A Method and System for AI Fusion Recognition and Analysis of Geological Multimodal Data

Abstract

This invention relates to the field of geological exploration data processing technology, and discloses a method and system for AI fusion, identification, and analysis of multimodal geological data. The method aggregates and structures multimodal data such as physical fields, chemical elements, geological structures, and drilling core samples; constructs and iteratively optimizes a knowledge topology network based on geological semantics to express semantic relationships and spatial constraints between data entities; actively detects high-value information links and contradictions within the knowledge network; generates multiple parallel fusion decision-making plans accordingly, clarifying data coordination rules and feature enhancement directions; introduces external intervention commands to screen and reconstruct the plans, forming a target fusion operation sequence; executes the sequence and synchronously evolves the knowledge network. This method realizes a transformation from static data association to dynamic knowledge guidance, and from fixed algorithms to human-machine collaborative proactive decision-making, improving the reliability of geological target identification and the level of intelligent analysis.

Description

Technical Field

[0001] This invention relates to the field of geological exploration data processing technology, specifically to a method and system for AI fusion, identification, and analysis of geological multimodal data. Background Technology

[0002] In current geological exploration, the analysis of multimodal data, including physical fields, chemical elements, geological structural morphology, and deep drilling cores, mainly relies on the independent interpretation of various data types or simple superposition and statistical correlation based on spatial coordinates. Existing fusion technologies mostly use fixed algorithm models to integrate preprocessed features, and their correlation mechanisms remain at the data level of matching or correlation analysis. This approach cannot systematically construct and utilize the implicit geological semantic connections and spatial constraints behind the data, resulting in a lack of geological interpretive guidance in the fusion process.

[0003] The shortcomings of existing technical solutions lie in the fact that the correlation between multimodal data is static and shallow, making it difficult to express complex geological logics such as "control" and "symbiosis." The analysis path is usually pre-set and singular. The entire process cannot proactively assess the value and contradictions of different evidence chains before analysis, and it also lacks an effective mechanism to incorporate expert judgment and dynamically adjust the fusion strategy during the analysis process, which limits its reliability and applicability in solving complex and uncertain geological problems.

[0004] This invention aims to construct a unified knowledge framework that can dynamically express the relationship between geological semantics and spatial constraints, thereby supporting deep fusion with geological interpretability. Simultaneously, it needs to implement a flexible analysis and decision-making process capable of proactively detecting information paths and supporting external intervention and adjustment. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for AI fusion, identification and analysis of geological multimodal data, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, this invention provides an AI fusion and identification analysis method for geological multimodal data, the method comprising:

[0007] The process of collecting and preliminarily structuring geological multimodal raw data, wherein the geological multimodal raw data includes at least physical field data, chemical element data, geological structural morphology data, and deep drilling core data;

[0008] The process of constructing and iteratively optimizing a knowledge topology network based on geological semantics, wherein the knowledge topology network is used to express the semantic associations and spatial constraints between geological data entities of different modalities;

[0009] An active path detection process is initiated in the knowledge topology network to discover high-value information flow links and potential convergence points of interpretive contradictions;

[0010] Based on the detection results, multiple parallel fusion decision-making plans are generated. Each fusion decision-making plan includes data coordination rules and feature enhancement directions for specific geological targets.

[0011] External intervention commands are introduced to screen and reconstruct the fusion decision-making plan, forming a target fusion operation sequence;

[0012] The target fusion operation sequence is executed, and the knowledge topology network is synchronously evolved based on the state variables generated during the execution process.

[0013] Preferably, the aggregation and preliminary structuring of the geological multimodal raw data is achieved through the following steps:

[0014] The raw data stream is asynchronously retrieved from distributed geological observation nodes and centralized geological databases, and the raw data stream is then formatted and time-stamped.

[0015] Modality-specific analysis was performed on the interpreted and aligned data. For physical field data, the field strength distribution characteristics and anomalous contour morphology were extracted; for chemical element data, the element abundance sequence and isotope ratio combination were extracted; for geological structural morphology data, the interface occurrence parameters and spatial distribution trend were extracted; and for deep drilling core data, the lithological columnar section and micropore structure parameters were extracted.

[0016] Each parsed data unit is assigned a globally unique geological data entity identifier and encapsulated into a structured data package. The structured data package contains an ontology description of the data unit, a quality assessment coefficient, a spatiotemporal coordinate index, and a pointer to the original data source.

[0017] The encapsulated structured data packets are injected into a temporary geological data buffer pool according to their spatiotemporal coordinate indices, awaiting invocation during the construction and iterative optimization process of the knowledge topology network.

[0018] Preferably, the construction and iterative optimization process of the knowledge topology network based on geological semantics is completed in the following way:

[0019] Structured data packets are extracted in batches from the geological data buffer pool according to a spatiotemporal sliding window.

[0020] Based on the ontology description in the structured data packet, each data unit is instantiated as a semantic node in the knowledge topology network. The semantic node has an attribute set, which includes at least modality type, data quality assessment coefficient, geological confidence score, information freshness index and fusionable state label.

[0021] Based on a pre-defined geological knowledge rule base, semantic edges are dynamically established between semantic nodes. The geological knowledge rule base defines the causal relationships, spatial inclusion relationships, time series relationships, and physical property response relationships that may exist between different types of geological data entities. Each semantic edge records the type of relationship, the strength weight of the relationship, and a summary of the geological basis for the relationship.

[0022] After the semantic edges are established, the network self-optimization loop is started: the average path length and clustering coefficient of the entire knowledge topology network are calculated. If the average path length exceeds the threshold or the clustering coefficient is lower than the threshold, the rule base is supplemented and the semantic edges are reconnected until the network structure indicators are stable within the preset range.

[0023] The optimized knowledge topology network, along with all its semantic nodes, attribute sets, and semantic edge information, will be persistently stored in the geological knowledge graph database for use in subsequent active pathfinding processes.

[0024] Preferably, the step of initiating the active path detection process in the knowledge topology network is specifically carried out as follows:

[0025] Load the current version of the knowledge topology network from the geological knowledge graph repository;

[0026] In the knowledge topology network, multiple detection start points and detection end points are preset. The detection start points correspond to semantic nodes with dense original observation data, and the detection end points correspond to semantic nodes corresponding to advanced geological interpretation or resource evaluation conclusions.

[0027] Concurrent exploration threads are started, with each thread starting from a different detection point and exploring along semantic edges at a limited depth. The exploration decision is based on the weighted sum of the strength weight of the semantic edge and the geological confidence score of the endpoint semantic node. During the exploration process, the sequence of semantic nodes traversed by the path, the sequence of semantic edges, and the overall confidence score of the path are recorded in real time.

[0028] After all exploration threads terminate, collect all valid paths that reach the exploration endpoint; sort the valid paths in descending order according to the cumulative confidence value of the paths, and mark the top-ranked valid paths as high-value information flow links.

