Intelligent ore prospecting prediction method and system based on deposit genetic type knowledge graph

Abstract

The application discloses an intelligent ore-prospecting prediction method and system based on a deposit genetic type knowledge graph, relates to the technical field of intelligent ore-prospecting prediction, and comprises the following steps: before starting reasoning of the knowledge graph, a cross-temporal constraint baseline is constructed, all genetic paths in the deposit genetic type knowledge graph are dynamically time-sequentially layered marked, and potential intersection points identified are solidified as multi-scale time anchor points to limit the overlapping range of the genetic paths in the same time sequence window, so that a time sequence structure that can be dynamically constrained is formed; under the support of the cross-temporal constraint baseline, a force factor residual error calculation mechanism is introduced based on the multi-scale time anchor points, signals generated by overlapping of the genetic paths are calculated layer by layer, and the trajectory deviation of the time anchor points is calculated. Through construction of the time sequence constraint and residual inversion mechanism, the application effectively controls genetic path cross interference, suppresses false enrichment signals, and improves the logical stability and prediction accuracy of ore-prospecting reasoning.

Description

Technical Field

[0001] This invention relates to the field of intelligent mineral exploration prediction technology, specifically to an intelligent mineral exploration prediction method and system based on a knowledge graph of mineral deposit genetic types. Background Technology

[0002] Intelligent mineral exploration prediction based on a knowledge graph of ore deposit genesis is an advanced mineral exploration method that combines geological genetic theory with artificial intelligence technology. It first uses ore deposit genetic classification theory as its core, structuring and organizing multi-source geological exploration information such as geological structure, mineral assemblages, magmatic activity, geochemical anomalies, and remote sensing imagery to construct a knowledge graph containing ore deposit types, ore-controlling factors, spatial distribution patterns, and evolutionary processes. Then, utilizing the relationships and logical reasoning rules within the knowledge graph, combined with machine learning or deep learning models, it intelligently compares and recognizes patterns in geological data from different regions, thereby automatically extracting potential mineral exploration clues. This approach not only overcomes the limitations of traditional mineral exploration relying on experience and single data points, but also enables multi-dimensional coupled analysis under complex geological conditions, improving the accuracy and reliability of predictions and providing a scientific basis for decision-making in mineral resource exploration.

[0003] The existing technology has the following shortcomings:

[0004] In the dynamic reasoning process based on the knowledge graph of ore deposit genetic types, when multiple genetic paths nonlinearly intersect within a specific time window, complex causal superposition effects are triggered, leading to artifact-like reconstructions of the relationships within the knowledge graph. These artifacts not only interfere with the original causal chains but may also generate false mineralization indicator signals at the reasoning level, causing fundamental misjudgments when identifying ore body control factors. Furthermore, the artifact superposition can be amplified by the model into anomalous enrichment features, resulting in false enrichment zones in the prediction results that are completely inconsistent with actual geological conditions. This directly misleads mineral exploration deployments, causing large-scale waste of exploration resources and potentially delaying the discovery of genuine mineralized areas.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent mineral exploration prediction method and system based on a knowledge graph of mineral deposit genetic types, so as to solve the problems in the background art mentioned above.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an intelligent mineral exploration prediction method based on a knowledge graph of ore deposit genetic types, comprising the following steps:

[0008] Before knowledge graph reasoning is initiated, a cross-temporal constraint baseline is constructed, all genetic paths in the knowledge graph of ore deposit genetic types are dynamically and temporally layered and labeled, and the identified potential intersections are solidified into multi-scale time anchors to limit the overlap range of each genetic path within the same temporal window, forming a temporal structure that can be dynamically constrained.

[0009] Supported by the cross-temporal constraint baseline, a force factor residual calculation mechanism is introduced based on multi-scale time anchor points to perform layer-by-layer calculation on the signals generated by the overlap of various causal paths, calculate the trajectory deviation of the time anchor points, and generate a traceable causal residual matrix to enhance the stability of the cross-temporal constraint baseline.

