-content lenses may block airflow, while fracture networks may become dominant channels. Existing numerical simulation methods rely heavily on accurate initial parameters and boundary conditions, which are often lacking in the early stages of exploration, resulting in high uncertainty in simulation results.

[0037] S203. Based on the mineral feature inference set, the discriminator locates methane leakage and microbial hotspots, generates grid sampling-fixed-point deep mining collaborative decision-making, and outputs a multimodal mineral prediction agent collaborative log that minimizes intrusion operations.

[0038] The discriminator can be a classifier based on a deep learning model (such as a convolutional neural network). Grid sampling-fixed-point deep mining collaborative decision-making can be an optimized exploration action plan. The multimodal mineral prediction agent collaborative log can be a complete document recording the entire agent's collaborative work process, intermediate results, final decisions, and the basis for them.

[0039] Specifically, the mineral feature inference set provides a variety of possible deep mineralization hypotheses, but it needs to be transformed into an executable, efficient, and environmentally intrusive field exploration plan. Directly conducting intensive drilling over a large area is neither economical nor environmentally friendly. Therefore, this step introduces a discriminator, which takes the mineral feature inference set (especially the surface projection area corresponding to the high-probability mineralization area) and high-resolution raw or processed multimodal data (such as methane concentration gradient maps and bacterial abundance hotspot maps) as input. Through its powerful pattern recognition capabilities, it further accurately locates the surface location that is most likely to directly reflect deep mineralization activity and has the strongest signal, namely the main methane leakage channel and the core area of ​​microbial hotspots.

[0040] The method provided in this embodiment utilizes multimodal data fusion analysis of surface biochemical signals to offer a new non-invasive indicator for the detection of concealed minerals in permafrost regions. By using LLM to simulate gas migration and intelligently infer mineralization characteristics, the rationality and reliability of deep prediction are significantly improved. Through intelligent discrimination to generate collaborative exploration decisions, a balance is achieved between accurate target area positioning and minimal environmental intrusion. The final output collaborative log ensures full transparency and traceability, forming a complete intelligent, accurate, and environmentally friendly mineral prediction technology chain.

[0041] In some embodiments, surface multimodal data includes surface methane concentration data and methanotrophic microbial distribution data. Based on surface methane concentration data, the spatial diffusion patterns and temporal fluctuation rhythms of surface methane are analyzed to identify methane anomaly characteristics with continuous evolutionary patterns, thereby obtaining surface methane dynamic anomaly information. Based on methanotrophic microbial distribution data, the synergistic changes in methanotrophic microbial population abundance and metabolic potential are analyzed to identify methanotrophic microbial community characteristics that are sensitive to gas changes and have active metabolic functions, thereby obtaining surface methanotrophic bacterial community response information. Based on surface methane dynamic anomaly information, combined with surface methanotrophic bacterial community response information, the bidirectional effects of methane anomalies and methanotrophic microbial responses in terms of spatial location and change trends are analyzed to obtain a permafrost microbial-gas coupling information set.

[0042] Surface methane concentration data can reflect the spatial diffusion and temporal fluctuation characteristics of methane. Methanophilic microbial distribution data can record the species, quantity, and spatial distribution of microbial populations on permafrost surfaces that can utilize methane as a carbon and energy source. Surface methane dynamic anomaly information can be identified by comprehensively considering the spatial diffusion patterns and temporal fluctuation rhythms of methane, revealing methane anomaly characteristics with continuously evolving patterns. Surface methophilic bacterial community response information can reflect the sensitivity of methophilic microbial populations to changes in methane gas and their metabolic activity.

[0043] Specifically, in the field of permafrost mineral prediction, existing mineral exploration methods often rely on single geological drilling and geophysical exploration techniques. These methods are not only highly invasive and costly, but also easily affected by the special geological conditions of permafrost (such as ice cover and complex soil structure), resulting in low prediction accuracy and vague target areas. Furthermore, during the formation or alteration of permafrost minerals, characteristic gases such as methane are released. When these gases migrate within the permafrost, they interact closely with the surface methanotrophic microbial community. Abnormal methane diffusion stimulates the growth, reproduction, and metabolic activities of methanotrophic microorganisms, and the active microbial community, in turn, influences the oxidation and migration processes of methane. This two-way interaction is a key "signal chain" reflecting mineral distribution. Addressing the aforementioned issues, it begins with synchronously acquired "surface multimodal data." "Surface methane concentration data" originates from a network of ground-deployed laser spectroscopic gas analyzers, recording concentration values ​​in a gridded manner (e.g., one measuring point every 100 meters). "Methanephilic microbial distribution data" is obtained through metagenomic sequencing of surface soil samples collected at the grid points. Firstly, for the methane concentration data, "spatial trend surface analysis" is employed to fit the concentration distribution surface across continuous geographic space, identifying major concentration ridges (indicating dominant diffusion channels) and abrupt change boundaries (e.g., concentration gradients exceeding, for example, 0.2). The data is analyzed to construct "spatial context information of methane diffusion" by identifying regions with concentrations in ppm / m. Simultaneously, "time series decomposition and sliding window statistics" methods are used to process continuous monitoring data spanning several months. After removing seasonal trends, the rate of change within the sliding window is calculated to identify regions with regular fluctuation rhythms (such as daily evening concentration peaks), forming "methane fluctuation temporal rhythm information." Superimposing these two methods allows for the location of "surface methane dynamic anomaly information" that exhibits stable spatial directionality and regular temporal evolution. Secondly, for microbial data, bioinformatics tools are used for "functional gene abundance statistics" to quantify the abundance of functional genes encoding key enzymes such as methane monooxygenase (pmoA) at each sampling point. The spatial co-occurrence patterns of these genes with the abundance of different microbial populations (such as Methylobacter) are analyzed to obtain "microbial functional synergistic combination information." Furthermore, by employing "node screening" algorithms (such as centrality analysis based on network topology), key populations that contribute most to maintaining the stability of the entire methane oxidation metabolic network are identified. Then, through "metabolic pathway dynamic simulation," the relative abundance changes of these key populations are correlated with the calculated values ​​of theoretical community metabolic potential, thereby extracting "surface methanotrophic bacterial community response information" that is metabolically active and sensitive to changes. Finally, the above two types of information are integrated: the spatial overlap index between methane anomaly areas and microbial hotspot areas is calculated through "spatial overlay analysis"; and the phase difference between the methane concentration fluctuation period and the microbial functional gene expression fluctuation period is quantified through "phase coupling analysis" (such as calculating cross-wavelet spectra).Based on regions with high spatial matching and high temporal synergy, the "mutual feedback mechanism identification" logic is used to explain how the diffusion path of methane is modulated by the consumption of active microorganisms below (forming a concentration "shaded area"), and how the hot spot distribution and activity rhythm of microorganisms are shaped by the path and pulse of gas supply. Finally, a series of coupling units in the "permafrost microorganism-gas coupling information set" are precisely defined and output.

[0044] The approach provided in this embodiment goes beyond simply overlaying multi-source data. Through in-depth mining and mechanism fusion, it transforms surface-observed methane signals and microbial responses into a coupled information set that reflects the dynamic balance between deep gas transport processes and surface biological consumption processes. This significantly improves the anti-interference capability and interpretability of clues for identifying mineral-associated gas leakage in permafrost regions, providing a solid, reliable, and mechanism-rich input foundation for subsequent intelligent path simulation and mineral type inference.

