Method for constructing mapping knowledge domain of metallogenic space-time events fused with large model

Abstract

The invention discloses a method for constructing a mapping knowledge domain of a metallogenic time-space event fused with a large model, which belongs to the technical field of computers and comprises the following steps of: acquiring multi-source heterogeneous geological data, generating a unified semantic primitive carrying a dynamic credibility weight through multi-modal semantic fusion, constructing a dynamic cognitive energy field based on a probability theory and a dynamics principle, and constructing a mapping knowledge domain of the dynamic cognitive energy field. Releasing a probability event probe in an energy field, generating an initial spatio-temporal event chain containing a plurality of possible paths through potential energy gradient descent and probability exploration, constructing a cognitive isomer system containing an expert simulator, a data insight body and a meta-cognitive arbiter, performing cognitive game on the initial event chain to obtain a game result, and performing cognitive game on the initial event chain. State prediction is carried out based on multi-dimensional health indexes, and predictive self-repairing is carried out on the knowledge graph; by adopting the technical scheme of constructing a dynamic cognitive energy field and a cognitive isomer system for coevolution, dynamic evolution, deep cognitive gaming and predictive self-repairing of the knowledge graph can be realized.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method for constructing a knowledge graph of mineralization spatiotemporal events that integrates large models. Background Technology

[0002] Metallogenic spatiotemporal event knowledge graphs are an intelligent technology that organizes and expresses knowledge related to metallogenic processes in a graph structure. They aim to reveal the evolutionary patterns of mineralization in time and space by integrating and analyzing multi-source data from geology, geophysics, and geochemistry. This technology represents a significant application of artificial intelligence in geological prospecting. By constructing a knowledge network encompassing geological entities, events, and their complex relationships, it provides decision support for mineral resource exploration. Its core lies in how to extract, integrate, and structure geological knowledge from massive, heterogeneous data.

[0003] In existing technologies, the construction of mineralization knowledge graphs typically employs rule-based or statistical methods. Rule-based methods rely primarily on predefined ontology and reasoning rules from geological experts to map data onto the knowledge graph. Statistical methods, especially the large-scale modeling techniques that have emerged in recent years, utilize deep learning models to automatically extract entities and relationships from massive amounts of text and data. Some advanced methods attempt to combine the two approaches, for example, by first using a large-scale model for preliminary extraction and then using expert rules for verification and correction, or simply by weighted fusion of the results from the two methods.

[0004] However, existing technical solutions have obvious shortcomings. Rule-based methods construct knowledge graph structures that are rigid and difficult to adapt to the dynamic changes of new data and new knowledge. Statistical methods, while more flexible, are easily affected by data noise and lack strong constraints from geological logic, which may produce "illusions" that do not conform to common sense in the domain. Simply combining the two is often just a superficial integration and fails to form a mechanism for deep interaction and collaborative optimization. As a result, when facing complex and uncertain mineralization problems, cognitive biases from either side may be transmitted and amplified. At the same time, the entire system lacks real-time monitoring and self-correction capabilities for its own knowledge quality, making it easy for errors and uncertainties to accumulate and affecting the long-term reliability of the knowledge graph. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a method for constructing a knowledge graph of mineralization spatiotemporal events that integrates large models. It employs a technical solution of constructing a dynamic cognitive energy field and a cognitive heterogeneous system for co-evolution, which enables dynamic evolution of the knowledge graph, deep cognitive game theory, and predictive self-repair.

[0006] The above objectives can be achieved through the following approach: A method for constructing a knowledge graph of mineralization spatiotemporal events that integrates large-scale models, the method comprising: Acquire multi-source heterogeneous geological data and perform multi-modal semantic fusion processing on the multi-source heterogeneous geological data to generate unified semantic primitives carrying dynamic credibility weights; A dynamic cognitive energy field is constructed based on the principles of probability theory and dynamics, and unified semantic primitives are mapped onto the dynamic cognitive energy field to obtain a dynamic cognitive energy field carrying semantic primitive mappings. In a dynamic cognitive energy field, probabilistic event probes are released, and spatiotemporal event chains are self-assembled through potential energy gradient descent and probabilistic exploration to generate an initial spatiotemporal event chain containing multiple alternative paths. Construct a cognitive heterogeneous system comprising an expert simulator, a data insight entity, and a metacognitive arbitrator. The metacognitive arbitrator organizes the expert simulator and the data insight entity to engage in cognitive game theory on the initial spatiotemporal event chain, generating game results with consensus and divergence point markers. Real-time monitoring of dynamic cognitive energy field and initial spatiotemporal event chain; calculation of multi-dimensional health indicators; generation of health status prediction based on multi-dimensional health indicators; triggering predictive self-repair intervention based on health status prediction; obtaining the repaired spatiotemporal event chain. Based on the game results and the repaired spatiotemporal event chain, the parameters of the dynamic cognitive energy field are optimized through a co-evolutionary algorithm to generate an optimized spatiotemporal event knowledge graph of mineralization.

[0007] Optionally, the step of performing multimodal semantic fusion processing on the multi-source heterogeneous geological data to generate a unified semantic primitive carrying dynamic credibility weights includes: extracting textual descriptive features, geophysical features, remote sensing image features, and geochemical features from the multi-source heterogeneous geological data to obtain multimodal raw features; performing feature alignment and normalization processing on the multimodal raw features to generate multimodal feature vectors; and performing weighted fusion of the multimodal feature vectors using an attention model for evaluating feature importance to generate the unified semantic primitive, wherein the dynamic credibility weights are dynamically adjusted based on the acceptance of the unified semantic primitive in subsequent cognitive games.

[0008] Optionally, the step of constructing a dynamic cognitive energy field based on probability theory and dynamics principles, and mapping unified semantic primitives to the dynamic cognitive energy field to obtain a dynamic cognitive energy field carrying semantic primitive mappings, includes: establishing a dynamic cognitive energy field mathematical model containing micro, meso, and macro layers; defining cognitive association operators for calculating semantic similarity, spatiotemporal proximity, and causal association, and using the cognitive association operators to calculate the association strength between the unified semantic primitives; and mapping the unified semantic primitives to the corresponding levels of the dynamic cognitive energy field mathematical model based on the association strength to generate the dynamic cognitive energy field carrying semantic primitive mappings.

[0009] Optionally, the step of releasing probabilistic event probes in a dynamic cognitive energy field carrying semantic primitive mappings, and generating an initial spatiotemporal event chain containing multiple alternative paths through potential energy gradient descent and probabilistic exploration to perform spatiotemporal event chain self-assembly includes: initializing the position of the probabilistic event probes according to the potential energy distribution of the dynamic cognitive energy field, and setting the exploration probability threshold of the probabilistic event probes; guiding the movement of the probabilistic event probes based on the potential energy gradient of the dynamic cognitive energy field, allowing the probabilistic event probes to perform probabilistic exploration according to the exploration probability threshold, and recording their movement trajectories; aggregating multiple movement trajectories, and performing clustering and pruning processing on the movement trajectories to generate the initial spatiotemporal event chain.

[0010] Optionally, the step of organizing the expert simulation and the data insight entity to conduct cognitive game on the initial spatiotemporal event chain through the metacognitive arbitrator to generate a game result with consensus and divergence point markers includes: the expert simulation entity performing symbolic reasoning on the initial spatiotemporal event chain based on a reasoning system with a built-in geological rule base and mineralization case base to generate expert reasoning results; the data insight entity integrating large models adapted to multiple domains performing pattern discovery and anomaly detection on the initial spatiotemporal event chain to generate data-driven results; and the metacognitive arbitrator fusing the confidence of the expert reasoning results and the data-driven results based on an improved confidence propagation algorithm to generate the game result.