[0029] Cross-comparison analysis was performed on all valid paths to identify semantic nodes that were frequently passed by multiple paths but whose geological confidence scores in their attribute sets were significantly lower than the network mean. These semantic nodes were marked as potential points of convergence of explanatory contradictions.

[0030] The detailed structure of the high-value information flow links output by the detection process and the list of nodes where potential contradictions in interpretation converge are packaged into a path detection report and passed to the subsequent fusion decision-making plan generation process.

[0031] Preferably, the process of generating multiple parallel fusion decision plans based on the detection results specifically includes the following steps:

[0032] Analyze the received path detection reports to separate the set of high-value information flow links and the set of potential points of convergence of contradictory interpretations;

[0033] For each high-value information flow link, a feature enhancement plan is generated. The core of the feature enhancement plan is to improve the information freshness index of key semantic nodes in the high-value information flow link and enhance the strength weight of weak semantic edges in the link. The implementation method is to schedule additional data collection resources to conduct supplementary observations of the geological areas involved in the high-value information flow link and inject new structured data packets into the knowledge topology network, thereby updating the attribute set of relevant nodes.

[0034] For each potential convergence point of interpretation contradictions, a set of data coordination plans is generated. The core of the data coordination plan is to perform multi-source verification and contradiction resolution on the geological data entity corresponding to the convergence point. The implementation method is to backtrack all input semantic edges of the semantic node, compare the consistency of data units from different sources on the semantic node attributes, and when the inconsistency exceeds the tolerance, an arbitration calculation based on geological and physical laws is initiated, and the arbitration result is used to overwrite the original attribute value, while the arbitration log is recorded.

[0035] All generated feature enhancement plans and data coordination plans are categorized and packaged according to their operational targets and estimated resource consumption to form multiple independent fusion decision-making plan files. Each file contains a list of executable operation instructions and a description of expected network state changes.

[0036] Preferably, the introduction of external intervention instructions to screen and reconstruct the fusion decision plan to form a target fusion operation sequence is achieved through the following mechanism:

[0037] A visual decision-making interface is provided, which presents the list of operation instructions and the expected changes in network status from multiple integrated decision-making plans to geological experts in an interactive manner.

[0038] Through the visual decision-making interface, external intervention instructions input by geological experts are received. These external intervention instructions include marking the adoption of a specific plan, adjusting the execution order of operation instructions, modifying parameters in operation instructions, and cross-combining instructions from different plans.

[0039] Establish a contingency plan execution engine, which receives external intervention instructions and original fusion decision contingency plan documents;

[0040] The contingency plan execution engine first filters out contingency plan files that are not selected based on the adoption flag;

[0041] Next, the contingency plan execution engine reorders and updates the operation instructions and parameters in the retained contingency plan file according to the adjusted execution order and modified parameters;

[0042] Then, the pre-plan execution engine merges the operation instructions from different pre-plan files into a unified execution list according to their logical dependencies, based on the cross-combination instructions;

[0043] Finally, the pre-plan execution engine performs logical verification and resource conflict checks on this unified execution list. After verification and checks, it solidifies it into the final target fusion operation sequence, which is a set of atomic operations arranged strictly according to time sequence and conditions.

[0044] Preferably, the execution of the target fusion operation sequence, and the synchronous evolution of the knowledge topology network based on the state variables generated during the execution process, specifically includes the following operations:

[0045] The contingency execution engine reads each atomic operation in the target fusion operation sequence sequentially.

[0046] For atomic operations that are data acquisition type, the engine calls the resource scheduling interface, dispatches the observation equipment to perform supplementary observation tasks, and waits for the return of new structured data packets. Then, it triggers the aggregation and preliminary structuring process of the geological multimodal raw data, and injects the newly generated structured data packets into the geological data buffer pool.

[0047] For atomic operations that are attribute arbitration type, the engine loads the current state of the relevant semantic nodes and their associated edges from the geological knowledge graph library, executes the arbitration calculation logic, directly updates the attribute set of the corresponding semantic nodes in the knowledge topology network with the calculation results, and writes the changes to persistent storage.

[0048] After each atomic operation is completed, the engine captures the operation execution status code and related intermediate variables, including the amount of data freshness improvement, the magnitude of attribute changes, and the amount of resource consumption.

[0049] Once all atomic operations in the target fusion operation sequence have been executed, the engine summarizes all operation execution status codes and intermediate variables to generate a fusion execution summary.

[0050] Based on the fusion execution summary, especially the information regarding the magnitude of attribute changes and the amount of data freshness improvement, a local reconstruction of the knowledge topology network is automatically triggered: the geological confidence scores and fusionable state labels of the affected semantic nodes are recalculated, and the strength weights of the semantic edges connected to them are re-evaluated.

[0051] The modified parts of the partially reconstructed knowledge topology network are synchronously updated to the geological knowledge graph database, completing a single evolution of the knowledge topology network.

[0052] Preferably, the construction and dynamic supplementation of the geological knowledge rule base are completed through the following independent steps:

[0053] In the initial stage, a series of "premise-conclusion" production rules were manually extracted and formalized from authoritative geology textbooks, published high-quality geological map descriptions, and geological stratification and interpretation specifications in industry standards, forming the initial core of the rule base;

[0054] In the construction and self-optimization loop of the knowledge topology network, when the rule base is supplemented, the rule mining routine is started: the rule mining routine analyzes the frequently co-occurring semantic node patterns and semantic edge patterns in the current network structure, and automatically abstracts the co-occurrence patterns of geological data entities that meet the support and confidence thresholds into new candidate production rules.

[0055] A rule review queue is established, in which all candidate production rules generated by automatic mining and new rules proposed by external geological experts through the interactive interface are entered into the rule review queue.

[0056] Regularly, multiple geological experts conduct back-to-back reviews of the rules in the rule review queue. Only rules that are approved by the majority of experts will be formally incorporated into the geological knowledge rule base.

[0057] Each formal rule records its source, confidence level, effective time, and last invocation time. Automatically mined rules are assigned an initial confidence level lower than that of manually generated and authoritative rules, and their confidence level gradually increases as they are successfully verified more often in subsequent network construction.

[0058] Preferably, after a single evolution of the knowledge topology network is completed, a network health assessment and feedback adjustment phase is set up, which includes the following steps:

[0059] The global health index of the evolved knowledge topology network is calculated. The global health index is a weighted sum of multiple sub-indicators, including: the average geological confidence score of semantic nodes, the average strength weight of semantic edges, the number of network connected subgraphs, and the average length of high-value information flow links.

[0060] The calculated global health index is compared with the historical health index sequence to determine whether the current evolution has led to an improvement, a decrease, or a basically unchanged health level.