[0045] In some embodiments, based on surface methane concentration data, spatial trend surface analysis is used to analyze the gradual and abrupt changes in methane concentration in continuous geographic space, obtaining spatial context information of methane diffusion; based on surface methane concentration data, time series decomposition and sliding window statistics are used to analyze the rate of change characteristics of methane concentration in multiple continuous time series, obtaining temporal rhythm information of methane fluctuations; based on the spatial context information of methane diffusion, combined with the temporal rhythm information of methane fluctuations, the coupling relationship between spatial diffusion trends and temporal fluctuation patterns in the same geographical location is analyzed, identifying areas with stable source orientation in space and regular enhancement or weakening characteristics in time, obtaining dynamic anomaly information of surface methane.

[0046] Spatial diffusion patterns can refer to the gradual or abrupt distribution characteristics of methane across a continuous geographic space on the permafrost surface. Temporal fluctuation rhythms can refer to the rate and regularity of change in methane concentration over multiple continuous time series. Methane anomalies with continuous evolution patterns can be methane concentration anomalies that, after excluding accidental disturbances, exhibit a stable spatial distribution pattern and a traceable temporal trend.

[0047] Specifically, in permafrost mineral prediction, the dynamic behavior of surface methane is a key apparent signal revealing underground gas migration and potential mineral occurrence. However, the permafrost environment is complex, and methane concentration is affected by various instantaneous factors such as ambient temperature and microbial activity, exhibiting high spatial heterogeneity and temporal fluctuations. If only single-point or instantaneous methane concentrations are analyzed, random fluctuations or local disturbances are easily misjudged as mineral-induced anomalies, leading to inaccurate predictions and high exploration costs. To address these issues: First, spatial trend surface analysis, a geostatistical tool, is used to perform spatial interpolation and surface fitting on discretely collected surface methane concentration data to filter out random noise and extract spatial distribution patterns that clearly reflect the overall diffusion direction, concentration gradient changes, and possible abrupt boundary changes (such as sudden increases along fault lines) of methane at the regional scale, thereby obtaining information on the spatial network of methane diffusion. Then, time series decomposition methods (such as STL decomposition) are used to analyze long-term monitored methane data. Concentration time series data are decomposed into trend, seasonal, and residual components. Combined with a sliding window statistical method, the rate of change, periodic fluctuation characteristics, and statistical properties within local time windows of methane concentration at different time scales are quantitatively characterized, thus obtaining methane fluctuation temporal rhythm information. Finally, by combining spatial overlay analysis and temporal series correlation analysis using a geographic information system, the spatial contextual features (e.g., whether it is a diffusion source, gradient direction) corresponding to each geographic unit (e.g., raster pixel) are comprehensively correlated and coupled with their corresponding temporal rhythm features (e.g., whether the trend is continuously rising, whether the rate of change is stable). This identifies geographic regions that exhibit clear and stable diffusion source orientation in space (e.g., the highest concentration point with outward gradient radiation), and whose concentration at that source point shows a statistically significant regular evolutionary trend in time. Integrating the information from these regions yields highly reliable surface methane dynamic anomaly information.

[0048] The method provided in this embodiment comprehensively utilizes multi-dimensional analysis techniques, such as spatial trend surface analysis, time series decomposition, and sliding window statistics, to deeply explore the connotation of surface methane concentration data from two aspects: spatial diffusion pattern and temporal evolution law. Through detailed analysis of the coupling relationship between the two, random environmental interference can be effectively eliminated, and surface methane dynamic anomalies with continuity and regularity can be accurately identified, which are more likely to point to associated gas sources of underground minerals.

[0049] In some embodiments, based on the distribution data of methophiles, the overlap patterns in the spatial distribution of different methophile microbial populations and the abundance relationship of the methane oxidation functional genes they carry are analyzed through functional gene abundance statistics to obtain information on the synergistic combination of microbial functions. Based on the information on the synergistic combination of microbial functions, node screening is used to analyze the contribution of different methophile microbial populations to maintaining the overall methane oxidation metabolic function of the community, to obtain information on the contribution of microbial populations. Based on the information on the contribution of microbial populations, dynamic simulation of metabolic pathways is used to analyze the correlation between the abundance changes of methophile microbial populations with prominent buffering and maintenance functions and the changes in the overall methane oxidation metabolic potential of the methophile microbial community, to obtain information on the response of surface methophile bacterial communities.

[0050] Functional gene abundance statistics can be an analytical method used to quantify the number of specific functional genes carried by different microbial populations. Microbial functional synergistic information can be a set of information consisting of the spatial overlap patterns of different methophile microbial populations and the abundance correlations of the methane oxidation functional genes they carry. Node screening can be an analytical method based on community functional contribution. Microbial population contribution information can be a quantitative representation of the role played by different methophile microbial populations in maintaining the methane oxidation metabolic function of the entire methophile microbial community. Metabolic pathway dynamic simulation can be a method that constructs a microbial metabolic network model.

[0051] Specifically, in permafrost mineral prediction, direct detection of deep ore bodies is costly and highly invasive, while surface gas chemical exploration in heterogeneous permafrost often presents ambiguities due to complex gas transport pathways and weak or dispersed surface concentration signals. Methanophilic microorganisms, as a class of microorganisms that rely solely on methane for carbon and energy, are extremely sensitive to minute changes in surface methane concentration in their community structure and metabolic activity, and can serve as "biosensors" to reveal underground gas transport and mineral deposits. To address the above issues, the research began with high-throughput sequencing and bioinformatics analysis of soil samples collected in the field to obtain raw data on the distribution of methanophilic microorganisms (this data includes an abundance of 8.5 × 10^5 cells per gram of soil for the genus *Methylocystis* at sampling point A, and a copy number of the pmoA functional gene carried by these microorganisms of 3.2 × 10^4 per gram of soil). First, using functional gene abundance statistics, spatial association analysis was performed on all detected methanophilic microbial populations (such as Methylomonas and Methylobacter). The overlap index of their distribution at different sampling points was calculated, and the relative abundance ratio of key methane oxidation functional genes (such as pmoA and mmoX) carried by them was statistically analyzed, thereby generating information on microbial functional synergistic combinations. This information can reveal potential collaborative patterns, such as "in region B, the high abundance of the genus *Methylococcus* is always accompanied by the high abundance of a specific pmoA genotype in the genus *Methylosarcina*". Next, using a node screening technique based on network analysis, the above synergistic combination information was constructed into a network model with microbial populations as nodes and functional gene association strength as edges. By calculating topological indices such as degree centrality and betweenness centrality of nodes, the importance of each population in maintaining the connectivity and functional stability of the entire network was quantitatively assessed. For example, "Methylococcus sp. Strain A" was identified as a core hub in the network across multiple sampling points, thus outputting quantitative information on the contribution of microbial populations.Finally, dynamic simulation of metabolic pathways was conducted: Based on the methane oxidation pathway in the Kyoto Encyclopedia of Genes and Genomes (KEGG) database, a computational model incorporating key enzymatic reactions was constructed. The abundance changes of the aforementioned high-contribution core populations (e.g., simulating a 30% decrease in their abundance in response to a methane concentration gradient) were used as input parameters. The model was run to simulate the dynamic response of metabolic potential indicators such as methane oxidation flux and intermediate product concentration of the entire community. By analyzing the strong coupling relationship between the fluctuations in the abundance of these core populations and the simulated changes in metabolic potential, the response information of surface methanotrophic bacterial communities that are sensitive to gas changes and have active metabolic functions was accurately identified and output. For example, "a specific community combination dominated by Methylococcus sp. Strain A and synergistically composed of Methylosarcina sp. Strain B can provide early warning of a slight increase in underground methane flux due to structural changes."