[0011] Optionally, the real-time monitoring of the dynamic cognitive energy field and the initial spatiotemporal event chain, the calculation of multi-dimensional health indicators, the generation of a health status prediction based on the multi-dimensional health indicators, and the triggering of predictive self-repair intervention based on the health status prediction to obtain the repaired spatiotemporal event chain include: calculating the activity index of the unified semantic primitive, the logical consistency index of the initial spatiotemporal event chain, and the evolutionary stability index of the dynamic cognitive energy field to constitute the multi-dimensional health indicators; inputting the multi-dimensional health indicators into a health prediction model for predicting the trend of time series changes to generate the health status prediction; when the health status prediction indicates an abnormal risk, triggering the corresponding predictive self-repair intervention based on the type and severity of the abnormal risk to repair the initial spatiotemporal event chain to obtain the repaired spatiotemporal event chain.

[0012] Optionally, the parameter optimization of the dynamic cognitive energy field using a co-evolutionary algorithm includes: analyzing the game result and converting it into a potential energy field parameter adjustment instruction to perform top-down cognitive-driven optimization of the dynamic cognitive energy field; extracting new patterns and new associations from the repaired spatiotemporal event chain and feeding them back to the rule base of the cognitive heterogeneous system for data-driven optimization; identifying high-quality cognitive features based on the consensus markers in the game result and the health indicators corresponding to the repaired spatiotemporal event chain, storing these high-quality cognitive features in a cognitive gene bank, and applying an elite retention strategy during the optimization process to ensure the inheritance of these high-quality cognitive features.

[0013] Optionally, the metacognitive arbitrator, based on an improved confidence propagation algorithm, integrates the confidence levels of the expert reasoning results and the data-driven results to generate the game result, including: constructing the initial spatiotemporal event chain into a confidence network with events as nodes and logical dependencies as edges; inputting the expert reasoning results and the data-driven results as external evidence into the confidence network, and iteratively updating the confidence values ​​of each node based on a message passing mechanism until convergence; marking each node as a consensus or divergence point according to the converged confidence values ​​and a preset confidence threshold, and summarizing to generate the game result.

[0014] Optionally, triggering the corresponding predictive self-healing intervention based on the type and severity of the anomaly risk includes: defining an anomaly classification system that includes data-level anomalies, logical anomalies, and evolutionary anomalies; when the anomaly risk is a data-level anomaly, triggering semantic primitive replacement and dynamic credibility weight reassessment; when the anomaly risk is a logical anomaly, initiating event chain reconstruction and association strength recalculation; and when the anomaly risk is an evolutionary anomaly, performing potential field parameter reset and path reinitialization.

[0015] Optionally, the system also includes constructing a co-evolutionary engine to establish a two-way linkage between the metacognitive arbitrator and the predictive self-repair intervention; creating a collaborative knowledge base to bidirectionally store game result data generated by the metacognitive arbitrator in previous cognitive games and repair cases of self-repair intervention; feeding back successful repair cases to the metacognitive arbitrator to optimize the initial confidence of subsequent games, while converting persistent divergences that recur in the game results into high-priority health risks to trigger the pre-deployment of self-repair; and based on the data in the collaborative knowledge base, jointly fine-tuning the propagation parameters of the metacognitive arbitrator and the strategy logic of the self-repair intervention to drive the overall co-evolution of the system.

[0016] Compared with the prior art, the present invention has the following advantages: This invention realizes the transformation of knowledge graphs from static construction to dynamic evolution by constructing a dynamic cognitive energy field and a cognitive heterogeneous system. This enables the knowledge graph to respond to new input data and cognition in real time and continuously self-optimize through a co-evolutionary algorithm, thereby fundamentally solving the limitations of traditional knowledge graphs and ensuring the timeliness of knowledge and its long-term applicability in complex geological environments.

[0017] This invention organizes expert knowledge and big model data insights into a deep cognitive game by establishing a metacognitive arbitrator, which goes beyond simple technology superposition. It enables the logical rigor of symbolic reasoning and the data-driven pattern discovery capability to mutually verify and stimulate each other, significantly improving the accuracy of reasoning about complex and uncertain mineralization processes, and discovering potential mineralization patterns that transcend traditional cognition.

[0018] This invention introduces a predictive self-repair intervention mechanism based on health status prediction, which endows the system with forward-looking self-monitoring and repair capabilities. It can proactively intervene before data anomalies or logical conflicts actually occur, thereby improving the robustness and anti-interference ability of the entire knowledge graph construction system and ensuring that it can maintain high-quality and high-stability output when facing continuously dynamically changing data streams. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram illustrating the construction of the dynamic cognitive energy field of the present invention; Figure 3 This is a schematic diagram of the self-assembly process of the probability event probe of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0022] Reference Figure 1As shown, one embodiment of the present invention proposes a method for constructing a knowledge graph of mineralization spatiotemporal events that integrates large models. The method includes: Acquire multi-source heterogeneous geological data and perform multi-modal semantic fusion processing on the multi-source heterogeneous geological data to generate unified semantic primitives carrying dynamic credibility weights; A dynamic cognitive energy field is constructed based on the principles of probability theory and dynamics, and unified semantic primitives are mapped onto the dynamic cognitive energy field to obtain a dynamic cognitive energy field carrying semantic primitive mappings. In a dynamic cognitive energy field, probabilistic event probes are released, and spatiotemporal event chains are self-assembled through potential energy gradient descent and probabilistic exploration to generate an initial spatiotemporal event chain containing multiple alternative paths. Construct a cognitive heterogeneous system comprising an expert simulator, a data insight entity, and a metacognitive arbitrator. The metacognitive arbitrator organizes the expert simulator and the data insight entity to engage in cognitive game theory on the initial spatiotemporal event chain, generating game results with consensus and divergence point markers. Real-time monitoring of dynamic cognitive energy field and initial spatiotemporal event chain; calculation of multi-dimensional health indicators; generation of health status prediction based on multi-dimensional health indicators; triggering predictive self-repair intervention based on health status prediction; obtaining the repaired spatiotemporal event chain. Based on the game results and the repaired spatiotemporal event chain, the parameters of the dynamic cognitive energy field are optimized through a co-evolutionary algorithm to generate an optimized spatiotemporal event knowledge graph of mineralization.

[0023] Specifically, through multimodal semantic fusion technology, dispersed heterogeneous geological data is transformed into standardized unified semantic primitives carrying dynamic credibility weights. These primitives are then placed into a dynamic cognitive energy field that simulates the cognitive decision-making process. This energy field represents the certainty of knowledge through the level of potential energy. The system uses probabilistic event probes to explore this energy field, autonomously generating multiple alternative initial spatiotemporal event chains by considering both high-probability and low-probability innovation paths. Simultaneously, a cognitive heterogeneous system composed of expert simulators, data insight entities, and metacognitive arbitrators conducts deep cognitive game theory on these event chains, i.e., cross-validation of symbolic knowledge and data patterns, generating game results with consensus and disagreement markers. Furthermore, the system continuously monitors its cognitive health status through multi-dimensional health indicators and triggers self-repair interventions in advance based on state predictions to prevent the accumulation of anomalies. The system utilizes game results and repair feedback to bidirectionally optimize the dynamic cognitive energy field and built-in rules through a co-evolutionary algorithm, thereby driving the entire knowledge graph to continuously evolve towards a more accurate and complete direction.

[0024] This transforms the constructed knowledge graph from a static collection of knowledge into a dynamic entity that grows with new data and insights, solving the problem of traditional knowledge graphs becoming outdated as soon as they are built. It significantly improves the timeliness and application value of knowledge. Through a deep cognitive game mechanism, it achieves complementary advantages and deep integration between the rigorous logic of expert experience and the data insight of large models, overcoming the limitations of single methods in dealing with complex and uncertain problems. It greatly improves the accuracy and reliability of mineralization prediction and decision support. Through the introduction of a predictive self-repair mechanism, it endows the system with unprecedented robustness and self-healing ability, enabling it to operate stably for a long time in real data environments full of noise and change. This method transforms the knowledge graph construction process from a passive manual process into an active and intelligent evolutionary process with continuous learning and innovative discovery capabilities.