[0061] If the improvement is due to health status, the pattern of atomic operations in the target fusion operation sequence that led to this improvement will be strengthened, and its operation parameters and execution order will be used as successful experience templates and stored in the contingency plan template library for priority reference when generating fusion decision contingency plans in the future.

[0062] If the health level declines, back-analyze the target fusion operation sequence to identify the atomic operations that cause a significant drop in the geological confidence score of key semantic nodes or breakage of important semantic edges. Add these operations to the list of cautious operations and apply higher trigger thresholds or avoidance suggestions when generating fusion decision plans in the future.

[0063] The health assessment results, successful experience templates, and cautionary operation checklists are used as feedback information streams and input into the process of generating multiple parallel fusion decision plans based on the detection results, so as to influence the generation strategy of subsequent plans.

[0064] Preferably, the present invention also includes a geological multimodal data AI fusion identification and analysis system, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the steps of the geological multimodal data AI fusion identification and analysis method described above.

[0065] Compared with the prior art, the beneficial effects of the present invention are:

[0066] A knowledge topology network integrating geological semantics and spatial constraints was constructed, and its iterative optimization mechanism was designed. Geological data entities of different modalities, along with their semantic relationships and spatial constraints, are encoded in an integrated manner to form a dynamic knowledge structure with geological interpretability. This network can explicitly transform implicit geological laws into computable and inferable relational graphs, providing a domain-knowledge-rich contextual framework for multimodal data fusion. The network structure can evolve synchronously based on the state feedback generated by the fusion analysis, enabling the knowledge representation to continuously improve and correct itself along with the analysis process. This enhances the system's understanding and adaptability to complex and uncertain geological scenarios, placing the fusion process on a continuously evolving geological cognitive foundation.

[0067] Active path detection was initiated within the knowledge topology network, and a process was established to generate multiple parallel, interventionist fusion decision-making plans. Before executing specific fusion tasks, the system proactively probes and evaluates high-value information flow links connecting different pieces of evidence within the knowledge network, and identifies potential inconsistencies in reasoning, thus shifting from "passively executing computations" to "actively planning analytical strategies." Based on the detection results, the system generates multiple fusion decision-making plans containing specific data coordination rules and feature enhancement directions, providing various analytical perspectives and solution paths for the same geological problem. By introducing external intervention commands to filter and reconstruct the plans, human-machine collaborative decision-making is achieved, enabling experts' domain experience to directly and flexibly guide and correct the automatic analysis process, forming a dynamic and controllable closed-loop analysis that integrates domain intelligence and machine computing power. Attached Figure Description

[0068] Figure 1 This is a schematic diagram illustrating the working principle of the geological multimodal data AI fusion identification and analysis method described in this invention.

[0069] Figure 2 A flowchart for the aggregation and preliminary structured processing of raw geological multimodal data;

[0070] Figure 3 This is a flowchart illustrating the construction and iterative optimization of a knowledge topology network based on geological semantics.

[0071] Figure 4 A comparison diagram of the state changes of the knowledge topology network for geological multimodal data;

[0072] Figure 5 This is a quality and coverage assessment map for geological multimodal data. Detailed Implementation

[0073] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0074] Please see Figure 1This invention provides an AI-based fusion and identification analysis method for geological multimodal data. The method includes: a process for aggregating and initially structuring raw geological multimodal data, responsible for integrating multimodal data such as physical field data, chemical element data, geological structural morphology data, and deep drilling core data, and transforming them into data packets with unified identifiers and structures; a process for constructing and iteratively optimizing a knowledge topology network based on geological semantics, using the processed structured data to construct a network model expressing the semantic and spatial relationships between geological data entities, and stabilizing the network structure through a self-optimization mechanism; an active path detection process initiated in the knowledge topology network, which automatically discovers high-value information links connecting observational data and geological interpretation through network exploration algorithms, and locates nodes that may contain contradictions; multiple parallel fusion decision-making plans are generated based on the detection results, forming a set of executable plans aimed at strengthening features or coordinating data for the discovered links and contradictions; external intervention commands are introduced to screen and reconstruct the fusion decision-making plans, and geological experts evaluate and adjust the plans through an interactive interface, ultimately forming a logically self-consistent target fusion operation sequence. The target fusion operation sequence is executed, and the knowledge topology network is synchronously evolved based on the state variables generated during the execution process. This process executes operation instructions one by one, updates the network state, and realizes the dynamic evolution of the knowledge network based on the execution feedback.

[0075] In one embodiment of the present invention, see [reference] Figure 2 The system asynchronously retrieves raw data streams from distributed geological observation nodes and centralized geological databases, and performs format decoding and time-stamp alignment on these raw data streams. Modality-specific analysis is then performed on the decoded and aligned data. For physical field data, field strength distribution characteristics and anomalous contour morphology are extracted; for chemical element data, elemental abundance sequences and isotope ratio combinations are extracted; for geological structural morphology data, interface occurrence parameters and spatial distribution trends are extracted; and for deep drilling core data, lithological columnar sections and micropore structure parameters are extracted. Each parsed data unit is assigned a globally unique geological data entity identifier and encapsulated into a structured data package. This structured data package contains an ontology description, quality assessment coefficients, spatiotemporal coordinate index, and a pointer to the original data source. Multiple encapsulated structured data packages are then injected into a temporary geological data buffer pool according to their spatiotemporal coordinate indices.

[0076] In practice, the aggregation and preliminary structuring of geological multimodal raw data is achieved through a clearly defined sequence of steps. The system asynchronously retrieves raw data streams from distributed geological observation nodes and a centralized geological database. The distributed observation nodes include time-series data files collected in real-time by geophysical equipment such as gravimeters, magnetometers, and transient electromagnetic instruments deployed in the exploration area. The centralized geological database includes a relational database and document library storing historical exploration reports, digital geological maps, and laboratory test results. The format interpretation and time-stamp alignment operations performed on the asynchronously retrieved raw data streams involve parsing the proprietary binary formats of instruments from different manufacturers, converting coordinate systems to a unified geodetic datum, and correcting the time stamps of all data to the same standard time system.

[0077] In some embodiments, modality-specific analysis is performed on the interpreted and aligned data, and different feature extraction modules are invoked. For physical field data, the feature extraction module reads the gridded field value data and extracts its field strength distribution characteristics and anomalous contour morphology through gradient calculation and contour tracking algorithms. For chemical element data, the feature extraction module analyzes spectral or mass spectrometry data sequences, extracts its elemental abundance sequence through peak identification and area integration calculation, and extracts isotope ratio combinations through isotope peak intensity ratio calculation. For geological structural morphology data, the feature extraction module processes 3D seismic interpretation layers or geological mapping vector data, extracts its interface attitude parameters and spatial distribution trends through surface fitting and trend surface analysis. For deep drilling core data, the feature extraction module analyzes core scanning images and porosity experimental data, extracts its lithological columnar section and micropore structure parameters through image layer recognition and pore network modeling.