[0010] Under the influence of the causal residual matrix, a phase conjugate inversion chain is established, and phase inversion modeling is performed on the overlapping signals extracted from the causal residual matrix. The correction signal generated by the inversion modeling is dynamically injected into the knowledge graph reasoning process to suppress the artifact reconstruction interference generated in the causal residual matrix.

[0011] Under the constraint of the phase conjugate inversion chain, the counterfactual replay mechanism is run to compare the corrected inference results with the historical causal trajectory layer by layer, generate verification fingerprints for anomaly elimination, and write back the verification fingerprints to the cross-temporal constraint baseline to form a closed-loop steady-state control mechanism. This enables continuous suppression of artifact interference, avoids the generation of false enrichment bands, and improves the accuracy and reliability of intelligent mineral exploration prediction.

[0012] Preferably, the steps of building a cross-temporal constraint baseline before knowledge graph reasoning begins include:

[0013] Based on the constructed knowledge graph of mineral deposit genesis, all causal chains containing mineralization paths are extracted, and each genetic path is quantitatively modeled in terms of time sequence dimension, and divided into multiple time sequence segments with time labels and event characteristics according to the geological evolution stage.

[0014] Based on time series segments, dynamic time series hierarchical labeling is implemented by integrating the temporal sequence, event intensity and spatial overlap between paths, identifying path intersection trends and recording the time interval and geological event characteristics of intersection points.

[0015] Based on the intersection information, time nodes with causal intersection potential are identified and solidified into multi-scale time anchors. The anchors include time labels, path hierarchy indexes and mineralization impact indicators.

[0016] By integrating the relationship between time anchors and time-series hierarchical paths, a directed graph structure with path levels as nodes and anchors as edges is constructed, forming a time-series structure that can be dynamically scheduled and constrained for path selection and anomaly identification and control in subsequent inference processes.

[0017] Preferably, the multi-scale time anchors are arranged in chronological order and weighted by combining their corresponding path hierarchy index and mineralization impact index. This is used to dynamically adjust the participation priority of each anchor during the reasoning process, so as to enhance the constraint effect of the path hierarchical structure at different reasoning stages.

[0018] Preferably, the steps for introducing a force factor residual calculation mechanism based on multi-scale time anchor points include:

[0019] Standardize the spatial trajectory of the causal path at each time anchor point, construct the theoretical trajectory surface and calculate its residual vector to form the time series residual distribution curve;

[0020] A causal influence weight model is constructed based on the residual distribution curve to identify the geological causal relationship corresponding to the trajectory deviation, and a causal propagation map is constructed to generate the initial value of the residual influence matrix.

[0021] An adaptive solution mechanism is introduced based on the residual influence matrix. The matrix elements are optimized layer by layer by the residual correction iteration method and the multi-path fusion debiasing method to obtain a structurally stable causal residual matrix.

[0022] Based on the causal residual matrix, the residual results are mapped to a dynamic temporal hierarchical structure, the path hierarchical results are adjusted, and high-interference anchor points are marked as restrictive nodes to enhance the stability of cross-temporal constraint baselines.

[0023] Preferably, the residual correction iterative method used in the adaptive solution mechanism simultaneously considers the trajectory symmetry, spatial overlap and causal propagation directionality of the time anchor point. It obtains the residual estimate through multiple rounds of iterative calculation, and performs principal component analysis on the residual superposition of multiple paths at the same time anchor point to extract the principal residual components for subsequent inference correction.

[0024] Preferably, the steps of establishing a phase conjugate inversion chain and performing phase inversion modeling include:

[0025] High-interference residual signals are extracted based on the causal residual matrix, and overlapping signals are identified through frequency domain transformation and phase analysis to construct a candidate set of overlapping signals;

[0026] Based on the frequency and phase characteristics of the overlapping signals, a phase-inverse mirror function is constructed, and an inverted signal sequence with a path mapping index is generated to form a set of correction signals;

[0027] The correction signal is dynamically injected into the reasoning path structure of the original knowledge graph according to the path mapping index, and local reasoning reconstruction is performed to adjust the causal path weights.