[0025] Optionally, the step of performing multimodal semantic fusion processing on the multi-source heterogeneous geological data to generate a unified semantic primitive carrying dynamic credibility weights includes: extracting textual descriptive features, geophysical features, remote sensing image features, and geochemical features from the multi-source heterogeneous geological data to obtain multimodal raw features; performing feature alignment and normalization processing on the multimodal raw features to generate multimodal feature vectors; and performing weighted fusion of the multimodal feature vectors using an attention model for evaluating feature importance to generate the unified semantic primitive, wherein the dynamic credibility weights are dynamically adjusted based on the acceptance of the unified semantic primitive in subsequent cognitive games.

[0026] Specifically, the system acquires textual descriptive data from geological reports, geophysical data from gravity, magnetic, and electrical exploration, remote sensing image data from multispectral satellite imagery, and geochemical data from regional geochemical measurements. For these four types of multi-source, heterogeneous geological data, the system uses different feature extraction models in parallel. For example, it uses a natural language processing model pre-trained in the geological field to extract geological entities, events, and relationships from the textual descriptive data, forming textual descriptive features; it uses convolutional neural networks to analyze the gridded images of geophysical data, identifying tectonic anomalies and physical property interfaces to generate geophysical features; it applies image segmentation and classification models to process remote sensing image data, extracting spatial distribution information such as lithology and alteration zones to obtain remote sensing image features; and it performs statistical analysis and... Spatial interpolation methods identify anomalous combinations and zonation patterns of elements from geochemical data to obtain geochemical features. After obtaining the four independent features mentioned above, feature alignment and normalization are required because their dimensions, scales, and semantic spaces are different. A joint embedding model is adopted, which uses a multi-input single-output neural network architecture to map textual descriptive features, geophysical features, remote sensing image features, and geochemical features into a unified high-dimensional semantic space. In this space, different modal features describing the same geological phenomenon are closer in vector distance. After feature alignment, Z-score standardization is performed on all feature vectors to ensure that features of each dimension have the same contribution weight, thereby generating a unified and standardized multimodal feature vector.

[0027] Subsequently, the multimodal feature vectors are weighted and fused based on a pre-defined credibility assessment model. This model utilizes an attention mechanism to dynamically calculate the importance of each modal feature according to the current analysis objective, and uses this importance as a weight for weighted summation, ultimately generating a highly condensed unified semantic primitive. Crucially, each newly generated unified semantic primitive is assigned an initial dynamic credibility weight. This weight is not static in subsequent system operation but is dynamically adjusted based on its performance in the cognitive game phase. Its update process can be represented by the following formula: , in, It is the updated dynamic credibility weight. This represents the current dynamic confidence weight, where λ is a constant learning rate that controls the update speed. It is the acceptance score of the unified semantic primitive in the most recent cognitive game. When the event chain it participates in is judged as cognitive consensus, Higher values ​​correspond to lower values, and all symbols are dimensionless weights or scores, ensuring logical consistency in the calculations.

[0028] Through the above method, this invention transforms raw, heterogeneous, and uncertain geological data into a standardized unified semantic primitive carrying adaptive credibility information. This not only solves the technical challenge of multi-source data fusion, but also provides an inherent self-learning data quality control capability for the entire knowledge graph construction process by introducing a dynamic credibility weight mechanism. This mechanism enables the system to review and correct the credibility of its most basic data units based on subsequent reasoning results, thereby significantly improving the robustness of the entire knowledge graph and its adaptability to complex geological environments, and realizing the initial transformation from data to credible knowledge.

[0029] Optionally, the step of constructing a dynamic cognitive energy field based on probability theory and dynamics principles, and mapping unified semantic primitives to the dynamic cognitive energy field to obtain a dynamic cognitive energy field carrying semantic primitive mappings, includes: establishing a dynamic cognitive energy field mathematical model containing micro, meso, and macro layers; defining cognitive association operators for calculating semantic similarity, spatiotemporal proximity, and causal association, and using the cognitive association operators to calculate the association strength between the unified semantic primitives; and mapping the unified semantic primitives to the corresponding levels of the dynamic cognitive energy field mathematical model based on the association strength to generate the dynamic cognitive energy field carrying semantic primitive mappings.

[0030] Specifically, the first step is to establish a dynamic cognitive energy field mathematical model based on probability distribution. This model is a multi-dimensional abstract space, where each dimension corresponds to a geological attribute. The system divides this space into three logical levels: the micro-level, the foundational level, directly contains all unified semantic primitives, with each primitive represented as a point; the meso-level, which characterizes geological events or phenomena composed of multiple micro-level primitives, such as "hydrothermal activity" or "fault structures"; and the macro-level, the highest level, which represents complete metallogenic systems composed of multiple meso-level events, such as "porphyry copper deposit systems." To achieve the mapping from micro to macro, the system defines a cognitive association operator. This operator quantifies the association strength between any two unified semantic primitives. The calculation of the association strength integrates prior geological knowledge and the statistical characteristics of the data itself, and its calculation formula is as follows: , in, Representative element and basic elements The strength of the correlation between them The semantic similarity score is obtained by calculating the cosine similarity of the multimodal feature vectors of the two entities. The spatiotemporal proximity score is calculated based on the spatiotemporal coordinates recorded in geological data; the closer the distance, the higher the score. The causal correlation score is obtained by querying a pre-set geological causal relationship knowledge base. For example, there is a high causal correlation score between "magmatic intrusion" and "contact metamorphism". , and These are adjustable weighting coefficients, all of which are dimensionless normalized parameters used to balance the influence of different factors. When the calculated correlation strength... When the correlation exceeds a preset threshold, the system determines that the two unified semantic primitives are strongly correlated and aggregates them at the micro level to form a meso level event node. Similarly, the correlation strength between meso level event nodes can also be calculated using this operator, thereby aggregating them to form macro level mineralization system nodes.

[0031] Finally, the system constructs a potential energy gradient calculation module to generate a potential energy distribution map, intuitively displaying the deterministic and uncertain regions of knowledge. In this model, "potential energy" is defined as a measure of cognitive uncertainty, inversely proportional to the distribution density and association strength of unified semantic primitives. Specifically, in the dynamic cognitive energy field, regions with dense unified semantic primitives and high association strength represent knowledge clusters with high cognitive consensus and logical consistency, and are therefore assigned lower potential energy values, forming "potential energy depressions." Conversely, regions with sparse primitives or chaotic associations correspond to higher potential energy values. This module calculates the potential energy value of each point in the field using a kernel density estimation method, combined with the spatial distribution of unified semantic primitives and their association strength network, ultimately generating a visualized potential energy distribution map.

[0032] By constructing such a hierarchical, dynamic cognitive energy field, referencing... Figure 2 As shown, the originally flat and discrete semantic primitives are transformed into a structured and visualized knowledge topographic map. This representation method not only realizes multi-scale knowledge organization from microscopic geological facts to macroscopic metallogenic systems, but more importantly, by introducing the concept of "potential energy," it provides a clear guide for subsequent reasoning and exploration. The potential energy distribution map clearly marks the "stable zone" and "exploration zone" of knowledge, providing a dynamic and probabilistic navigation basis for the subsequent self-assembly of spatiotemporal event chains. Thus, the knowledge graph construction process is transformed from a simple information linking process into an intelligent process of exploratory reasoning that is closer to human cognition.