[0078] Optionally, each parsed data unit is assigned a globally unique geological data entity identifier, generated using a hash algorithm based on spatiotemporal stamps and device encoding. The process of encapsulating this identifier into a structured data packet integrates and encapsulates the data unit's ontology description, quality assessment coefficients, spatiotemporal coordinate index, and original data source pointer. The data unit's ontology description uses a resource description framework triple, declaring the modal category to which the data unit belongs, the geological object it represents, and its basic attributes. The quality assessment coefficient is calculated using an evaluation function that comprehensively considers factors such as data acquisition accuracy, signal-to-noise ratio, and coverage integrity.

[0079] ;

[0080] in: Represents the quality assessment coefficient. This represents a score indicating the accuracy of data collection. C represents the signal-to-noise ratio, and C represents the spatial or temporal coverage integrity score. , , These are the preset weighting coefficients for each item. The spatiotemporal coordinate index establishes the mapping relationship between data units and specific geographical locations and time periods. The original data tracing pointer, in the form of a Uniform Resource Identifier, points to the storage location and access path of the original data file.

[0081] In practice, multiple encapsulated structured data packets are injected into a temporary geological data buffer pool according to their spatiotemporal coordinate indexes. This buffer pool is an in-memory database with spatiotemporal four-dimensional indexing capabilities, supporting rapid retrieval and batch reading of the injected structured data packets based on geographical scope and time intervals, awaiting invocation during the construction and iterative optimization process of the knowledge topology network. When structured data packets are injected into the geological data buffer pool, the system verifies the format validity of the spatiotemporal coordinate indexes and their consistency with the data ontology description. Structured data packets that fail verification trigger alarms and are transferred to an exception handling queue for manual review.

[0082] In one embodiment of the present invention, see [reference] Figure 3 In practical implementation, the construction and iterative optimization process of the knowledge topology network based on geological semantics is completed in the following way: The system extracts structured data packets in batches from the geological data buffer pool according to the spatiotemporal sliding window. The size of the spatiotemporal sliding window is preset according to the size of the exploration area and the data acquisition density. For example, a window may cover a plane area of ​​ten square kilometers and a time span of one geological year. The extracted structured data packets are sent to the network construction module. Based on the ontology description in the structured data packets, the network construction module instantiates each data unit as a semantic node in the knowledge topology network. The semantic node has an attribute set, which includes at least modality type, data quality assessment coefficient, geological confidence score, information freshness index, and fusionable state mark. The modality type is directly derived from the ontology description of the structured data packets. The data quality assessment coefficient is directly inherited from the structured data packets. The geological confidence score is initialized based on the authority of the data source and historical verification records. The information freshness index is calculated based on the difference between the data acquisition time and the current time. The fusionable state mark is initially set to "to be evaluated" by default.

[0083] In some embodiments, semantic edges are dynamically established between semantic nodes based on a preset geological knowledge rule base. The geological knowledge rule base is stored in the form of a graph database and defines the causal relationships, spatial inclusion relationships, time series relationships, and physical property response relationships that may exist between different types of geological data entities. The network construction module traverses the semantic nodes extracted in the current batch, uses the modal type and attributes of the semantic nodes as conditions, and matches the rules that meet the conditions in the geological knowledge rule base. Each activated rule guides the creation of a directed or undirected semantic edge between two or more semantic nodes. Each semantic edge records the type of relationship, the strength weight of the relationship, and a summary of the geological basis for the relationship. The strength weight of the relationship is initialized based on the product of the confidence of the rule itself and the data quality evaluation coefficient of the semantic node.

[0084] Optionally, after semantic edges are established, the system initiates a network self-optimization loop. This loop calculates the average path length and clustering coefficient of the entire knowledge topology network. The average path length is the average number of semantic edges traversed by the shortest path between all possible pairs of semantic nodes. The clustering coefficient is an indicator that measures the degree of semantic edge connection between a semantic node's neighboring nodes. The network self-optimization loop compares the calculated average path length and clustering coefficient with preset thresholds. If the average path length exceeds the threshold or the clustering coefficient is below the threshold, it triggers the addition of rules to the rule base and the reconnection of semantic edges. The reconnection of semantic edges includes removing some semantic edges with strength weights below the average and attempting to establish new semantic edges between semantic nodes that were not originally directly connected based on the supplemented geological knowledge rule base. This process is repeated until the network structure indicators stabilize within a preset range.

[0085] ;

[0086] in: Represents the deviation of the network structure. This represents the average path length of the current network. and The predefined reasonable interval boundary representing the average path length The clustering coefficients represent the current network. and This represents the pre-defined reasonable interval boundary for the clustering coefficients. When... When the value is greater than zero, the system determines that an optimization operation needs to be triggered.

[0087] It is understandable that the optimized knowledge topology network, along with all its semantic nodes, attribute sets, and semantic edge information, is persistently stored in the geological knowledge graph repository. The geological knowledge graph repository adopts a storage system that supports graph data models and version management. During persistent storage, the system generates a globally unique version identifier for the current version of the knowledge topology network and associates the version identifier with the spatiotemporal sliding window parameters used during network construction and the version number of the geological knowledge rule base used, so that subsequent active path detection processes can load and use it by specifying the version identifier.

[0088] In practice, the construction and dynamic supplementation of the geological knowledge rule base are completed through the following independent steps. In the initial stage, geological experts manually extract and formalize a series of "premise-conclusion" production rules from authoritative geological textbooks, published high-quality geological map descriptions, and geological stratification and interpretation specifications in industry standards, which constitute the initial core of the rule base. Each rule is defined in a machine-readable scripting language, which clarifies the semantic node modality type and attribute constraints corresponding to its preconditions, as well as the semantic relationship type and weight calculation logic corresponding to its conclusion.

[0089] In some embodiments, when the rule base is supplemented during the construction and self-optimization loop of the knowledge topology network, the system starts a rule mining routine. The rule mining routine analyzes the frequently co-occurring semantic node patterns and semantic edge patterns in the current network structure. The rule mining routine uses a graph pattern mining algorithm to traverse all subgraph structures of the knowledge topology network, counts the frequency of specific types of semantic edges between semantic nodes of specific modal combinations, and automatically abstracts the co-occurrence patterns of geological data entities that meet the preset support and confidence thresholds into new candidate production rules.