[0028] By comparing the inference results with the historical causal trajectories, the path change information caused by the correction signal repair is extracted, the inference priority is updated or the restrictive path is marked, and a closed-loop feedback mechanism is constructed.

[0029] Preferably, the correction signal generated by the phase inverse mirror function must be ensured through dependency analysis before being injected into the knowledge graph reasoning path structure to prevent it from causing causal jumps or path short circuits, and dynamic injection is only performed in regions where there are phase overlaps or frequency resonance features at the path nodes.

[0030] Preferably, the steps of running the counterfactual playback mechanism under the constraints of the phase conjugate inversion chain include:

[0031] Counterfactual modeling is performed on the corrected reasoning path to construct reasoning branches that are opposite to the actual path, and then compared layer by layer with the historical causal trajectory.

[0032] Based on the comparison, artifact-related reconstruction regions are identified, and event response differences, spatial distribution differences, and causal consistency features are extracted to generate anomaly elimination verification fingerprints.

[0033] The verification fingerprint is written back to the cross-temporal constraint baseline based on the path number and time anchor information, and the path priority coefficient and anchor stability weight are adjusted.

[0034] A closed-loop self-adjustment mechanism is introduced to update the inversion strategy based on the verification fingerprint score, and to perform consistency verification and fingerprint rollback on the structural rules.

[0035] Preferably, the anomaly elimination verification fingerprint includes path node number, time anchor position, comparison difference vector and elimination level score, and the process of writing back to the cross-temporal constraint baseline includes dynamically adjusting the participation weight of the path in subsequent inference based on the elimination level score.

[0036] The intelligent mineral exploration prediction system based on the knowledge graph of mineral deposit genesis includes a time-constrained modeling module, a residual solution and analysis module, a phase inversion correction module, and a playback verification and feedback module.

[0037] The temporal constraint modeling module constructs a cross-temporal constraint baseline before knowledge graph reasoning is initiated. It dynamically and temporally hierarchically labels all genetic paths in the knowledge graph of ore deposit genetic types and solidifies the identified potential intersections into multi-scale time anchors to limit the overlap range of each genetic path within the same temporal window, forming a temporal structure that can be dynamically constrained.

[0038] The residual solution analysis module, supported by the cross-temporal constraint baseline, introduces a force factor residual calculation mechanism based on multi-scale time anchor points, performs layer-by-layer calculation on the signals generated by the overlap of various causal paths, calculates the trajectory deviation of the time anchor points, and generates a traceable causal residual matrix.

[0039] The phase inversion correction module establishes a phase conjugate inversion chain under the action of the causal residual matrix, performs phase inversion modeling on the overlapping signals extracted from the causal residual matrix, and dynamically injects the correction signal generated by the inversion modeling into the knowledge graph reasoning process to suppress the artifact reconstruction interference generated in the causal residual matrix.

[0040] The playback verification feedback module, under the constraint of the phase conjugate inversion chain, runs a counterfactual playback mechanism, compares the corrected inference results with the historical causal trajectory layer by layer, generates verification fingerprints for anomaly elimination, and writes the verification fingerprints back to the cross-temporal constraint baseline to form a closed-loop steady-state control mechanism to achieve continuous suppression of artifact interference.

[0041] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0042] This invention, by constructing a cross-temporal constraint baseline and a multi-scale time anchor mechanism, achieves precise stratification and cross-limitation of ore deposit genetic paths in the temporal dimension, significantly improving the temporal consistency and logical stability of causal chains during knowledge graph reasoning. By dynamically stratifying and marking the overlapping behavior of genetic paths, combined with time anchor constraints, effective control over nonlinear intersection regions of paths is achieved, avoiding logical reconstruction artifacts caused by temporal misalignment. Simultaneously, the dynamic adjustability of the temporal structure enhances the adaptability of the reasoning model to different evolutionary scenarios, improving the accuracy of mineral exploration reasoning under complex geological conditions.