[0033] Optionally, the step of releasing probabilistic event probes in a dynamic cognitive energy field carrying semantic primitive mappings, and generating an initial spatiotemporal event chain containing multiple alternative paths through potential energy gradient descent and probabilistic exploration to perform spatiotemporal event chain self-assembly includes: initializing the position of the probabilistic event probes according to the potential energy distribution of the dynamic cognitive energy field, and setting the exploration probability threshold of the probabilistic event probes; guiding the movement of the probabilistic event probes based on the potential energy gradient of the dynamic cognitive energy field, allowing the probabilistic event probes to perform probabilistic exploration according to the exploration probability threshold, and recording their movement trajectories; aggregating multiple movement trajectories, and performing clustering and pruning processing on the movement trajectories to generate the initial spatiotemporal event chain.

[0034] Specifically, based on the dynamic cognitive energy field constructed in the previous step, the system automatically generates an initial spatiotemporal event chain with multiple possibilities, conforming to geological spatiotemporal logic, by simulating the exploration mechanism of an intelligent agent. The system initializes a set of probabilistic event probes. Each probe can be considered a virtual intelligent agent roaming within the dynamic cognitive energy field, with its initial position randomly placed in several high-potential-energy regions within the energy field. This represents starting the exploration from regions of knowledge uncertainty. Simultaneously, the system sets an exploration probability threshold for each probabilistic event probe. This threshold is a dimensionless parameter between 0 and 1, used to control the probe's behavior pattern. During the exploration process, the movement of the probabilistic event probes follows a dual mechanism. Its primary movement mode is based on potential energy gradient descent; that is, the probe preferentially moves along the direction that causes the potential energy value to decrease the fastest. This simulates the tendency in the cognitive process to seek explanatory paths with complete and logically consistent evidence chains. During each step, the probe selects the unified semantic primitive with the lowest potential energy in its neighborhood as the next node. To avoid getting trapped in local optima, the system introduces a probabilistic exploration mechanism. At each decision step, the probe generates a random number. If this random number is lower than a preset exploration probability threshold, the probe ignores the potential energy gradient and randomly selects a unified semantic primitive from its neighborhood as the next target, even if this causes it to move to a region with higher potential energy. This mechanism allows the probe to "jump out" of conventional, highly deterministic logical paths to explore new associations that are low-probability but geologically plausible, thereby discovering potential unconventional mineralization models. The triggering condition for probabilistic exploration can be formally expressed as: , when At that time, probability exploration is triggered.

[0035] in, It is the probability of exploration. It is a random number that is uniformly distributed between 0 and 1. The current location of the probe The normalized potential energy value; the higher the potential energy, the closer its value is to 0, and the higher the exploration probability. The larger the value, the higher the probability that the probe will randomly explore in areas of high uncertainty. Each probabilistic event probe moves continuously in the dynamic cognitive energy field until it reaches the preset step limit or enters a "steady state" with extremely low potential energy (i.e., there are no nodes with lower potential energy to choose from in the surrounding neighborhood). All the unified semantic primitives that the probe passes through in this process are connected in the order of their visits to form a single movement trajectory. The system releases a large number of probabilistic event probes in parallel to obtain multiple such movement trajectories. Finally, the system aggregates these large number of movement trajectories through trajectory clustering and pruning algorithms, merges similar trajectories, removes redundant or logically conflicting paths, and finally generates a representative initial spatiotemporal event chain containing multiple alternative paths.

[0036] Through this self-assembly method that combines "potential energy guidance" and "probabilistic exploration," referencing Figure 3 As shown, it overcomes the limitations of traditional methods that rely on manually set rules or exhaustive search. It can automatically generate multiple mineralization evolution paths from complex knowledge terrain without human intervention. These paths conform to mainstream cognition (potential gradient descent) and contain innovative possibilities (probabilistic exploration). The generated initial spatiotemporal event chain is not a single fixed answer, but a set containing multiple possible hypotheses. This provides rich and high-quality input for subsequent cognitive games and model verification, greatly improving the intelligence level and innovation discovery capability of the knowledge graph construction process.

[0037] Optionally, the step of organizing the expert simulation and the data insight entity to conduct cognitive game on the initial spatiotemporal event chain through the metacognitive arbitrator to generate a game result with consensus and divergence point markers includes: the expert simulation entity performing symbolic reasoning on the initial spatiotemporal event chain based on a reasoning system with a built-in geological rule base and mineralization case base to generate expert reasoning results; the data insight entity integrating large models adapted to multiple domains performing pattern discovery and anomaly detection on the initial spatiotemporal event chain to generate data-driven results; and the metacognitive arbitrator fusing the confidence of the expert reasoning results and the data-driven results based on an improved confidence propagation algorithm to generate the game result.

[0038] Specifically, in order to deeply verify and optimize the initial spatiotemporal event chain generated in the previous stage, the system activates an expert simulation body by simulating a collaborative decision-making process jointly participated in by expert knowledge and data-driven intelligence. This simulation body is a system based on symbolic reasoning, which integrates a geological rule base accumulated by geologists over a long period of time and a large number of classic mineralization case libraries. When an initial spatiotemporal event chain is received, the expert simulation body will check whether the sequence of events and causal relationships in the chain are consistent with the axioms and theorems in the geological rule base, and compare them with similar mineralization processes in the case library to evaluate their geological rationality. After completing the analysis, the expert simulation body generates a confidence score based on symbolic logic for each link of the event chain, forming the expert reasoning result.

[0039] Simultaneously, the Data Insights Platform is activated. This system integrates multiple domain-adapted large models trained on geological data, such as language models for text mining and visual models for image analysis. After receiving the same initial spatiotemporal event chain, the Data Insights Platform performs pattern discovery and anomaly detection tasks. It reverse-engineers the original multi-source heterogeneous geological data, verifies whether each event in the event chain has sufficient data evidence to support it, and detects whether there are anomalous associations in the event sequence that contradict the statistical patterns of large-scale data. Based on the strength of evidence and the degree of pattern matching, the Data Insights Platform generates a data-driven confidence score for the event chain, forming a data-driven result.

[0040] Subsequently, a metacognitive arbitrator intervenes, organizing a cognitive game between an expert simulation and a data insight entity. The arbitrator first collects the expert reasoning and data-driven results from both sides. Instead of simply averaging or voting on the scores, it employs an improved confidence propagation algorithm. This algorithm treats the event chain as a Bayesian network, where each event is a node and the relationships between events are edges. Expert reasoning and data-driven results are input into the network as external evidence. The arbitrator propagates and updates the confidence of each node in the network through an iterative message passing process until the entire network reaches a stable confidence distribution. During this process, a node that receives high confidence support from both sides will have its final confidence significantly enhanced; conversely, if one side supports it while the other opposes it, the node's confidence will be weakened. Based on the improved confidence propagation algorithm, the metacognitive arbitrator constructs the initial spatiotemporal event chain into a confidence network, where each event node... The confidence level is calculated through iterative updates. , in, Represents a node Updated confidence level This is a normalization constant to ensure the normalization of the probability distribution. The fusion function is based on the combined results of expert reasoning. and data-driven results , To the neighbor node Passed to node Message value, This represents the product of message values ​​from all neighboring nodes. The algorithm iterates through multiple rounds of message passing until the confidence levels of all nodes in the network converge. Finally, consensus and divergence points are marked based on the converged confidence levels. After confidence propagation is completed, the metacognitive arbitrator generates the game result based on the final confidence level of each node. Events or associations with confidence levels above a certain high threshold are marked as "consensus," while those below a certain low threshold are marked as "divergence points." The final output game result is an event chain evaluation report with consensus and divergence point markings and deep cross-validation.