[0052] The method provided in this embodiment deeply analyzes the response mechanism of methanephilic microbial communities to changes in the gas environment from three levels: functional synergy, key contributions, and dynamic metabolism. It accurately screens out the key microbial characteristics that truly play an indicative role, greatly enhancing the specificity and reliability of biogeochemical indicators in permafrost mineral prediction.

[0053] In some embodiments, based on methane anomaly characteristics and combined with methophilic microbial community characteristics, spatial overlay analysis is used to analyze the degree of spatial overlap between methane anomaly regions with stable source orientation and methophilic microbial hotspot regions with prominent metabolic maintenance functions, obtaining spatial matching information. Based on methane fluctuation temporal rhythm information and combined with surface methophilic bacterial community response information, phase coupling analysis is used to analyze the degree of temporal synergy between the phase of regular methane concentration fluctuations and the phase of changes in the metabolic activity of methophilic microbial communities, obtaining trend synergy information. Based on spatial matching information and trend synergy information, mutual feedback mechanisms are used to identify the spatial diffusion patterns of methane anomalies modulated by the metabolic activities of methophilic microorganisms at corresponding locations, and the metabolic response trends of methophilic microbial communities driven by the corresponding methane concentration fluctuation rhythms, thereby identifying stable mutual indicative and reinforcing coupling units between gas migration pathways and methophilic microbial metabolic hotspots, obtaining a permafrost microbial-gas coupling information set.

[0054] Spatial overlay is a technique that combines different types of geospatial feature layers to analyze their spatial overlap and correlation. Spatial matching information can be quantitative data on the degree of spatial overlap between methane anomaly regions with stable source orientation and methophilic microbial hotspots with prominent metabolic maintenance functions. Phase coupling analysis is a mathematical analysis method that analyzes the synchronicity and synergy of two time series in the changing phases. Trend synergy information can be quantitative data on the synergy of the phases of regular fluctuations in methane concentration and the phases of changes in the metabolic activity of methophilic microbial communities in the temporal trend. Mutual feedback mechanism identification can be an analytical method that reveals the mutual influence and mutual regulation relationships between two or more systems. Gas migration pathways can be the propagation trajectories of gases diffusing from potential sources to the surface in heterogeneous permafrost. Coupled units can be functional regions where there is a stable mutual indication and mutual reinforcement between gas migration pathways and methophilic microbial metabolic hotspots.

[0055] Specifically, in the field of permafrost mineral prediction, methane anomalies and microbial community responses observed on the surface are both indirect representations of deep mineral information. However, their mechanisms are complex and may be affected by various environmental factors. If only the spatial patterns of gas anomalies or the metabolic characteristics of microbial communities are analyzed independently, misjudgments are very likely to occur. For example, short-term methane releases that are not of mineral origin or ordinary microbial enrichment areas unrelated to minerals may be misjudged as mineral-induced anomalies, leading to inaccurate positioning of subsequent exploration target areas and high drilling decision risks. To address the above problems, firstly, spatial overlay analysis technology of Geographic Information System (GIS) is used to overlay the spatial distribution raster layers of the above two. By calculating the product of the intensity of the two types of anomalies on each pixel or setting a threshold to determine the overlap state, spatial matching degree information is quantified. For example, a matching degree raster map can be output, in which areas with a matching degree higher than a set threshold (such as 70%) are initially marked as potential associated areas. Next, phase coupling analysis methods from the field of signal processing are introduced. For these potentially correlated regions, the corresponding long-term methane concentration data and microbial metabolic activity indices (such as potential oxidation rates calculated based on functional gene abundance) are extracted. Using methods such as Hilbert transform or calculating windowed cross-correlation functions, the phase difference and synchronicity of the two sequences on key periods (such as daily or seasonal cycles) are analyzed, quantifying trend coherence information. For example, regions with a phase synchronization index greater than 0.8 indicate a high degree of consistency in their gas-bacterial change rhythms. Finally, and most importantly, a feedback mechanism identification method based on time-series causal inference (such as Granger causality test or convergent cross-mapping algorithm) is employed. For candidate regions that simultaneously satisfy high spatial matching and high trend coherence, the lead-lag relationship and mutual predictive ability between their methane concentration change sequences and microbial activity change sequences are analyzed in depth. For example, the analysis may reveal that within a specific coupling unit, an increase in methane concentration in the previous time series can significantly predict an increase in microbial activity in the next time series (Granger causality holds), and conversely, an increase in microbial activity can lead to a slowdown in the subsequent methane concentration increase trend (feedback regulation exists). By identifying and confirming this stable bidirectional dynamic correlation, the coupling units that have a stable mutual indication and enhancement effect between gas migration pathways and microbial metabolic hotspots were finally accurately located from many anomaly areas. The set of all coupling units constitutes a highly reliable permafrost microbial-gas coupling information set, providing core inputs that have been doubly verified and supported by mechanisms for subsequent path simulation and mineral inference.

[0056] The method provided in this embodiment effectively avoids false positive interference that may be caused by single indicator or simple superposition analysis, improves the accuracy and reliability of identifying surface microbial geochemical anomalies in mineral deposits, and provides a more solid and high-quality input information foundation for subsequent use of LLM to simulate gas migration paths and infer mineral characteristics.

[0057] In some embodiments, based on coupling units, the spatial arrangement orientation and intensity gradient of different coupling units are analyzed to generate an initial migration trend network that traces back from methanotrophic microbial hotspots to potential gas sources. Based on the initial migration trend network, and according to prior knowledge of the geological structure of heterogeneous permafrost layers, the influence of geological interfaces and fracture networks on gas migration trends is analyzed. Advantageous channels that conform to geological constraints are screened and strengthened in the initial migration trend network to form a modified gas migration path. Based on the gas migration path, and according to the characteristic gas fingerprint knowledge associated with the formation or alteration of different types of minerals, the correspondence between the changing patterns of gas-microbe coupling intensity on the path and mineral types is analyzed to obtain a mineral feature inference set.

[0058] Mineral types can refer to various minerals that may exist beneath the permafrost layer. Association mapping can be an analytical method for establishing the correspondence between gas migration path characteristics and mineral types. Spatial arrangement direction can refer to the geographical distribution direction of different coupled units. Intensity gradient can be the spatial variation trend of the gas-microbe interaction intensity embodied by the coupled units. Potential gas sources can be the original locations of gas generation or release. The initial migration trend network can be a network structure constructed based on the spatial characteristics of coupled units, initially reflecting the direction of gas tracing from microbial hotspots to potential sources. Prior geological structural knowledge can be existing geological structural information about the permafrost layer and underlying strata. Geological interfaces can be the boundaries between different geological layers. Fracture networks can be naturally formed or later developed interconnected fracture systems in the permafrost layer. Dominant pathways can be preferred paths with low gas migration resistance and good connectivity in heterogeneous permafrost. Modified gas migration paths can be gas migration trajectories that better conform to actual geological conditions after being screened and strengthened by geological structural constraints. Characteristic gas fingerprint knowledge can be information about gases with unique composition or concentration characteristics that are released during the formation or alteration of different types of minerals.