[0090] Optionally, the system establishes a rule review queue. All candidate production rules automatically mined by rule mining routines and new rules proposed by external geological experts through the interactive interface enter the rule review queue. Each record in the rule review queue contains the rule's logical expression, information about the mining source or proposer, and preliminary statistical support and confidence data. The status of the rule review queue is visible to all authorized geological expert users. It is understood that the system periodically has multiple geological experts conduct back-to-back reviews of the rules in the rule review queue. The review process is conducted through an independent online interface, where geological experts judge the geological rationality and universality of each rule and provide acceptance or rejection opinions. Only rules that receive majority expert approval are formally incorporated into the geological knowledge rule base. The incorporation operation includes storing the rule script in the core library and updating the index. Each formal rule records its source, confidence level, effective time, and last call time. The system assigns an initial confidence level to automatically mined rules that is lower than that of manually mined rules and authoritative rules. The confidence level of automatically mined rules gradually increases as they are successfully verified more often in subsequent network construction. Each successful verification means that after the rule is activated and used to establish semantic edges, the established semantic edges are not removed in the network self-optimization loop.

[0091] In one embodiment of the present invention, the active path detection process initiating in the knowledge topology network is specifically carried out as follows: the path detection module loads the current version of the knowledge topology network from the geological knowledge graph library. The loading operation requires specifying the version identifier of the knowledge topology network to ensure that the detection process is based on a consistent data snapshot. In the loaded knowledge topology network, multiple detection start points and detection end points are preset. The detection start points correspond to semantic nodes with dense original observation data, such as semantic nodes instantiated from high-density gravity measurement data and dense chemical sampling data. The detection end points correspond to semantic nodes corresponding to advanced geological interpretation or resource evaluation conclusions, such as interpretation conclusion nodes such as "favorable mineral-bearing area" and "fault existence". Concurrent exploration threads are initiated to perform path search. Each thread starts from a different exploration point and explores along semantic edges at a limited depth. The depth limit is used to prevent the search from going on indefinitely in an excessively large knowledge topology network. The exploration decision is based on the weighted sum of the strength weight of the semantic edge and the geological confidence score of the endpoint semantic node. During the exploration process, each thread records the sequence of semantic nodes traversed by the path, the sequence of semantic edges, and the cumulative overall confidence value of the path in real time. The cumulative overall confidence value of the path is calculated by accumulating the strength weight of all semantic edges on the path and the geological confidence scores of the semantic nodes traversed according to a predetermined algorithm.

[0092] In some embodiments, after all exploration threads terminate, the path detection module collects all valid paths leading to the detection endpoint. A valid path refers to an ordered sequence of semantic nodes and semantic edges that start from the detection starting point, are within a depth limit, and ultimately reach any preset detection endpoint. The path detection module sorts the valid paths in descending order based on the cumulative overall confidence value of the paths, marking the top-ranked valid paths as high-value information flow links. The preset proportion is set by the user according to task requirements. A cross-analysis is performed on all valid paths to identify semantic nodes that are frequently traversed by multiple paths but whose geological confidence scores in their attribute sets are significantly lower than the network average. The criterion for "significantly lower" is a threshold set to be a certain percentage below the network average. These semantic nodes are marked as potential points of convergence for interpretive contradictions. The path detection module encapsulates the detailed structure of the high-value information flow links output by the detection process and the node list of potential explanatory contradictions into a path detection report. The path detection report adopts a structured data format, which records in detail the node and edge sequence and their confidence level of each high-value information flow link, as well as the node identifier, the number of paths passing through it, and the current geological confidence score of each potential explanatory contradiction convergence point, and passes it to the subsequent fusion decision-making plan generation process.

[0093] Optionally, the process of generating multiple parallel fusion decision-making plans based on the detection results includes the following steps: The plan generation engine parses the received path detection report, separating a set of high-value information flow links and a set of potential interpretive contradiction convergence points. For each high-value information flow link, the plan generation engine generates a feature enhancement plan. The core of the feature enhancement plan is to improve the information freshness index of key semantic nodes in the high-value information flow link and enhance the strength weight of weak semantic edges in the link. Key semantic nodes are determined by analyzing the centrality index of all nodes in the link, and weak semantic edges refer to edges in the link whose strength weight is lower than the average weight of the link by a certain proportion. The feature enhancement plan is implemented by scheduling additional data acquisition resources to conduct supplementary observations of the geological areas involved in the high-value information flow links. The supplementary observation instructions specify the data modalities, spatial locations, and accuracy requirements to be collected, and inject new structured data packets into the knowledge topology network to update the attribute sets of relevant nodes.

[0094] Understandably, for each potential convergence point of interpretive contradictions, the contingency plan generation engine generates a data coordination contingency plan. The core operation of the data coordination contingency plan is to perform multi-source verification and contradiction resolution on the geological data entity corresponding to the convergence point. The implementation of the data coordination contingency plan involves tracing back all input semantic edges of the semantic node, where each input semantic edge points to an edge of the semantic node of the potential convergence point of interpretive contradictions. The consistency of semantic node attributes between data units from different sources is compared; for example, comparing density values ​​obtained through gravity inversion with density values ​​obtained through core measurements. When inconsistencies exceed the tolerance, the system initiates arbitration calculations based on geophysical laws. The arbitration calculation model is selected according to the specific data modality and geological background, such as using weighted average, error propagation theory, or a deterministic geological model for calculation. The arbitration result overwrites the original attribute value, and an arbitration log is recorded. The contingency plan generation engine categorizes and packages all generated feature enhancement contingency plans and data coordination contingency plans according to their operational objects and estimated resource consumption, forming multiple independent fusion decision contingency plan files. Each fusion decision contingency plan file contains a list of executable operation instructions and a description of expected network state changes. The list of operation instructions is a script that can be parsed and executed by the contingency plan execution engine. The description of expected network state changes includes the expected increase in the geological confidence score of the target semantic node and the expected increment of the target semantic edge strength weight.

[0095] In one embodiment of the present invention, external intervention instructions are introduced to filter and reconstruct the fusion decision-making plan, forming a target fusion operation sequence, which is achieved through the following mechanism: A visual decision-making interface is provided, presenting the list of operation instructions and expected network state changes from multiple fusion decision-making plan files to geological experts in an interactive form. Through the visual decision-making interface, external intervention instructions input by geological experts are received. These external intervention instructions include marking the adoption of a specific plan, adjusting the execution order of operation instructions, modifying parameters in operation instructions, and cross-combining instructions from different plans. A plan execution engine is established, which receives the external intervention instructions and the original fusion decision-making plan files. The plan execution engine first filters out unselected plan files based on the adoption mark. Then, based on the adjusted execution order and modified parameters, the plan execution engine reorders and updates the parameters of the remaining plan files. Finally, based on the cross-combination instructions, the plan execution engine merges the operation instructions from different plan files into a unified execution list according to logical dependencies. Finally, the pre-plan execution engine performs logical verification and resource conflict checks on this unified execution list. After verification and checks, it solidifies it into the final target fusion operation sequence, which is a set of atomic operations arranged strictly according to time sequence and conditions.