[0041] By introducing this cognitive game mechanism, a deep integration far exceeding simple technological superposition is achieved. It establishes a platform where expert knowledge and data intelligence mutually "question" and "support" each other. The logical rigor of expert knowledge constrains the potential "illusions" of large models, while the powerful pattern-finding capabilities of large models supplement and challenge traditional expert cognition. This process not only significantly improves the accuracy and reliability of initial spatiotemporal event chain assessments, but more importantly, it accurately identifies "consensus zones" and "unknown zones" (i.e., points of divergence) within the knowledge system, providing clear and actionable targets for subsequent self-repair and knowledge graph evolution. Optionally, the real-time monitoring of the dynamic cognitive energy field and the initial spatiotemporal event chain, the calculation of multi-dimensional health indicators, the generation of a health status prediction based on the multi-dimensional health indicators, and the triggering of predictive self-repair intervention based on the health status prediction to obtain the repaired spatiotemporal event chain include: calculating the activity index of the unified semantic primitive, the logical consistency index of the initial spatiotemporal event chain, and the evolutionary stability index of the dynamic cognitive energy field to constitute the multi-dimensional health indicators; inputting the multi-dimensional health indicators into a health prediction model for predicting the trend of time series changes to generate the health status prediction; when the health status prediction indicates an abnormal risk, triggering the corresponding predictive self-repair intervention based on the type and severity of the abnormal risk to repair the initial spatiotemporal event chain to obtain the repaired spatiotemporal event chain.

[0042] Specifically, the system comprehensively quantifies the current state of the dynamic cognitive energy field and the initial spatiotemporal event chain by calculating multi-dimensional health indicators in real time. These indicators include the activity index of the unified semantic primitives at the micro level, which is calculated by monitoring the frequency of each primitive being invoked in recent cognitive games and the rate of change of dynamic credibility weights; the logical consistency index of the initial spatiotemporal event chain at the meso level, which is assessed by detecting whether there are logical errors such as causal contradictions and temporal misordering within the event chain; and the evolutionary stability index at the macro level, which is measured by analyzing the overall drastic changes in the potential energy distribution of the dynamic cognitive energy field and the stability of the knowledge graph network topology.

[0043] After acquiring these multi-dimensional health indicators at discrete time points, the system uses them to train a health prediction model. This model typically employs time series analysis algorithms, such as Long Short-Term Memory (LSTM) networks, to learn the evolution patterns of historical health indicator data and predict the possible trends of various indicators within a future time window, thereby generating a health status prediction. This prediction not only includes the numerical prediction of the indicators but also an assessment of the probability of the indicators entering the "warning" or "dangerous" range. For example, the model might predict that the logical consistency index of a certain core event chain has a high probability of falling below a safe threshold within several iterations. When the health status prediction indicates a potential abnormal risk, i.e., before the actual problem occurs, the system proactively triggers predictive self-healing intervention. This is a rule-based decision-making system. Based on the predicted anomaly type and severity, the system automatically selects and executes corresponding repair strategies. Specifically, if the activity index of the unified semantic primitive is predicted to continue to decline, the system will initiate a semantic primitive replacement strategy, automatically retrieving new and relevant multi-source heterogeneous geological data to update or replace the primitive. If the logical consistency index of the initial spatiotemporal event chain is predicted to have problems, the system will trigger event chain reconstruction, performing local re-reasoning on the problematic chain segment or introducing new probabilistic event probes for exploration. If the evolutionary stability index of the entire dynamic cognitive energy field is predicted to fluctuate significantly, which usually means that the system's knowledge structure may undergo disruptive changes or fall into chaos, the highest level of potential energy field parameter reset will be triggered to restore the potential energy distribution in some areas to a more stable initial state to prevent system collapse.

[0044] By establishing a closed-loop process of monitoring and proactive intervention, the traditional passive repair mechanism is upgraded to an active self-healing capability. This predictive self-repair intervention mechanism can significantly improve the robustness and anti-interference ability of the knowledge graph construction system, enabling it to maintain long-term stability and reliability when facing continuously input new data and constantly evolving cognitive models. This ensures the healthy evolution of the knowledge graph, avoids the accumulation and spread of errors and uncertainties, and thus guarantees the overall quality of the final output mineralization spatiotemporal event knowledge graph.

[0045] Optionally, the parameter optimization of the dynamic cognitive energy field using a co-evolutionary algorithm includes: analyzing the game result and converting it into a potential energy field parameter adjustment instruction to perform top-down cognitive-driven optimization of the dynamic cognitive energy field; extracting new patterns and new associations from the repaired spatiotemporal event chain and feeding them back to the rule base of the cognitive heterogeneous system for data-driven optimization; identifying high-quality cognitive features based on the consensus markers in the game result and the health indicators corresponding to the repaired spatiotemporal event chain, storing these high-quality cognitive features in a cognitive gene bank, and applying an elite retention strategy during the optimization process to ensure the inheritance of these high-quality cognitive features.

[0046] Specifically, a bidirectional co-evolutionary algorithm continuously optimizes the dynamic cognitive energy field, enabling it to evolve adaptively. In a top-down cognitive-driven optimization path, the system transforms the game results generated in the cognitive game of previous steps into adjustment instructions for the parameters of the dynamic cognitive energy field. Specifically, for event chains or event associations marked as "consensus," the system reduces the potential energy value of the corresponding path or region in the energy field, forming deeper "potential energy depressions," making them easier to access and reinforce in subsequent explorations. For regions marked as "divergence points," the system appropriately increases their potential energy value and increases the complexity of the potential energy gradient in that region, thereby encouraging subsequent probabilistic event probes to conduct deeper and more diverse explorations in that region. In a bottom-up data-driven optimization path, the system feeds back the spontaneously emerging new patterns and associations in the repaired spatiotemporal event chains to the rule base of the cognitive game stage. If a rule becomes logically sound after self-repair and gains... A new event chain strongly supported by data, whose structure differs from any existing mineralization case in the expert simulation, will be abstracted by the system into a new mineralization pattern candidate and added to the case library of the expert simulation, or the attention preferences of the large model of the data insight body will be updated. This mechanism ensures that the system can learn from data-driven exploration, continuously enriching and improving its built-in prior knowledge system. In order to ensure the stability and effectiveness of the evolutionary process, the co-evolutionary algorithm also introduces a cognitive gene pool and an elite retention strategy. The cognitive gene pool is a database used to store core cognitive features that have been verified through multiple rounds of iteration and have shown high stability and strong explanatory power. During each optimization of the potential field parameters, the elite retention strategy ensures that the energy field structure corresponding to these features stored in the cognitive gene pool is protected and will not be destroyed by random perturbations during the optimization process. This is equivalent to preserving the optimal "genes" in the genetic algorithm, preventing the accidental loss of high-quality knowledge during the evolutionary process.

[0047] Through this two-way optimization mechanism of interaction and co-evolution between potential energy fields and event chains, a closed-loop system capable of self-improvement and continuous learning is constructed. Top-down cognitive drive ensures that the system's evolutionary direction conforms to the logic and consensus of domain experts, while bottom-up data drive injects innovative power into the system to discover new knowledge and break through traditional cognitive frameworks. The cognitive gene pool and elite retention strategy provide stability guarantees for the entire evolutionary process. This whole mechanism makes the final generated spatiotemporal knowledge graph of mineralization no longer a static knowledge warehouse, but a dynamic knowledge ecosystem that can evolve synchronously with new data and new cognition and has vitality.

[0048] Optionally, the metacognitive arbitrator, based on an improved confidence propagation algorithm, integrates the confidence levels of the expert reasoning results and the data-driven results to generate the game result, including: constructing the initial spatiotemporal event chain into a confidence network with events as nodes and logical dependencies as edges; inputting the expert reasoning results and the data-driven results as external evidence into the confidence network, and iteratively updating the confidence values ​​of each node based on a message passing mechanism until convergence; marking each node as a consensus or divergence point according to the converged confidence values ​​and a preset confidence threshold, and summarizing to generate the game result.