[0043] This invention introduces a force-factor residual calculation and phase conjugate inversion mechanism, which plays a crucial role in identifying and correcting inference errors caused by overlapping paths, effectively eliminating artifact-like enrichment signals formed by path interference. By establishing a causal residual matrix to achieve quantitative tracking of abnormal trajectories, and combining phase inversion correction and counterfactual playback processes to generate verification fingerprints, a closed-loop feedback mechanism is further constructed, enabling the inference structure to dynamically self-correct and continuously optimize. In areas with dense intersections of multiple paths or in high-interference scenarios, this mechanism can significantly suppress the generation of false enrichment zones, ensuring that mineral exploration prediction results are more geologically sound and engineeringally practical. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0045] Figure 1 This is a flowchart of the intelligent mineral exploration prediction method based on a knowledge graph of mineral deposit genesis types, as described in this invention.

[0046] Figure 2 This is a schematic diagram of the modules of the intelligent mineral exploration prediction system based on the knowledge graph of mineral deposit genesis type according to the present invention. Detailed Implementation

[0047] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0048] This invention provides, for example Figure 1 The intelligent mineral exploration prediction method based on a knowledge graph of ore deposit genetic types, as shown, includes the following steps:

[0049] Before knowledge graph reasoning is initiated, a cross-temporal constraint baseline is constructed, all genetic paths in the knowledge graph of ore deposit genetic types are dynamically and temporally layered and labeled, and the identified potential intersections are solidified into multi-scale time anchors to limit the overlap range of each genetic path within the same temporal window, forming a temporal structure that can be dynamically constrained.

[0050] Before initiating knowledge graph reasoning, the first step is to extract all causal chains containing mineralization pathways based on the constructed knowledge graph of ore deposit genetic types, and then perform initial temporal dimension quantitative modeling for each genetic pathway. This quantitative modeling uses geological history evolution stages as the basic time scale unit, combining time node information such as tectonic periods, magmatic activity periods, metamorphic stages, and mineralization stages recorded in regional geological survey data, to divide each genetic pathway into several continuous temporal segments along the time dimension. Each temporal segment must be accompanied by corresponding geological event characteristics, spatial extent annotations, and formation mechanism descriptions to provide a structured basis for subsequent layered labeling. During this process, it is necessary to ensure that the time axis of each genetic pathway is completely mapped to a unified geological evolution time frame, and to adopt a coding mechanism with consistent temporal resolution to ensure the consistency and comparability of subsequent analyses.

[0051] Based on the obtained structured temporal paths, dynamic temporal stratification is implemented for all genetic paths. This stratification needs to comprehensively consider the occurrence sequence, influence intensity, and evolutionary trend of geological events on the timeline, and combine the intersection of multiple paths within the regional scale. An adaptive stratification algorithm is used to divide each path into levels. Each level represents the continuous or discontinuous evolutionary state of a certain type of mineralization under a specific evolutionary stage, and there are traceable temporal boundaries between adjacent levels. During the stratification process, the temporal overlap and spatial overlap probability between each path within adjacent levels need to be dynamically calculated to identify the set of paths with potential intersection or coupling relationships. Especially when the boundaries of multiple path time intervals show convergence or convergence trends, key marking should be implemented, and the temporal boundary intersection attributes should be recorded, including the path number to which the intersection point belongs, the time overlap interval, and the co-location identifier of key geological events within the overlap period, to provide data support for the next step of intersection point identification and solidification.

[0052] After completing the path stratification, multi-dimensional analysis is performed on the marked potential intersections to identify time nodes with causal intersection potential, which are then solidified as multi-scale time anchors. The construction of time anchors is based on the intersection intervals identified in the previous stage, combined with the results of geological event weighting analysis for fine screening, retaining those time nodes with significant mineralization control significance. Each time anchor should include its projection position in different genetic paths, time label, summary of associated geological events, spatial projection coordinates, and quantitative indicators of the intersection's impact on the mineralization system. During the anchor selection process, historical mineral exploration data is also used to retrospectively verify the predictive effectiveness of multiple time anchors, eliminating interfering anchors and retaining only key anchors that significantly affect subsequent reasoning. Furthermore, all anchors must be arranged chronologically and labeled with their respective genetic path hierarchy indexes to ensure unified scheduling across the time dimension.