[0059] Specifically, in permafrost mineral prediction, directly inferring underground ore bodies from surface gas or microbial anomalies faces significant challenges. This is because permafrost exhibits strong heterogeneity and anisotropy; gas migration within it is not simply vertical ascent or uniform diffusion, but is strictly controlled by multiple geological factors such as lithological variations, fracture systems, and water-ice phases. Ignoring these geological constraints leads to exploration failure. To address these issues, spatial gradient analysis is employed to calculate the rate of change and direction of the "coupling strength" attribute of each coupled unit in two-dimensional geographic space (e.g., using surface gradient tools in ArcGIS or a self-developed differential algorithm). The gradient vectors of all units naturally connect and extend spatially, automatically generating an "initial migration trend network" composed of line segments pointing from high-value areas to low-value areas. Subsequently, the crucial correction step introduced "prior knowledge of the geological structure of heterogeneous permafrost layers." Specifically, this involved integrating regional geological maps, borehole lithology logs, and high-resolution seismic profile interpretation results to construct a three-dimensional digital geological model. This model clearly marked the spatial distribution of highly permeable gravel lenses, closed clay layers, and major tensional fracture zones. Next, network analysis and optimization algorithms from graph theory were applied (for example, in Python's NetworkX library, the initial network was treated as a graph, and differentiated weights were assigned to edges passing through different lithological strata or fractures based on the geological model). For example, the path weight for the gravel layer is set to 0.9, and the path weight for the clay layer is set to 0.1. Shortest or optimal path searches are performed to filter and strengthen channels that are highly consistent with the orientation of high-permeability geological structures, forming modified "gas migration paths." Finally, mineral type association mapping is performed: We establish a knowledge base containing gas fingerprints and microbial response patterns of multiple mineral types (for example, the knowledge base rules state that "natural gas hydrate" release paths often exhibit a continuous and stable increase in coupling strength with depth; "hydrothermal sulfide" related paths may show a sudden peak in coupling strength in specific alteration zones). Through pattern recognition and rule reasoning engines, the longitudinal variation curve shape of the coupling strength value on each optimized path is analyzed and matched with the rules in the knowledge base. For example, if a path has a steep peak plateau in its coupling strength curve at a depth of about 150 meters in the model inversion, and this location corresponds to a silicification alteration zone in the geological model, the system will map this path feature as "possibly indicating the presence of hydrothermal polymetallic sulfide mineralization at depth" and summarize such inferences into a structured "mineral feature inference set" to provide a direct basis for subsequent decision-making.

[0060] The method provided in this embodiment improves the accuracy of spatial location prediction of hidden mineral deposits under the permafrost layer, effectively reduces the blindness and cost of exploration, and achieves preliminary intelligent identification of mineral types. It provides a reliable scientific basis and decision support for subsequent deployment of more targeted and costly deep verification projects (such as drilling), reflecting the core value of multimodal intelligent agent collaborative prediction.

[0061] In some embodiments, based on the gas migration path, the gas-microbe coupling strength characterized by the permafrost microbe-gas coupling information set at different spatial nodes along the path is analyzed to obtain a coupling strength variation sequence distributed along the migration path; based on the coupling strength variation sequence, according to the measured gas-microbe coupling pattern library of historically proven mineral samples, the similarity between the spatial distribution pattern of the coupling strength variation sequence and the characteristic coupling patterns corresponding to known types of minerals is analyzed to obtain the pattern matching degree; based on the pattern matching degree, the correlation between gas accumulation trend and path geometric features is analyzed to infer the possible types and spatial distribution characteristics of minerals, and a mineral feature inference set is obtained.

[0062] Gas-microbe coupling strength can be a quantitative indicator characterizing the intensity of interaction between methanophilic microbial communities and methane gas at a specific spatial location, representing the concentrated information on gas-microbe coupling in permafrost. The coupling strength change sequence can be a continuous sequence of gas-microbe coupling strength values ​​arranged sequentially along a gas migration path. The measured gas-microbe coupling pattern library of historically proven mineral samples can be a standardized pattern database constructed by collecting and organizing measured gas concentrations, microbial community characteristics, and coupling relationships data from various proven mineral areas. Pattern matching degree can be a quantitative value of the similarity between the spatial distribution pattern of the coupling strength change sequence and the coupling patterns of known mineral characteristics. Gas accumulation pattern can be the concentrated distribution state of gas along a migration path. Path geometric characteristics can be the spatial morphological attributes of the gas migration path. Possible mineral types can be the mineral types inferred through pattern matching and association analysis that best match the coupling strength change pattern. Spatial distribution characteristics can be attributes such as the inferred geographical distribution range, extension direction, and morphology of minerals.

[0063] Specifically, existing permafrost mineral prediction technologies mainly rely on single physical (such as resistivity) or chemical (such as gas concentration) anomalies to delineate target areas. However, this inference based on isolated indicators faces fundamental defects in complex heterogeneous permafrost environments: on the one hand, surface gas anomalies may originate from biological activities or shallow gas escape from non-mineral factors, leading to false positives; on the other hand, different types of minerals (such as natural gas hydrates and metal sulfides) may produce similar gas concentration anomalies, which cannot be effectively distinguished by concentration alone, resulting in ambiguous prediction results and poor directionality. To address the aforementioned issues: First, spatial sequential sampling and multi-source information fusion technology is employed. A series of geographic reference nodes are set along the path at preset intervals (e.g., 50 meters). At each node, the "permafrost microorganism-gas coupling information set" generated upstream is invoked and parsed. This information set is essentially a data package containing multi-dimensional quantitative indicators such as spatial matching degree (e.g., overlap score based on overlay analysis) and trend synergy (e.g., correlation coefficient based on phase analysis). Through a preset weighted fusion model (e.g., setting the spatial matching degree weight to 0.6 and the trend synergy weight to 0.4), a standardized "comprehensive coupling strength value" is calculated for each node. This transforms the entire path into a quantifiable "coupling strength-spatial location" change curve, i.e., a coupling strength change sequence. Then, the system activates a template matching engine based on a case library. This engine calls upon the "Measured Gas-Microbe Coupling Pattern Library of Historical Proven Mineral Samples," a data warehouse that has undergone rigorous knowledge extraction. Each record in the library represents a typical coupling pattern "template" for a known type of mineral (taking "natural gas hydrate type" as an example). This template is generated by extracting features (such as the number of peaks, gradient change rate, and specific morphological inflection points) and standardizing the measured coupling strength curves above multiple proven mineral deposits of the same type. The matching engine uses dynamic time warping algorithm or morphological similarity correlation coefficient calculation to compare the currently detected sequence with all templates in the library one by one, and calculates the "pattern matching degree" score (a value between 0 and 1) for each comparison result. For example, the matching degree of the current sequence with the "natural gas hydrate type" template may be calculated as 0.88, while the matching degree with the "polymetallic sulfide type" template is only 0.45. Finally, the system performs spatial-attribute joint inference.It not only identifies the highest matching degree (e.g., 0.88) and its corresponding most likely mineral type (e.g., "natural gas hydrate"), but also analyzes the spatial aggregation pattern of high matching degree nodes (e.g., nodes with matching degree > 0.8) (whether they are continuous or discretely distributed), and correlates the analysis with the geometric characteristics of the gas migration path itself (e.g., the path shows obvious convergence at a certain coordinate, or extends along a certain directional fault zone). Combining this information, the system infers the "spatial distribution characteristics" of the mineral, for example: "The inferred mineral type is natural gas hydrate, its possible occurrence form is lenticular, the main distribution direction is 1-20 degrees northeast, and the core area corresponds to the continuous high matching degree area from path node number A-15 to A-35."

[0064] The method provided in this embodiment can effectively correlate and map the complex gas-microbe coupling dynamics observed on the surface to specific underground mineral types and their spatial occurrence, improving the accuracy and directionality of mineral prediction. This advances the prediction results from the vague level of "existing gas source anomaly" to the relatively accurate level of "suspected type of mineral with a certain spatial morphology".