[0049] Specifically, the system constructs a confidence network, a graph structure, where nodes represent each core event in the initial spatiotemporal event chain (such as "magmatic intrusion" or "hydrothermal alteration") or their relationships (such as causal relationships or temporal relationships). Directed edges in the network represent the logical dependencies between these events or relationships. These dependencies are generated based on a pre-defined geological rule base. The metacognitive arbitrator injects expert reasoning results and data-driven results as external evidence into this confidence network. Specifically, for each node, two initial confidence values ​​are associated: one from the expert simulation's assessment of its geological rationality, and the other from the data insight's assessment of its data support strength. The system initiates a confidence propagation process based on a message passing mechanism. In this process, each node sends a "message" to its neighboring nodes. The content of the message is an "inference" made by the node about the state of its neighboring nodes based on its current confidence and the other neighboring messages it has received. This process iterates repeatedly until the confidence values ​​of all nodes in the network converge to a stable state. A node's final stable confidence value combines its own initial evidence with the collective "opinion" of all relevant nodes in the network.

[0050] After iterative convergence, the system calculates the consensus degree based on the final stable confidence value of each node. The consensus degree is a comprehensive indicator that reflects not only the node's own confidence level but also its importance in the network (such as the node's degree centrality). The formula for calculating the consensus degree can be designed as follows: , in, It is a node The degree of consensus It is a node The final stable confidence value after confidence propagation It is a node In a confidence network, centrality is measured to indicate the importance of a node in the entire logical chain. All variables are dimensionless normalized values. The metacognitive arbitrator sets two preset thresholds: a high confidence threshold and a low confidence threshold. When a node's consensus is higher than the high threshold, the event or association represented by that node is considered to be a highly reliable consensus. When the consensus is lower than the low threshold, it indicates that there is a significant conflict between expert knowledge and data evidence at that point, or that neither side can provide strong support. The node is therefore marked as a divergence point.

[0051] By deepening the evaluation of event chains from simple, independent scoring comparisons to a networked, global confidence reasoning process, this approach can more precisely capture the complex interactions between expert knowledge and data-driven evidence. It goes beyond simply judging right or wrong; it quantifies the strength of consensus and accurately pinpoints the root causes of conflict, marking the final results as points of consensus and disagreement. This provides clear targets for subsequent knowledge graph optimization, enabling the system to concentrate resources on resolving the most controversial and uncertain knowledge links, thereby greatly improving the efficiency and depth of the cognitive game process.

[0052] Optionally, triggering the corresponding predictive self-healing intervention based on the type and severity of the anomaly risk includes: defining an anomaly classification system that includes data-level anomalies, logical anomalies, and evolutionary anomalies; when the anomaly risk is a data-level anomaly, triggering semantic primitive replacement and dynamic credibility weight reassessment; when the anomaly risk is a logical anomaly, initiating event chain reconstruction and association strength recalculation; and when the anomaly risk is an evolutionary anomaly, performing potential field parameter reset and path reinitialization.

[0053] Specifically, the system defines an anomaly classification system, dividing potential anomalies into three levels. Data-level anomalies are the lowest level, mainly referring to problems with the unified semantic primitives that form the foundation of the knowledge graph, such as outdated data sources or significant conflicts with other primitives, leading to a continuous decline in their dynamic credibility weight. Logic-level anomalies are intermediate level anomalies, referring to structural problems in the initial spatiotemporal event chain composed of multiple semantic primitives, such as reversed event sequence, broken causal chains, or violations of basic geological principles. Evolution-level anomalies are the highest level, referring to problems with the macroscopic state of the entire dynamic cognitive energy field, such as large-scale violent oscillations in potential energy distribution, or the fragmentation of the knowledge graph's network structure, indicating that the overall evolution process of the system has fallen into an unstable state.

[0054] For anomaly predictions at different levels, the system automatically triggers corresponding repair processes. When a health status prediction indicates an impending data-level anomaly, the system initiates a semantic primitive replacement and dynamic credibility weight reassessment procedure for the low-activity unified semantic primitives involved in the prediction. The system performs a reverse query of multi-source heterogeneous geological databases to find the latest data or alternative data sources related to the semantic primitive, and re-executes the multimodal semantic fusion process to generate new semantic primitives to replace the old ones. Simultaneously, a punitive reassessment of its dynamic credibility weight is performed to reduce its short-term impact.

[0055] When a logical-level anomaly is predicted, the system will initiate event chain reconstruction and recalculation of association strength. It will identify the initial spatiotemporal event chain segment with logical risks as indicated in the prediction report and apply a local repair algorithm. This may include invoking repair rules from an expert simulation to adjust the event sequence, or releasing a set of locally targeted probabilistic event probes to explore the dynamic cognitive energy field surrounding the segment to find more reasonable connection paths to complete the event chain reconstruction. Simultaneously, the cognitive association operator will be re-invoked to recalculate the association strength of events within and around the segment, thus reinforcing the repaired logical structure.

[0056] When the highest-level evolutionary anomaly is predicted, it indicates that the system may face the risk of overall instability. At this time, the potential energy field parameters will be reset and the path will be reinitialized. The system will selectively reset the parameters of the most volatile regions in the dynamic cognitive energy field, restore their potential energy distribution to an earlier and more stable state, clear out a large number of low-quality or conflicting initial spatiotemporal event chains, and reinitialize the distribution of probabilistic event probes, guiding the system to start a new round of evolutionary exploration from a healthier and more stable basic state.

[0057] Through this layered and categorized refined self-repair strategy, this invention establishes a sound and efficient system "immune response" mechanism. It can take intervention measures of varying intensity, from "local fine-tuning" to "system-level reset," depending on the severity and nature of the problem. This ensures that the repair behavior can effectively solve the problem without causing unnecessary impact on the overall stability of the system. This intelligent self-repair capability is the key to ensuring the long-term, healthy, and autonomous evolution of the knowledge graph, giving it the core competitiveness of maintaining high quality and high availability in complex and dynamic environments.

[0058] Optionally, the system also includes constructing a co-evolutionary engine to establish a two-way linkage between the metacognitive arbitrator and the predictive self-repair intervention; creating a collaborative knowledge base to bidirectionally store game result data generated by the metacognitive arbitrator in previous cognitive games and repair cases of self-repair intervention; feeding back successful repair cases to the metacognitive arbitrator to optimize the initial confidence of subsequent games, while converting persistent divergences that recur in the game results into high-priority health risks to trigger the pre-deployment of self-repair; and based on the data in the collaborative knowledge base, jointly fine-tuning the propagation parameters of the metacognitive arbitrator and the strategy logic of the self-repair intervention to drive the overall co-evolution of the system.

[0059] Specifically, by constructing a co-evolutionary engine as the central hub, the core task of which is to create and maintain a collaborative knowledge base, designed with a bidirectional storage structure. On the one hand, it records in detail the cognitive memories from the metacognitive arbitrator, including the complete process of each round of cognitive game, especially the events that are ultimately marked as consensus and intractable divergence points and their reasoning basis. On the other hand, it systematically archives all repair cases from predictive self-repair interventions, with each case containing the health risk characteristics that triggered the intervention, the specific repair strategies adopted, and the effect evaluation after the intervention.