[0096] In practical implementation, the introduction of external intervention instructions to screen and reconstruct the fusion decision-making plan and form the target fusion operation sequence is achieved through the following mechanism: The system provides a visual decision-making interface, which presents the list of operation instructions and the expected network state changes from multiple fusion decision-making plan files to geological experts in a graphical manner; the list of operation instructions is displayed in the form of draggable and editable process nodes, each node is labeled with operation type, target geological entity identifier, and estimated resource consumption; the expected network state changes are displayed in the form of comparison charts, showing the expected changes in the geological confidence score, information freshness index, and semantic edge strength weight of key semantic nodes before and after the execution of the plan.

[0097] In some embodiments, external intervention instructions input by geological experts are received through a visual decision-making interface. These external intervention instructions include marking the adoption of a specific plan, adjusting the execution order of operation instructions, modifying parameters in operation instructions, and cross-combining instructions from different plans. Geological experts mark the adoption of a plan by checking boxes, adjust the execution order of operation instructions by dragging process nodes, modify parameters in operation instructions by pop-up forms (e.g., adjusting the accuracy level of data acquisition or modifying the selection of arbitration calculation models), and cross-combining nodes from different plans to create new instruction sequences.

[0098] Optionally, a contingency plan execution engine is established. This engine receives external intervention instructions and the original fusion decision contingency plan files. First, the engine filters out unselected contingency plan files based on adoption flags. These unselected files are archived but not included in subsequent processing. Next, the engine reorders and updates the parameters of the remaining contingency plan files according to the adjusted execution order and modified parameters. This reordering process ensures that data dependencies between operational instructions are not disrupted; for example, attribute arbitration operations can only be executed after their dependent data acquisition operations are completed. Finally, the engine merges operational instructions from different contingency plan files into a unified execution list based on logical dependencies, determined by analyzing the data entities processed by the operational instructions and their read / write states.

[0099] In practice, when geological experts evaluate multiple integrated decision-making plans on the visualized decision-making interface, the system can provide a comparison table of key parameters of the plans to assist in decision-making. This table directly presents the core information of different plans, as shown in Table 1.

[0100] Table 1: Comparison of Key Parameters of Integrated Decision-Making Plans

[0101]

[0102] In some embodiments, the execution engine generates an execution plan summary for final confirmation before finalizing the target fusion operation sequence. The execution plan summary lists all atomic operations to be performed, the total estimated time, the total resource consumption, and the expected changes in the overall network state. After confirmation by geological experts, the target fusion operation sequence is formally submitted to the execution module, thereby completing the transformation process from multiple parallel fusion decision plans to a single executable target fusion operation sequence.

[0103] See Figure 4 This is a comparative chart of the state changes in the knowledge topology network of geological multimodal data. Its core purpose is to evaluate the improvement effect of different fusion decision-making plans on key network indicators. Dimensionlessness is used to achieve unified evaluation of multiple indicators and general analysis across scenarios, while adapting to the AI ​​and topology network computation needs of geological multimodal data fusion. After the implementation of all plans, the network indicators were higher than the baseline values ​​before implementation, proving the effectiveness of the plans. This chart serves as a guide for plan selection in geological multimodal data fusion decision-making. If network confidence needs to be improved, PA-04 is preferred; if data freshness needs to be improved, PA-02 is preferred. Feature enhancement and data coordination plans can be combined to achieve comprehensive optimization of network indicators. Quantifying the value of plans through indicator changes provides data support for subsequent plan iterations.

[0104] In one embodiment of the present invention, in a specific implementation, executing the target fusion operation sequence and synchronously evolving the knowledge topology network according to the state variables generated during the execution process specifically includes the following operations: the pre-execution engine reads each atomic operation in the target fusion operation sequence in sequence; for atomic operations that are data acquisition type, the pre-execution engine calls the resource scheduling interface to dispatch observation equipment to perform supplementary observation tasks, such as scheduling an aircraft equipped with a high-precision gravimeter to perform supplementary aeromagnetic measurements in a designated area, and waiting for the return of new structured data packets, and then triggering the aggregation and preliminary structuring process of geological multimodal raw data, injecting the newly generated structured data packets into the geological data buffer pool, and the newly generated structured data packets contain updated field strength distribution characteristics. For atomic operations that are attribute arbitration type, the contingency plan execution engine loads the current state of relevant semantic nodes and their associated edges from the geological knowledge graph library. Loading relevant semantic nodes means loading the arbitration calculation target node and all the semantic nodes that provide data from its upstream. The arbitration calculation logic is executed. For example, two different density values ​​from gravity inversion and core testing are calculated using an error-based weighted average algorithm. The calculation results are used to directly update the density attribute value in the attribute set of the corresponding semantic node in the knowledge topology network, and the changes are written to persistent storage.

[0105] In some embodiments, after each atomic operation is completed, the contingency execution engine captures the operation execution status code and related intermediate variables. The operation execution status code indicates success, failure, or partial completion. The intermediate variables include the amount of data freshness improvement, the magnitude of attribute change, and resource consumption. The amount of data freshness improvement is calculated by comparing the timestamp of the newly injected data with the timestamp of the original data. The magnitude of attribute change is represented by the absolute or relative difference between the attribute value calculated by arbitration and the original attribute value. Resource consumption is recorded as equipment man-hours, computation core hours, or monetary cost. When all atomic operations in the target fusion operation sequence have been completed, the contingency execution engine summarizes all operation execution status codes and intermediate variables to generate a fusion execution summary. The fusion execution summary lists the execution result of each atomic operation, the resources consumed, and the specific changes it caused to the network state in a structured report format. Based on the information from the fused execution summary, a partial reconstruction of the knowledge topology network is automatically triggered. This partial reconstruction recalculates the geological confidence scores and fusion-ready status markers of the affected semantic nodes. The recalculation of geological confidence scores comprehensively considers data quality assessment coefficients, information freshness indicators, and the consistency of multi-source data. It also re-evaluates the strength weights of the semantic edges connected to these nodes, based on the updated geological confidence scores of the two connected semantic nodes. The changed portions of the partially reconstructed knowledge topology network are synchronously updated to the geological knowledge graph database. This synchronous update operation is performed in a transactional manner to ensure data consistency, completing a single evolution of the knowledge topology network.

[0106] Optionally, after a single evolution of the knowledge topology network is completed, a network health assessment and feedback adjustment phase is set up. This phase calculates the global health index of the evolved knowledge topology network. The global health index is a weighted composite of multiple sub-indicators, including: the average geological confidence score of semantic nodes, the average strength weight of semantic edges, the number of connected subgraphs in the network, and the average length of high-value information flow links. The calculated global health index is compared with a historical health index sequence, which stores the index values ​​recorded after several previous evolutions. This comparison determines whether the current evolution has led to an improvement, decline, or near-period in health, based on a set threshold for the magnitude of change.