[0065] In some embodiments, by accurately identifying priority exploration areas for methane leakage and microbial hotspots, the targeting and efficiency of mineral exploration are improved. A collaborative mode of perimeter grid sampling for boundary demarcation and internal intensified deep excavation for verification is adopted to achieve a balance between comprehensive exploration coverage and accurate core verification. Simultaneously, operational paths are adjusted based on geographical environment and permafrost ecological sensitivity to mitigate ecological risks and balance exploration benefits with ecological protection. The resulting multimodal collaborative log provides structured support for subsequent exploration decisions, helping to achieve the goals of efficient, accurate, and eco-friendly mineral exploration.

[0066] A continuous region can be a geographically connected area where all points within it have a pattern matching degree higher than the set threshold. Priority exploration targets can be areas selected based on pattern matching degree that are considered most likely to contain minerals or indicate mineral deposits. Methane seepage zones can be geographical areas with significantly abnormal surface or near-surface methane concentrations, and whose gas-microbe coupling patterns highly match mineral characteristics. Microbial hotspots can be geographical areas with abnormally active methanotrophic microbial communities, and whose distribution and variation patterns highly match mineral characteristics. Efficient coverage and core verification can be a high-level summary of the objectives of subsequent field exploration work. The need for different sampling methods can be the analysis conclusions regarding the applicable scenarios and specific requirements of regular grid sampling and high-precision fixed-point sampling after determining priority exploration targets, in order to achieve the objectives of "efficient coverage" and "core verification." A collaborative operation plan can be a grid sampling-fixed-point deep excavation collaborative decision-making, referring to a specific exploration action plan that integrates grid sampling and fixed-point deep excavation methods, which are spatially complementary and logically sequential. Geographical environment and permafrost ecological sensitivity can refer to the specific natural environmental conditions of the exploration target area, especially the spatial distribution and susceptibility to disturbance of ecologically vulnerable elements such as permafrost, vegetation, and water systems. Adjusting the work route can be a local optimization and modification of the pre-set sampling point locations or connecting routes during the actual implementation of a collaborative work plan to avoid irreversible damage to ecologically vulnerable areas. Avoiding ecologically vulnerable points is the core purpose of adjusting the work route, referring to proactively avoiding areas requiring special protection, such as areas sensitive to permafrost thawing, rare vegetation communities, animal habitats, or water sources, during route planning. Structured records can be standardized archives of content including, but not limited to, the final work plan, reasons for route adjustments, actual work points, field observation records, and sample information, according to pre-set, unified fields and formats.

[0067] Specifically, existing mineral exploration methods in permafrost regions often face enormous challenges. On the one hand, the permafrost environment is fragile, and large-scale invasive exploration (such as intensive drilling) can easily damage the permafrost structure, triggering environmental disasters such as thermal thaw and landslides, and the exploration costs are high. On the other hand, surface geochemical and geophysical anomaly signals are often weak, scattered, and complex, making it difficult for a single exploration method to accurately locate the ore body, resulting in low exploration efficiency and a low success rate of exploration.To address the aforementioned issues: The discriminator first employs a density-based spatial clustering algorithm (such as DBSCAN) to process the entire area's pattern matching raster data. It automatically identifies spatially continuous patches where the matching values ​​of all pixels within the patch are higher than a preset empirical threshold (e.g., a threshold of 0.85). These patches are then designated as "priority exploration targets" and further subdivided into "methane leakage zones" and "microbial hotspot zones" based on the dominant characteristics of their internal coupling information (gas anomaly-dominated or microbial response-dominated). Subsequently, for the delineated target area, the system retrieves raw multimodal signals from its interior and surrounding areas (such as surface methane concentration point data and microbial functional gene abundance heatmaps) for spatial analysis. Inter-area pattern analysis, for example, when signal points around the target area are found to be discretely scattered but their vector directions all point to an undefined center, it is determined that there is a need to efficiently determine the anomaly spatial range through "regular grid sampling," and sampling grid suggestions such as 500m × 500m are automatically generated; simultaneously, within the target area, a "converging core" is identified where methane concentration and microbial activity both show peak values ​​and spatially overlap. If this "converging core" coincides with the inferred mineral target location, it triggers the need for "intensified fixed-point deep excavation" verification. Based on this, shallow drilling or trenching points will be planned centered on the "converging core" and arranged radially or in a quincunx pattern (for example, designing 5 boreholes spaced 50 meters apart). The discriminator automatically integrates the analysis results of the two requirements mentioned above through spatial overlay and logical sorting to generate a detailed "grid sampling-fixed-point deep excavation collaborative decision-making" document. This document clearly defines the layout range, density, and sampling medium of the grid sampling, as well as the specific coordinates, depth, and sampling requirements of the fixed-point deep excavation. Subsequently, it integrates high-resolution remote sensing imagery, digital elevation models, and regional ecologically sensitive area vector layers to construct a comprehensive evaluation model of "geographical environment and permafrost ecological sensitivity." This model automatically identifies "ecologically vulnerable points" that need to be avoided on the path through overlay analysis (such as overlaying the preset work path with the ecologically vulnerable area layer). (For example, it identifies that the path will pass through a known ice wedge polygon tundra). The project automatically offsets and optimizes the travel route and some sampling point locations to generate a new "adjusted operation path". Finally, all input parameters (such as the threshold of 0.85), intermediate products (such as the identified continuous area boundary coordinates), generated collaborative schemes (including the final geographic coordinates of grids and points), environmental adjustment basis (such as avoiding a certain type of vulnerable habitat), and expected operation sequence in the entire decision chain are encoded, associated, and archived according to a predefined structured template, outputting a "multimodal mineral prediction intelligent agent collaborative log" containing spatial information, attribute information, and metadata. This log can be directly imported into a field mobile terminal to guide the exploration team to perform precise operations with minimal environmental intrusion.

[0068] The standardized and structured collaborative logs generated through the method provided in this embodiment ensure that the information is traceable and decisions are reviewable throughout the entire process from intelligent analysis to field operations. This not only provides precise action guidelines for current exploration, but also accumulates valuable and standardized digital assets for subsequent method optimization, effect evaluation, and environmental auditing.

[0069] In some embodiments, based on the methane leakage area and combined with the microbial hotspot area, the spatial distribution information and aggregation pattern of signal points are analyzed to obtain the spatial distribution pattern information of the target area; based on the spatial distribution pattern information, the grid sampling requirement information is obtained by analyzing the peripheral area where the methane leakage and microbial hotspot signals are discretely distributed but point to an undefined range; based on the spatial distribution pattern information, the fixed-point deep mining requirement information is obtained by analyzing the internal area where the methane leakage and microbial hotspot signals are highly converged and the convergence position coincides with the target point position indicated by the mineral feature inference set; based on the grid sampling requirement information and the fixed-point deep mining requirement information, the sequential relationship in the operational logic and the complementary relationship in space between the two types of requirements are analyzed to obtain the requirements for different sampling methods used to generate collaborative operation schemes.

[0070] Spatial distribution pattern information can be a comprehensive description reflecting the spatial dispersion, aggregation morphology, and relative positional relationship of methane leakage points and microbial hotspots. Grid sampling requirement information can describe the need for systematic, regular grid-like sampling to delineate anomaly boundaries in peripheral areas where signals are discretely distributed but collectively indicate a potential range. Targeted deep drilling requirement information can describe the need for precise, intensified drilling or trenching to directly verify the existence of mineral deposits in core areas where signals are highly convergent and coincide with predicted target points. Sequential relationships can refer to the logical order in which grid sampling and targeted deep drilling are implemented. Complementary relationships can refer to the complementary relationship between grid sampling and targeted deep drilling in terms of spatial coverage.