[0060] Based on this collaborative knowledge base, a two-way linkage mechanism is realized. In the feedback path from repair to cognition, the co-evolutionary engine analyzes successful repair cases in the knowledge base and extracts event chain structures or data association patterns that have been proven effective in practice. When the metacognitive arbitrator starts a new round of game, the engine will use these patterns as prior knowledge to adjust the initial confidence of the corresponding nodes in the confidence network. That is, it will give higher initial trust to those hypotheses that have been successfully repaired and verified, thereby optimizing the efficiency and accuracy of subsequent games. In the feedforward path from cognition to repair, the engine continuously tracks cognitive memory and identifies stubborn divergences that repeatedly appear in multiple iterations. These points are considered bottlenecks in the system's cognitive ability or deep-seated contradictions in the knowledge system. The engine marks them as high-priority health risks and proactively pushes them to the predictive self-repair intervention module. This enables the self-repair mechanism to be deployed in advance, focusing on monitoring these potential problem areas and even preparing repair plans in advance, transforming passive response into proactive management. Ultimately, the co-evolution engine utilizes the rich data accumulated in the collaborative knowledge base to perform joint fine-tuning tasks. It uses the game results in cognitive memory and the intervention effects in repair cases as joint training labels to construct a unified optimization objective function. It then collaboratively adjusts the propagation parameters of the confidence propagation algorithm in the metacognitive arbitrator and the strategy selection logic in the predictive self-repair intervention module, thereby finding a set of parameter configurations that optimize the overall performance of the system. This means that high-quality consensus can be reached more quickly in the game, while risks can be resolved more accurately and efficiently in self-repair. This drives the entire knowledge graph construction system to co-evolve as an organic whole.

[0061] A bridge has been built between cognitive assessment and system maintenance, forming a complete and self-consistent learning and evolution loop. Through this two-way linkage and collaborative fine-tuning, the system is no longer a simple chain of two independent modules, but an intelligent agent capable of "reflective learning." It can extract cognition from its own repair practices to guide future reasoning and judgment; it can also start from the deep confusion in the cognitive process to guide preventive system maintenance. This mechanism greatly enhances the system's adaptability and intelligence level, enabling it to continuously and autonomously optimize its core algorithms and knowledge structure, ensuring that the final generated spatiotemporal knowledge graph of mineralization is not only of high quality in the present, but also has the ability to grow towards the future and continuously improve itself.

[0062] In this embodiment, to verify the feasibility of the invention in practice, it is applied to a mineral exploration project of a regional geological survey institute targeting a polymetallic metallogenic prediction area in South China. This area has complex geological structures and diverse mineralization information. Traditional mineral exploration methods rely on expert experience, making it difficult to efficiently integrate massive amounts of geological reports, geophysical, remote sensing, and geochemical data, resulting in low efficiency and significant uncertainty in mineral exploration prediction. The geological survey institute hopes to use this invention to construct a self-evolving and continuously learning knowledge graph of spatiotemporal events related to mineralization, thereby improving the accuracy and intelligence of mineral exploration prediction.

[0063] In this embodiment, the Geological Survey Institute, utilizing the method proposed in this invention, first collected multi-source heterogeneous geological data accumulated over the past thirty years within the prediction area, including detailed geological survey reports (textual descriptive data), regional gravity, magnetic, and electrical exploration results (geophysical data), multispectral satellite imagery (remote sensing image data), and 1:50,000 stream sediment measurement data (geochemical data). The system, through multimodal semantic fusion processing, transformed these raw data into unified semantic primitives carrying dynamic credibility weights. Subsequently, these semantic primitives were mapped into a dynamic cognitive energy field, and spatiotemporal event chains were self-assembled by releasing probabilistic event probes. The generated initial event chains were evaluated through game theory by the cognitive heterogeneous system, while the system ensured the stability of the evolutionary process through a cognitive health monitoring and self-repair module. Finally, the knowledge graph was continuously optimized using a co-evolutionary algorithm.

[0064] In the multimodal semantic fusion stage, the system extracts geological entities such as "Yanshanian granite intrusion" and "NE-trending fault zone" from geological reports; identifies annular low-magnetic anomalies related to the rock mass from aeromagnetic anomaly maps using convolutional neural networks; extracts iron staining and hydroxyl alteration zones from remote sensing images using image segmentation models; and delineates anomaly combinations of Cu, Mo, and W elements based on geochemical data. These features from different modalities are mapped to a unified semantic space through a joint embedding model and fused to generate unified semantic primitives such as "Semantic primitive A: Yanshanian rock mass intrusion, accompanied by annular low magnetic anomalies and Cu-Mo anomalies." Each primitive is assigned an initial dynamic confidence weight, for example, 0.7.

[0065] During the spatiotemporal event chain self-assembly stage, the system releases probabilistic event probes. Most probes follow the potential energy gradient descent, moving along classic mineralization paths such as "magmatic intrusion → hydrothermal circulation → wall rock alteration → metal precipitation," generating multiple highly deterministic event chains. For example, event chain 001 completely reproduces the known porphyry copper mineralization model in this area. Simultaneously, a few probes trigger probabilistic exploration in areas of higher potential energy uncertainty. For instance, when probe P-128 moves to a "NE-trending fault zone" element, it randomly jumps to a spatially non-adjacent but geochemically similar "hidden structural anomaly" element, thus generating a novel initial spatiotemporal event chain 015, suggesting the possible existence of a fault-controlled hidden ore body at depth.

[0066] During the cognitive game phase, the expert simulation, based on its built-in porphyry mineralization model, gave event chain 001 a high geological plausibility score, while giving event chain 015 a lower score due to its lack of classical model support. However, when the data insight entity reviewed the original data, it discovered a weak but persistent deep geophysical anomaly and surface geochemical halo in the area indicated by event chain 015, thus giving it a higher data support score. The metacognitive arbitrator, through a confidence propagation algorithm, fused the results from both sides and ultimately marked event chain 001 as "consensus," while marking the "fault-hidden tectonic association" link in event chain 015 as a "disagreement point," with its consensus score below a preset threshold.

[0067] During the cognitive health monitoring and self-repair phase, the system detected that some early-formed event chains were experiencing a decline in their "logical consistency index" due to newly added borehole data input, predicting a logical-level anomaly. The system then triggered event chain reconstruction, releasing local probes to explore around the chain segment, successfully integrating new borehole data (such as the discovery of new alteration zones) into the chain and correcting the logical inconsistencies.

[0068] Finally, during the co-evolution phase, event chain 001, marked as "consensus," reinforced the corresponding potential energy depression in the energy field. Meanwhile, the corresponding region of event chain 015, marked as "divergence point," had its potential energy value increased to incentivize further exploration by more probes. After multiple iterations, a repaired and verified new event chain (an optimized version of the original event chain 015) was confirmed to be highly probable, and the "fracture-controlled deep-seated concealed mineralization mode" it represented was abstracted and stored in the cognitive gene pool.

[0069] Through the application of this invention, the project not only verified the mineralization model of known mineral deposits, but also successfully predicted three new target areas with mineral exploration potential. One of these target areas is highly consistent with the innovative mineralization model represented by event chain 015.

[0070] It should be noted that the electrical connections between the various units described above do not necessarily represent direct or indirect connections. Any indirect connection method can be applied to the embodiments of the present invention as long as it achieves the purpose of the present invention. The above are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the present invention.

[0071] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A method for constructing a spatiotemporal knowledge graph of mineralization events that integrates large-scale models, characterized in that, The method includes: Acquire multi-source heterogeneous geological data and perform multi-modal semantic fusion processing on the multi-source heterogeneous geological data to generate unified semantic primitives carrying dynamic credibility weights; A dynamic cognitive energy field is constructed based on the principles of probability theory and dynamics, and unified semantic primitives are mapped onto the dynamic cognitive energy field to obtain a dynamic cognitive energy field carrying semantic primitive mappings. In a dynamic cognitive energy field, probabilistic event probes are released, and spatiotemporal event chains are self-assembled through potential energy gradient descent and probabilistic exploration to generate an initial spatiotemporal event chain containing multiple alternative paths. Construct a cognitive heterogeneous system comprising an expert simulator, a data insight entity, and a metacognitive arbitrator. The metacognitive arbitrator organizes the expert simulator and the data insight entity to engage in cognitive game theory on the initial spatiotemporal event chain, generating game results with consensus and divergence point markers. Real-time monitoring of dynamic cognitive energy field and initial spatiotemporal event chain; calculation of multi-dimensional health indicators; generation of health status prediction based on multi-dimensional health indicators; triggering predictive self-repair intervention based on health status prediction; obtaining the repaired spatiotemporal event chain. Based on the game results and the repaired spatiotemporal event chain, the parameters of the dynamic cognitive energy field are optimized through a co-evolutionary algorithm to generate an optimized spatiotemporal event knowledge graph of mineralization.