[0107] It is understandable that if the overall network health index comparison result shows an improvement in health, then the patterns of atomic operations in the target fusion operation sequence that led to this improvement will be strengthened. Strengthening operations involves extracting the type combinations, parameter settings, and execution order characteristics of these atomic operations, using their operation parameters and execution order as successful experience templates, and storing them in the contingency plan template library. Records stored in the contingency plan template library will be associated with the characteristics of the original path detection report that triggered this operation sequence, providing priority reference when generating future fusion decision contingency plans. If the overall network health index comparison result shows a decline in health, then the target fusion operation sequence will be back-analyzed. Back-analysis will use log analysis to determine the immediate impact of each atomic operation on the geological confidence scores of relevant semantic nodes and the semantic edge connection status. Atomic operations that cause a significant drop in the geological confidence scores of key semantic nodes or the breakage of important semantic edges will be identified. These operations will be included in a list of cautious operations, and higher trigger thresholds or avoidance suggestions will be applied when generating subsequent fusion decision contingency plans. The health assessment results, successful experience templates, and caution operation checklists are used as feedback information streams to generate multiple parallel fusion decision-making plans based on the detection results. The feedback information streams are transmitted through a shared database or message queue to influence the generation strategy of subsequent plans. For example, when generating feature enhancement plans, parameters in the successful experience templates are prioritized, or high-risk arbitration models recorded in the caution operation checklist are avoided when generating data coordination plans.

[0108] See Figure 5 This is a quality and coverage assessment map for geological multimodal data, primarily used to analyze the performance of different data modalities in terms of "data quality" and "spatial coverage." This map serves to formulate fusion strategies for geological multimodal data, combining "high-quality, small-scale" chemical element data with "large-scale, high-quality" geological structural morphology data to ensure accuracy while expanding coverage. To address the "coverage shortcomings" of chemical element data, its sampling density can be increased to improve spatial coverage. In the construction of knowledge topology networks, "high-quality + wide-coverage" geological structural morphology data can be given higher weights to improve the overall reliability of the network.

[0109] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0110] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for AI fusion, identification, and analysis of geological multimodal data, characterized in that, The process includes the following: The process of collecting and preliminarily structuring geological multimodal raw data, wherein the geological multimodal raw data includes at least physical field data, chemical element data, geological structural morphology data, and deep drilling core data; The process of constructing and iteratively optimizing a knowledge topology network based on geological semantics, wherein the knowledge topology network is used to express the semantic associations and spatial constraints between geological data entities of different modalities; An active path detection process is initiated in the knowledge topology network to discover high-value information flow links and potential convergence points of interpretive contradictions; Based on the detection results, multiple parallel fusion decision-making plans are generated. Each fusion decision-making plan includes data coordination rules and feature enhancement directions for specific geological targets. External intervention commands are introduced to screen and reconstruct the fusion decision-making plan, forming a target fusion operation sequence; The target fusion operation sequence is executed, and the knowledge topology network is synchronously evolved based on the state variables generated during the execution process; The process of generating multiple parallel fusion decision plans based on the detection results specifically includes the following steps: Analyze the received path detection reports to separate the set of high-value information flow links and the set of potential points of convergence of contradictory interpretations; For each high-value information flow link, a feature enhancement plan is generated. The core of the feature enhancement plan is to improve the information freshness index of key semantic nodes in the high-value information flow link and enhance the strength weight of weak semantic edges in the link. The implementation method is to schedule additional data collection resources to conduct supplementary observations of the geological areas involved in the high-value information flow link and inject new structured data packets into the knowledge topology network, thereby updating the attribute set of relevant nodes. For each potential convergence point of interpretation contradictions, a set of data coordination plans is generated. The core of the data coordination plan is to perform multi-source verification and contradiction resolution on the geological data entity corresponding to the convergence point. The implementation method is to backtrack all input semantic edges of the semantic node, compare the consistency of data units from different sources on the semantic node attributes, and when the inconsistency exceeds the tolerance, an arbitration calculation based on geological and physical laws is initiated, and the arbitration result is used to overwrite the original attribute value, while the arbitration log is recorded. All generated feature enhancement plans and data coordination plans are categorized and packaged according to their operational targets and estimated resource consumption to form multiple independent fusion decision-making plan files. Each file contains a list of executable operation instructions and a description of expected network state changes.

2. The geological multimodal data AI fusion recognition and analysis method according to claim 1, characterized in that, The aggregation and preliminary structuring of the geological multimodal raw data are specifically achieved through the following steps: The raw data stream is asynchronously retrieved from distributed geological observation nodes and centralized geological databases, and the raw data stream is then formatted and time-stamped. Modality-specific analysis was performed on the interpreted and aligned data. For physical field data, the field strength distribution characteristics and anomalous contour morphology were extracted; for chemical element data, the element abundance sequence and isotope ratio combination were extracted; for geological structural morphology data, the interface occurrence parameters and spatial distribution trend were extracted; and for deep drilling core data, the lithological columnar section and micropore structure parameters were extracted. Each parsed data unit is assigned a globally unique geological data entity identifier and encapsulated into a structured data package. The structured data package contains an ontology description of the data unit, a quality assessment coefficient, a spatiotemporal coordinate index, and a pointer to the original data source. The encapsulated structured data packets are injected into a temporary geological data buffer pool according to their spatiotemporal coordinate indices, awaiting invocation during the construction and iterative optimization process of the knowledge topology network.

3. The geological multimodal data AI fusion identification and analysis method according to claim 2, characterized in that, The construction and iterative optimization process of the geological semantic-based knowledge topology network is accomplished in the following way: Structured data packets are extracted in batches from the geological data buffer pool according to a spatiotemporal sliding window. Based on the ontology description in the structured data packet, each data unit is instantiated as a semantic node in the knowledge topology network. The semantic node has an attribute set, which includes at least modality type, data quality assessment coefficient, geological confidence score, information freshness index and fusionable state label. Based on a pre-defined geological knowledge rule base, semantic edges are dynamically established between semantic nodes. The geological knowledge rule base defines the causal relationships, spatial inclusion relationships, time series relationships, and physical property response relationships that may exist between different types of geological data entities. Each semantic edge records the type of relationship, the strength weight of the relationship, and a summary of the geological basis for the relationship. After the semantic edges are established, the network self-optimization loop is started: the average path length and clustering coefficient of the entire knowledge topology network are calculated. If the average path length exceeds the threshold or the clustering coefficient is lower than the threshold, the rule base is supplemented and the semantic edges are reconnected until the network structure indicators are stable within the preset range. The optimized knowledge topology network, along with all its semantic nodes, attribute sets, and semantic edge information, will be persistently stored in the geological knowledge graph database for use in subsequent active pathfinding processes.