[0071] Specifically, in mineral exploration in permafrost regions, the permafrost environment is fragile. Large-scale, blind exploration is not only costly but also prone to ecological damage. Existing single sampling models (such as relying entirely on regular grids or targeting only a few anomalies) often cannot balance coverage efficiency and verification accuracy. To address these issues, the approach first relies on the fusion processing of spatial point data from methane seepage areas and microbial hotspots, employing spatial point model analysis (such as Ripley's method). K-function and kernel density estimation (KDE) techniques are used to quantify the discrete, clustered, or random distribution of signal points within the study area and generate a density surface map showing continuous changes in signal intensity. This objectively extracts the spatial distribution pattern information of the target area. For the discrete signal areas on the periphery of the pattern, density gradient analysis (calculating the first derivative of the density surface) combined with trend surface extrapolation methods (such as polynomial fitting) automatically identifies the "sharp transition zone" where the signal density transitions from background values ​​to outliers. This zone delineates the approximate outline of the anomaly, but the internal structure is blurred. Therefore, the system determines that there is a need for grid sampling here, which involves "densification control and delineation of boundaries through regular grid sampling (such as setting a 500m × 500m grid)". Meanwhile, for high-intensity convergence signal areas within the pattern, hotspot clustering analysis (such as Getis-Ord Gi* statistics) is applied to accurately identify statistically significant hotspot cores. Through spatial overlay and position verification (such as calculating the distance between the centroid of the hotspot core and the coordinates of the preferred target points predicted by the mineral feature inference set; if it is less than a set tolerance such as 100 meters, it is considered to be coincident), the spatial correspondence between it and the deep mineral target is confirmed. This leads to the determination that there is a need for "implementing encrypted fixed-point drilling or trenching (such as setting up 3 verification holes with the hotspot core as the center and a radius of 50 meters) to obtain direct verification evidence." Finally, based on the above two types of needs, decision trees or logical rules are used to model and clarify the sequential relationship between the two in the exploration logic and the complementary relationship in space (the grid covers the outer planar area, and the deep excavation focuses on the internal point-like target center), thereby outputting a structured and differentiated sampling requirement instruction set, providing accurate input for the intelligent agent to generate collaborative operation schemes.

[0072] The method provided in this embodiment effectively avoids the blindness and uniformity of sampling deployment, and achieves an organic combination of efficient general survey of potential mineral areas and accurate verification of high-value targets with limited exploration costs and minimal environmental disturbance. This significantly improves the efficiency and success rate of field verification work for mineral prediction in permafrost areas, while strengthening the protection of the fragile permafrost ecological environment.

[0073] Figure 3 This is a schematic diagram of the structure of a multimodal mineral prediction intelligent agent collaborative system based on LLM dynamic perception provided in an embodiment of this application, as shown below. Figure 3As shown, the multimodal mineral prediction intelligent agent collaborative system 300 based on LLM dynamic perception in this embodiment includes: a coupling analysis module 301, a feature inference module 302, and a collaborative prediction module 303.

[0074] The coupling analysis module 301 is used to acquire surface multimodal data of permafrost minerals, and based on the surface multimodal data, analyze the interaction between the surface methanophilic bacterial community structure and the permafrost respiration gas in the permafrost layer above the minerals to obtain a permafrost microbial-gas coupling information set; the feature inference module 302 is used to simulate the gas migration path in heterogeneous permafrost through an LLM generator based on the permafrost microbial-gas coupling information set, and map the migration path with the mineral type to obtain a mineral feature inference set; the collaborative prediction module 303 is used to locate methane leakage and microbial hotspots through a discriminator based on the mineral feature inference set, generate grid sampling-fixed-point deep mining collaborative decision, and output a multimodal mineral prediction agent collaborative log that minimizes intrusion operations.

[0075] Optionally, the coupling analysis module 301, when analyzing the interaction between the surface methanotrophic bacterial community structure and permafrost respiratory gases in the permafrost layer above the mineral deposit based on the surface multimodal data to obtain a permafrost microbial-gas coupling information set, is specifically used for: the surface multimodal data including surface methane concentration data and methanotrophic microbial distribution data; based on the surface methane concentration data, analyzing the spatial diffusion pattern and temporal fluctuation rhythm of surface methane, identifying methane anomaly characteristics with continuous evolution, and obtaining surface methane dynamic anomaly information; based on the methanotrophic microbial distribution data, analyzing the synergistic changes in methanotrophic microbial population abundance and metabolic potential, identifying methanotrophic microbial community characteristics that are sensitive to gas changes and have active metabolic functions, and obtaining surface methanotrophic bacterial community response information; based on the surface methane dynamic anomaly information, combined with the surface methanotrophic bacterial community response information, analyzing the bidirectional effect of methane anomaly and methanotrophic microbial response in terms of spatial location and change trend, and obtaining the permafrost microbial-gas coupling information set.

[0076] Optionally, the coupling analysis module 301, when analyzing the spatial diffusion pattern and temporal fluctuation rhythm of surface methane to identify methane anomaly characteristics with continuous evolutionary patterns and obtain surface methane dynamic anomaly information, specifically performs the following: based on the surface methane concentration data, it uses spatial trend surface analysis to analyze the gradual and abrupt changes in methane concentration in continuous geographical space to obtain methane diffusion spatial pattern information; based on the surface methane concentration data, it uses time series decomposition and sliding window statistics to analyze the rate of change characteristics of methane concentration in multiple continuous time series to obtain methane fluctuation temporal rhythm information; based on the methane diffusion spatial pattern information, combined with the methane fluctuation temporal rhythm information, it analyzes the coupling relationship between spatial diffusion trends and temporal fluctuation patterns in the same geographical location, identifies regions with stable source orientation in space and exhibit regular enhancement or weakening characteristics in time, and obtains the surface methane dynamic anomaly information.

[0077] Optionally, the coupling analysis module 301, when analyzing the synergistic changes in the abundance and metabolic potential of methophilic microbial populations, identifying the characteristics of methophilic microbial communities that are sensitive to gas changes and have active metabolic functions, and obtaining the response information of surface methophilic bacterial communities, is specifically used for: based on the methophilic microbial distribution data, analyzing the overlap patterns in the spatial distribution of different methophilic microbial populations and the abundance relationship of the methane oxidation functional genes they carry through functional gene abundance statistics, to obtain information on the synergistic combination of microbial functions; based on the information on the synergistic combination of microbial functions, using node screening, analyzing the contribution of different methophilic microbial populations in maintaining the overall methane oxidation metabolic function of the community, to obtain information on the contribution of microbial populations; based on the information on the contribution of microbial populations, using dynamic simulation of metabolic pathways, analyzing the correlation between the abundance changes of methophilic microbial populations with prominent buffering and maintenance functions and the changes in the overall methane oxidation metabolic potential of the methophilic microbial community, to obtain the response information of the surface methophilic bacterial community.