2. The method for constructing a knowledge graph of mineralization spatiotemporal events by integrating a large model according to claim 1, characterized in that, The step of performing multimodal semantic fusion processing on the multi-source heterogeneous geological data to generate a unified semantic primitive carrying dynamic credibility weights includes: extracting textual descriptive features, geophysical features, remote sensing image features, and geochemical features from the multi-source heterogeneous geological data to obtain multimodal raw features; performing feature alignment and normalization processing on the multimodal raw features to generate multimodal feature vectors; and performing weighted fusion of the multimodal feature vectors using an attention model for evaluating feature importance to generate the unified semantic primitive, wherein the dynamic credibility weights are dynamically adjusted based on the acceptance of the unified semantic primitive in subsequent cognitive games.

3. The method for constructing a knowledge graph of mineralization spatiotemporal events by integrating a large model according to claim 1, characterized in that, The method of constructing a dynamic cognitive energy field based on probability theory and dynamics principles, and mapping unified semantic primitives into the dynamic cognitive energy field to obtain a dynamic cognitive energy field carrying semantic primitive mappings, includes: establishing a dynamic cognitive energy field mathematical model containing micro, meso, and macro layers; defining cognitive association operators for calculating semantic similarity, spatiotemporal proximity, and causal association, and using the cognitive association operators to calculate the association strength between the unified semantic primitives; and mapping the unified semantic primitives to the corresponding levels of the dynamic cognitive energy field mathematical model based on the association strength to generate the dynamic cognitive energy field carrying semantic primitive mappings.

4. The method for constructing a knowledge graph of mineralization spatiotemporal events by integrating a large model according to claim 1, characterized in that, The step of releasing probabilistic event probes in a dynamic cognitive energy field carrying semantic primitive mappings, and generating an initial spatiotemporal event chain containing multiple alternative paths through potential energy gradient descent and probabilistic exploration to perform spatiotemporal event chain self-assembly, includes: initializing the position of the probabilistic event probes according to the potential energy distribution of the dynamic cognitive energy field, and setting the exploration probability threshold of the probabilistic event probes; guiding the movement of the probabilistic event probes based on the potential energy gradient of the dynamic cognitive energy field, allowing the probabilistic event probes to perform probabilistic exploration according to the exploration probability threshold, and recording their movement trajectories; aggregating multiple movement trajectories, and performing clustering and pruning processing on the movement trajectories to generate the initial spatiotemporal event chain.

5. The method for constructing a knowledge graph of mineralization spatiotemporal events by integrating a large model according to claim 1, characterized in that, The process of organizing the expert simulation and the data insight entity to conduct cognitive game theory on the initial spatiotemporal event chain through a metacognitive arbitrator to generate a game result with consensus and divergence point markers includes: the expert simulation, based on a reasoning system with a built-in geological rule base and mineralization case base, performs symbolic reasoning on the initial spatiotemporal event chain to generate expert reasoning results; the data insight entity integrates large models adapted to multiple domains to perform pattern discovery and anomaly detection on the initial spatiotemporal event chain to generate data-driven results; and the metacognitive arbitrator, based on an improved confidence propagation algorithm, merges the confidence of the expert reasoning results and the data-driven results to generate the game result.

6. The method for constructing a knowledge graph of mineralization spatiotemporal events by integrating a large model according to claim 5, characterized in that, The process involves real-time monitoring of the dynamic cognitive energy field and the initial spatiotemporal event chain, calculating multi-dimensional health indicators, generating a health status prediction based on these indicators, and triggering predictive self-repair interventions based on the health status predictions to obtain the repaired spatiotemporal event chain. This includes: calculating the activity index of the unified semantic primitive, the logical consistency index of the initial spatiotemporal event chain, and the evolutionary stability index of the dynamic cognitive energy field to constitute the multi-dimensional health indicators; inputting these multi-dimensional health indicators into a health prediction model for predicting time series trends to generate the health status prediction; and when the health status prediction indicates an abnormal risk, triggering corresponding predictive self-repair interventions based on the type and severity of the abnormal risk to repair the initial spatiotemporal event chain, thereby obtaining the repaired spatiotemporal event chain.

7. The method for constructing a knowledge graph of mineralization spatiotemporal events by integrating a large model according to claim 6, characterized in that, The optimization of parameters in the dynamic cognitive energy field using a co-evolutionary algorithm includes: analyzing the game results and converting them into potential energy field parameter adjustment instructions to perform top-down cognitive-driven optimization of the dynamic cognitive energy field; extracting new patterns and associations from the repaired spatiotemporal event chain and feeding these new patterns and associations back into the rule base of the cognitive heterogeneous system for data-driven optimization; identifying high-quality cognitive features based on the consensus markers in the game results and the health indicators corresponding to the repaired spatiotemporal event chain, storing these high-quality cognitive features in a cognitive gene bank, and applying an elite retention strategy during the optimization process to ensure the inheritance of these high-quality cognitive features.

8. The method for constructing a knowledge graph of mineralization spatiotemporal events by integrating a large model according to claim 6, characterized in that, The metacognitive arbitrator, based on an improved confidence propagation algorithm, integrates the confidence levels of the expert reasoning results and the data-driven results to generate the game outcome. This includes: constructing the initial spatiotemporal event chain into a confidence network with events as nodes and logical dependencies as edges; inputting the expert reasoning results and the data-driven results as external evidence into the confidence network, and iteratively updating the confidence values ​​of each node based on a message passing mechanism until convergence; and marking each node as a consensus or divergence point based on the converged confidence values ​​and a preset confidence threshold, and summarizing to generate the game outcome.

9. The method for constructing a knowledge graph of mineralization spatiotemporal events by integrating a large model according to claim 8, characterized in that, The step of triggering corresponding predictive self-healing interventions based on the type and severity of the anomaly risk includes: defining an anomaly classification system that includes data-level anomalies, logical anomalies, and evolutionary anomalies; when the anomaly risk is a data-level anomaly, triggering semantic primitive replacement and dynamic credibility weight reassessment; when the anomaly risk is a logical anomaly, initiating event chain reconstruction and association strength recalculation; and when the anomaly risk is an evolutionary anomaly, performing potential field parameter reset and path reinitialization.

10. The method for constructing a knowledge graph of mineralization spatiotemporal events by integrating a large model according to claim 9, characterized in that, It also includes building a co-evolutionary engine to establish a two-way linkage between the metacognitive arbitrator and the predictive self-healing intervention; by creating a collaborative knowledge base, it bidirectionally stores the game result data generated by the metacognitive arbitrator in each cognitive game and the repair cases of the self-healing intervention; by feeding back successful repair cases to the metacognitive arbitrator to optimize the initial confidence of subsequent games, and at the same time, it transforms the stubborn divergence points that repeatedly appear in the game results into high-priority health risks to trigger the pre-deployment of self-healing. Based on data from the collaborative knowledge base, the propagation parameters of the metacognitive arbitrator and the strategy logic of self-repair intervention are jointly fine-tuned to drive the overall collaborative evolution of the system.

Images