4. The geological multimodal data AI fusion identification and analysis method according to claim 3, characterized in that, The process of initiating active path detection in the knowledge topology network is specifically broken down into the following steps: Load the current version of the knowledge topology network from the geological knowledge graph repository; In the knowledge topology network, multiple detection start points and detection end points are preset. The detection start points correspond to semantic nodes with dense original observation data, and the detection end points correspond to semantic nodes corresponding to advanced geological interpretation or resource evaluation conclusions. Concurrent exploration threads are started, with each thread starting from a different detection point and exploring along semantic edges at a limited depth. The exploration decision is based on the weighted sum of the strength weight of the semantic edge and the geological confidence score of the endpoint semantic node. During the exploration process, the sequence of semantic nodes traversed by the path, the sequence of semantic edges, and the overall confidence score of the path are recorded in real time. After all exploration threads terminate, collect all valid paths that reach the exploration endpoint; sort the valid paths in descending order according to the cumulative confidence value of the overall path, and mark the top-ranked valid paths as high-value information flow links. Cross-comparison analysis was performed on all valid paths to identify semantic nodes that were frequently passed by multiple paths but whose geological confidence scores in their attribute sets were significantly lower than the network mean. These semantic nodes were marked as potential points of convergence of explanatory contradictions. The detailed structure of the high-value information flow links output by the detection process and the list of nodes where potential contradictions in interpretation converge are packaged into a path detection report and passed to the subsequent fusion decision-making plan generation process.

5. The geological multimodal data AI fusion identification and analysis method according to claim 4, characterized in that, The introduction of external intervention commands to screen and reconstruct the fusion decision plan, forming a target fusion operation sequence, is achieved through the following mechanism: A visual decision-making interface is provided, which presents the list of operation instructions and the expected changes in network status from multiple integrated decision-making plans to geological experts in an interactive manner. Through the visual decision-making interface, external intervention instructions input by geological experts are received. These external intervention instructions include marking the adoption of a specific plan, adjusting the execution order of operation instructions, modifying parameters in operation instructions, and cross-combining instructions from different plans. Establish a contingency plan execution engine, which receives external intervention instructions and original fusion decision contingency plan documents; The contingency plan execution engine first filters out contingency plan files that are not selected based on the adoption flag; Next, the contingency plan execution engine reorders and updates the operation instructions and parameters in the retained contingency plan file according to the adjusted execution order and modified parameters; Then, the pre-plan execution engine merges the operation instructions from different pre-plan files into a unified execution list according to their logical dependencies, based on the cross-combination instructions; Finally, the pre-plan execution engine performs logical verification and resource conflict checks on this unified execution list. After verification and checks, it solidifies it into the final target fusion operation sequence, which is a set of atomic operations arranged strictly according to time sequence and conditions.

6. The geological multimodal data AI fusion identification and analysis method according to claim 5, characterized in that, The execution of the target fusion operation sequence, and the synchronous evolution of the knowledge topology network based on the state variables generated during the execution process, specifically includes the following operations: The contingency execution engine reads each atomic operation in the target fusion operation sequence sequentially. For atomic operations that are data acquisition type, the engine calls the resource scheduling interface, dispatches the observation equipment to perform supplementary observation tasks, and waits for the return of new structured data packets. Then, it triggers the aggregation and preliminary structuring process of the geological multimodal raw data, and injects the newly generated structured data packets into the geological data buffer pool. For atomic operations that are attribute arbitration type, the engine loads the current state of the relevant semantic nodes and their associated edges from the geological knowledge graph library, executes the arbitration calculation logic, directly updates the attribute set of the corresponding semantic nodes in the knowledge topology network with the calculation results, and writes the changes to persistent storage. After each atomic operation is completed, the engine captures the operation execution status code and related intermediate variables, including the amount of data freshness improvement, the magnitude of attribute changes, and the amount of resource consumption. Once all atomic operations in the target fusion operation sequence have been executed, the engine summarizes all operation execution status codes and intermediate variables to generate a fusion execution summary. Based on the fusion execution summary, especially the information regarding the magnitude of attribute changes and the amount of data freshness improvement, a local reconstruction of the knowledge topology network is automatically triggered: the geological confidence scores and fusionable state labels of the affected semantic nodes are recalculated, and the strength weights of the semantic edges connected to them are re-evaluated. The modified parts of the partially reconstructed knowledge topology network are synchronously updated to the geological knowledge graph database, completing a single evolution of the knowledge topology network.

7. The geological multimodal data AI fusion identification and analysis method according to claim 3, characterized in that, The construction and dynamic addition of the geological knowledge rule base are accomplished through the following independent steps: In the initial stage, a series of "premise-conclusion" production rules were manually extracted and formalized from authoritative geology textbooks, published high-quality geological map descriptions, and geological stratification and interpretation specifications in industry standards, forming the initial core of the rule base; In the construction and self-optimization loop of the knowledge topology network, when the rule base is supplemented, the rule mining routine is started: the rule mining routine analyzes the frequently co-occurring semantic node patterns and semantic edge patterns in the current network structure, and automatically abstracts the co-occurrence patterns of geological data entities that meet the support and confidence thresholds into new candidate production rules. A rule review queue is established, in which all candidate production rules generated by automatic mining and new rules proposed by external geological experts through the interactive interface are entered into the rule review queue. Regularly, multiple geological experts conduct back-to-back reviews of the rules in the rule review queue. Only rules that are approved by the majority of experts will be formally incorporated into the geological knowledge rule base. Each formal rule records its source, confidence level, effective time, and last invocation time. Automatically mined rules are assigned an initial confidence level lower than that of manually generated and authoritative rules, and their confidence level gradually increases as they are successfully verified more often in subsequent network construction.

8. The geological multimodal data AI fusion identification and analysis method according to claim 6, characterized in that, After a single evolution of the knowledge topology network is completed, a network health assessment and feedback adjustment phase is set up, which includes the following steps: The global health index of the evolved knowledge topology network is calculated. The global health index is a weighted sum of multiple sub-indicators, including: the average geological confidence score of semantic nodes, the average strength weight of semantic edges, the number of network connected subgraphs, and the average length of high-value information flow links. The calculated global health index is compared with the historical health index sequence to determine whether the current evolution has led to an improvement, a decrease, or a basically unchanged health level. If the improvement is due to health status, the pattern of atomic operations in the target fusion operation sequence that led to this improvement will be strengthened, and its operation parameters and execution order will be used as successful experience templates and stored in the contingency plan template library for priority reference when generating fusion decision contingency plans in the future. If the health level declines, back-analyze the target fusion operation sequence to identify the atomic operations that cause a significant drop in the geological confidence score of key semantic nodes or breakage of important semantic edges. Add these operations to the list of cautious operations and apply higher trigger thresholds or avoidance suggestions when generating fusion decision plans in the future. The health assessment results, successful experience templates, and cautionary operation checklists are used as feedback information streams and input into the process of generating multiple parallel fusion decision plans based on the detection results, so as to influence the generation strategy of subsequent plans.

9. A geological multimodal data AI fusion recognition and analysis system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the geological multimodal data AI fusion identification and analysis method as described in any one of claims 1 to 8.

Images