[0078] Optionally, the coupling analysis module 301, when analyzing the bidirectional effect of methane anomalies and methophile microbial responses in terms of spatial location and changing trends to obtain the permafrost microbial-gas coupling information set, is specifically used for: based on the methane anomaly characteristics and combined with the methophile microbial community characteristics, analyzing the degree of spatial overlap between methane anomaly regions with stable source orientation and methophile microbial hotspot regions with prominent metabolic maintenance functions through spatial overlay, to obtain spatial matching information; based on the methane fluctuation time rhythm information and combined with the surface methophile bacterial community response information, analyzing through phase coupling analysis... The degree of temporal coordination between the phase of regular fluctuations in methane concentration and the phase of changes in the metabolic activity of methophilic microbial communities yields trend coordination information. Based on the spatial matching information and the trend coordination information, a mutual feedback mechanism is used to identify and analyze the spatial diffusion patterns of methane anomalies, which are modulated by the metabolic activities of methophilic microorganisms at corresponding locations, and the metabolic response trends of methophilic microbial communities, which are driven by the corresponding methane concentration fluctuation rhythms. This identifies stable mutual indicative and reinforcing coupling units between gas migration pathways and methophilic microbial metabolic hotspots, resulting in the permafrost microbial-gas coupling information set.

[0079] Optionally, the feature inference module 302, when simulating the gas migration path in heterogeneous permafrost using an LLM generator and mapping the migration path to mineral types to obtain a mineral feature inference set, specifically performs the following: Based on the coupling units, it analyzes the spatial arrangement orientation and intensity gradient of different coupling units to generate an initial migration trend network tracing back from methanotrophic microbial hotspots to potential gas sources; Based on the initial migration trend network, according to prior knowledge of the geological structure of the heterogeneous permafrost layer, it analyzes the influence of geological interfaces and fracture networks on gas migration trends, selects and strengthens advantageous channels that conform to geological constraints in the initial migration trend network, forming a modified gas migration path; Based on the gas migration path, according to the characteristic gas fingerprint knowledge associated with the formation or alteration of different types of minerals, it analyzes the correspondence between the changing patterns of gas-microbe coupling intensity on the path and the mineral type to obtain the mineral feature inference set.

[0080] Optionally, the feature inference module 302, when analyzing the correspondence between the variation pattern of gas-microbe coupling intensity along the path and the mineral type based on the characteristic gas fingerprint knowledge associated with the formation or alteration of different types of minerals, and obtaining the mineral feature inference set, is specifically used for: analyzing the gas-microbe coupling intensity characterized by the permafrost microbe-gas coupling information set at different spatial nodes along the gas migration path, based on the gas migration path, to obtain a coupling intensity variation sequence distributed along the migration path; based on the coupling intensity variation sequence, analyzing the similarity between the spatial distribution pattern of the coupling intensity variation sequence and the characteristic coupling patterns corresponding to known types of minerals according to the measured gas-microbe coupling pattern library of historically explored mineral samples, to obtain the pattern matching degree; based on the pattern matching degree, analyzing the correlation between gas accumulation trend and path geometric features, inferring possible mineral types and spatial distribution characteristics, and obtaining the mineral feature inference set.

[0081] Optionally, the collaborative prediction module 303, when locating methane leakage and microbial hotspot areas through a discriminator, generating grid sampling-fixed-point deep mining collaborative decisions, and outputting a multimodal mineral prediction agent collaborative log that minimizes intrusion operations, is specifically used for: analyzing the spatial distribution of the pattern matching degree based on the possible mineral types and the spatial distribution characteristics, identifying continuous areas where the pattern matching degree is higher than a set threshold, and obtaining methane leakage areas and microbial hotspot areas as priority exploration targets; based on the methane leakage areas and the microbial hotspot areas, analyzing the requirements for different sampling methods to achieve efficient coverage and core verification, generating a collaborative operation scheme of implementing regular grid sampling outside the hotspot area to determine abnormal boundaries and implementing encrypted fixed-point deep mining inside the hotspot area to verify the existence of minerals, and obtaining grid sampling-fixed-point deep mining collaborative decisions; based on the grid sampling-fixed-point deep mining collaborative decisions, analyzing the geographical environment and permafrost ecological sensitivity of the methane leakage and the microbial hotspot areas, adjusting the operation path to avoid ecologically vulnerable points, and performing structured recording, and outputting the multimodal mineral prediction agent collaborative log.

[0082] Optionally, the collaborative prediction module 303, when analyzing the requirements for different sampling methods to achieve efficient coverage and core verification, is specifically used for: analyzing the spatial distribution information and aggregation pattern of signal points based on the methane leakage area and the microbial hotspot area to obtain spatial distribution pattern information of the target area; based on the spatial distribution pattern information, analyzing the peripheral area where methane leakage and microbial hotspot signals are discretely distributed but point to an undefined range to obtain grid sampling requirement information; based on the spatial distribution pattern information, analyzing the internal area where methane leakage and microbial hotspot signals are highly converged and the convergence position coincides with the target point position indicated by the mineral feature inference set to obtain fixed-point deep mining requirement information; and based on the grid sampling requirement information and the fixed-point deep mining requirement information, analyzing the sequential relationship in operational logic and the complementary relationship in space between the two types of requirements to obtain the requirements for different sampling methods used to generate the collaborative operation scheme.

[0083] The system in this embodiment can be used to execute the methods of any of the above embodiments, and its implementation principle and technical effect are similar, so they will not be described again here.

Claims

1. A multimodal mineral prediction agent cooperative method based on LLM dynamic perception, characterized in that, include: Obtain surface multimodal data of permafrost mineral deposits, and based on the surface multimodal data, analyze the interaction between the surface methanophilic bacterial community structure and permafrost respiratory gases in the permafrost layer above the mineral deposits to obtain a permafrost microbial-gas coupling information set. Based on the frozen soil microbial-gas coupling information set, the gas migration path in heterogeneous frozen soil is simulated by an LLM generator, and the migration path is associated with mineral types to obtain a mineral feature inference set. Based on the mineral feature inference set, the discriminator locates methane leakage and microbial hotspots, generates grid sampling-fixed-point deep mining collaborative decision-making, and outputs a multimodal mineral prediction agent collaborative log that minimizes intrusion operations. Based on the aforementioned surface multimodal data, the interaction between the surface methanophilic bacterial community structure and permafrost respiratory gases in the permafrost layer above the mineral deposits is analyzed to obtain a permafrost microbial-gas coupling information set, including: The surface multimodal data includes surface methane concentration data and methanotrophic microbial distribution data; Based on the surface methane concentration data, the spatial diffusion pattern and temporal fluctuation rhythm of surface methane are analyzed to identify the methane anomaly characteristics with continuous evolution, and to obtain dynamic anomaly information of surface methane. Based on the distribution data of the methophile microorganisms, the synergistic changes in the abundance and metabolic potential of the methophile microorganism population were analyzed, and the characteristics of methophile microbial communities that are sensitive to gas changes and have active metabolic functions were identified, thus obtaining the response information of the surface methophile bacterial community. Based on the dynamic anomaly information of methane on the surface, combined with the response information of the surface methophilic bacterial community, the bidirectional effect of methane anomaly and methophilic microbial response in terms of spatial location and change trend is analyzed to obtain the permafrost microbial-gas coupling information set. The analysis of the bidirectional interaction between methane anomalies and methanotrophic microbial responses in terms of spatial location and variation trends yielded the permafrost microbial-gas coupling information set, including: Based on the methane anomaly characteristics and the methophile microbial community characteristics, the spatial matching information is obtained by analyzing the degree of overlap between the methane anomaly region with stable source orientation and the methophile microbial hotspot region with prominent metabolic maintenance function through spatial overlay. Based on the temporal rhythm information of methane fluctuations, combined with the response information of the surface methophilic bacterial community, phase coupling analysis was used to analyze the degree of coordination between the phase of the regular fluctuation of methane concentration and the phase of the change in the metabolic activity of the methophilic microbial community in the time trend, and to obtain the trend coordination information. Based on the spatial matching information and the trend synergy information, the spatial diffusion pattern of methane anomalies is identified and analyzed through a mutual feedback mechanism. The analysis shows that the metabolic activity of methophilic microorganisms at the corresponding locations is modulated by the spatial diffusion pattern of methane anomalies, and the metabolic response trend of methophilic microbial communities is driven by the corresponding methane concentration fluctuation rhythm. This identifies a stable coupling unit between gas migration paths and methophilic microbial metabolic hotspots that mutually indicate and enhance each other, thus obtaining the permafrost microorganism-gas coupling information set. The process involves simulating the gas migration path in heterogeneous frozen soil using an LLM generator, and then mapping these migration paths to mineral types to obtain a mineral characteristic inference set, including: Based on the aforementioned coupling units, the spatial arrangement orientation and intensity gradient of different coupling units are analyzed to generate an initial migration trend network that traces back from the methanotrophic microbial hotspot to the potential gas source. Based on the initial migration trend network, and according to prior knowledge of the geological structure of heterogeneous permafrost, the influence of geological interfaces and fracture networks on gas migration trends is analyzed. In the initial migration trend network, advantageous channels that conform to geological constraints are selected and strengthened to form a modified gas migration path. Based on the gas migration path, and according to the characteristic gas fingerprint knowledge associated with the formation or alteration of different types of minerals, the correlation between the changing pattern of gas-microbe coupling intensity along the path and the mineral type is analyzed to obtain the mineral feature inference set.

2. The method according to claim 1, characterized in that, The analysis of the spatial diffusion patterns and temporal fluctuations of methane on the Earth's surface identifies continuously evolving methane anomaly characteristics, yielding dynamic anomaly information on surface methane, including: Based on the surface methane concentration data, spatial trend surface analysis is used to analyze the gradual and abrupt changes in methane concentration in continuous geographic space, and to obtain information on the spatial network of methane diffusion. Based on the surface methane concentration data, time series decomposition and sliding window statistics are used to analyze the rate of change characteristics of methane concentration in multiple continuous time series, and to obtain methane fluctuation time rhythm information. Based on the spatial diffusion network information of methane and the temporal rhythm information of methane fluctuations, the coupling relationship between spatial diffusion trends and temporal fluctuation patterns in the same geographical location is analyzed. Regions with stable source orientation in space and regular enhancement or weakening characteristics in time are identified, thereby obtaining the dynamic anomaly information of surface methane.

3. The method according to claim 2, characterized in that, The analysis of the synergistic changes in the abundance and metabolic potential of methophilic microbial communities identified the characteristics of methophilic microbial communities that are sensitive to gas changes and have active metabolic functions, thus obtaining information on the response of surface methophilic bacterial communities, including: Based on the distribution data of the methophile microorganisms, the overlap pattern of spatial distribution of different methophile microbial populations and the abundance relationship of methane oxidation functional genes carried by them were analyzed by functional gene abundance statistics to obtain information on the synergistic combination of microbial functions. Based on the aforementioned information on the synergistic combination of microbial functions, node screening was used to analyze the contribution of different methanophilic microbial populations to maintaining the overall methane oxidation metabolism function of the community, thereby obtaining information on the contribution of microbial populations. Based on the microbial population contribution information, dynamic simulation of metabolic pathways was used to analyze the correlation between the abundance changes of methophilic microbial populations with prominent buffering and maintenance functions and the changes in the overall methane oxidation metabolic potential of the methophilic microbial community, thereby obtaining the response information of the surface methophilic bacterial community.

4. The method according to claim 3, characterized in that, The method involves analyzing the correlation between the variation patterns of gas-microbe coupling strength along the path and the mineral type, based on the characteristic gas fingerprint knowledge associated with the formation or alteration of different types of minerals, to obtain the mineral feature inference set, including: Based on the gas migration path, the gas-microbe coupling strength characterized by the permafrost microbe-gas coupling information set at different spatial nodes along the path is analyzed to obtain the coupling strength variation sequence distributed along the migration path. Based on the coupling intensity change sequence, and according to the measured gas-microbe coupling pattern library of historical proven mineral samples, the similarity between the spatial distribution pattern of the coupling intensity change sequence and the characteristic coupling patterns corresponding to various known minerals is analyzed to obtain the pattern matching degree. Based on the pattern matching degree, the correlation between gas accumulation patterns and path geometric features is analyzed to infer possible mineral types and spatial distribution characteristics, thus obtaining the mineral feature inference set.

5. The method according to claim 4, characterized in that, The process of locating methane leaks and microbial hotspots using a discriminator, generating a grid sampling-point deep mining collaborative decision, and outputting a multimodal mineral prediction agent collaborative log that minimizes intrusion operations includes: Based on the possible mineral types and spatial distribution characteristics, the spatial distribution of the pattern matching degree is analyzed, and continuous areas with the pattern matching degree higher than a set threshold are identified to obtain methane leakage areas and microbial hotspots as priority exploration targets. Based on the methane leakage area and the microbial hotspot area, the requirements for different sampling methods to achieve efficient coverage and core verification are analyzed, and a collaborative operation plan is generated to implement regular grid sampling outside the hotspot area to determine the abnormal boundary and to implement intensive fixed-point deep mining inside the hotspot area to verify the existence of minerals, resulting in grid sampling-fixed-point deep mining collaborative decision. Based on the grid sampling-fixed-point deep mining collaborative decision-making, the geographical environment and permafrost ecological sensitivity of the methane leakage and the microbial hotspot are analyzed, the operation path is adjusted to avoid ecologically vulnerable points, and structured records are made to output the collaborative log of the multimodal mineral prediction agent.

6. The method according to claim 5, characterized in that, The analysis addresses the need for different sampling methods to achieve efficient coverage and core verification, including: Based on the methane leakage area and the microbial hotspot area, the spatial distribution information and aggregation pattern of the signal points are analyzed to obtain the spatial distribution pattern information of the target area. Based on the spatial distribution pattern information, the analysis shows that methane leakage and microbial hotspot signals are discretely distributed but all point to an undefined peripheral area, thus obtaining grid sampling requirement information. Based on the spatial distribution pattern information, the internal region where methane leakage and microbial hotspot signals show high-intensity convergence, and the convergence location coincides with the target location indicated by the mineral feature inference set, is analyzed to obtain the information on the demand for targeted deep mining. Based on the grid sampling requirement information and the fixed-point deep mining requirement information, the sequential relationship in the operation logic and the complementary relationship in the space of the two types of requirements are analyzed to obtain the requirements for different sampling methods used to generate the collaborative operation scheme.

7. A multimodal mineral resource prediction intelligent agent collaborative system based on LLM dynamic perception, characterized in that, The method applied to any one of claims 1-6 includes: The coupling analysis module is used to acquire surface multimodal data of permafrost mineral deposits. Based on the surface multimodal data, the interaction between the surface methanophilic bacterial community structure and the permafrost respiratory gases in the permafrost layer above the mineral deposits is analyzed to obtain a permafrost microbial-gas coupling information set. The feature inference module is used to simulate the gas migration path in heterogeneous permafrost using an LLM generator based on the permafrost microbial-gas coupling information set, and associate the migration path with mineral types to obtain a mineral feature inference set. The collaborative prediction module is used to locate methane leakage and microbial hotspots based on the mineral feature inference set, generate grid sampling-fixed-point deep mining collaborative decision, and output a multimodal mineral prediction agent collaborative log that minimizes intrusion operations.

Images