[0053] Based on the aforementioned multi-scale temporal anchor set, the relationships between each temporal hierarchical path and its anchor points are integrated to form a dynamic and constrained temporal structure. This temporal structure should be expressed as a directed graph, where nodes are the hierarchical units of each path, edges represent the temporal evolution relationships between anchor points, and interactive attribute labels between anchor points are attached to indicate their constraint priority and inversion necessity during the inference process. To ensure the constraint capability of this structure in actual inference, an anchor point effectiveness measurement function needs to be introduced to weight and score the path intersection density in different time periods, thereby dynamically adjusting the participation intensity of anchor points in different inference stages. During the inference preparation stage, the system will call this temporal structure as the initial constraint basis to guide the path selection, causal diffusion, and anomaly identification processes of the knowledge graph, ensuring that multi-path intersections no longer cause confusion in the causal chain, and further providing a stable temporal constraint basis for subsequent residual calculation, phase inversion, and verification playback steps.

[0054] Through the above steps, the entire process from temporal modeling of causal paths, hierarchical labeling, cross-identification to anchor point solidification is completed before inference begins. This ultimately forms a well-defined, highly controllable temporal structure with dynamic constraints. This not only effectively limits causal interference caused by path cross-cutting but also provides a robust and adjustable temporal logical foundation for the entire inference process. The introduction of this temporal structure significantly alleviates the nonlinear cross-cutting interference problem faced by traditional knowledge graphs, fundamentally improving the stability and prediction accuracy of knowledge graph inference under complex geological conditions.

[0055] Supported by the cross-temporal constraint baseline, a force factor residual calculation mechanism is introduced based on multi-scale time anchor points to perform layer-by-layer calculation on the signals generated by the overlap of various causal paths, calculate the trajectory deviation of the time anchor points, and generate a traceable causal residual matrix to enhance the stability of the cross-temporal constraint baseline.

[0056] Supported by a cross-temporal constraint baseline, and based on established multi-scale time anchors, the spatial trajectory of the genetic path at each anchor point is standardized and modeled. This modeling process relies on a previously established dynamic temporal hierarchical structure, analyzing the geological event sequences, geological factor manifestations, and spatial distribution parameters of each genetic path within the time periods before and after the anchor point. A time-symmetric trajectory evolution window is constructed centered on the anchor point. Within the trajectory evolution window, key geological control factors related to the path, such as changes in tectonic activity intensity, fluctuations in hydrothermal activity amplitude, and magmatic intrusion cycle characteristics, are selected as principal control variables. A multidimensional function fitting method is used to construct the theoretical trajectory surface of the genetic path at the time anchor point. Subsequently, the spatial behavior data recorded in the actual path trajectory is projected onto this theoretical surface, and the residual vector between the actual trajectory and the theoretical trajectory is calculated. Multiple points are uniformly sampled before and after the anchor point to form a time-series residual distribution curve. The goal of this step is to identify whether the actual evolution behavior of the genetic path deviates at key time anchor points, providing a quantitative basis for subsequent residual structuring.

[0057] Based on the aforementioned trajectory residual curves, a causal influence weighting model is constructed to map each residual deviation to its corresponding geological causal relationship. This mapping process relies on the semantic layer relationships of the previously constructed knowledge graph to identify the upstream geological events associated with each trajectory deviation point and their corresponding mineralization chain locations. By constructing a causal propagation map, the interference transmission intensity and propagation range of each residual path on the entire causal chain can be tracked. In the propagation map, time anchors are introduced as propagation boundaries, and the causal contribution rate of each influencing path is dynamically allocated based on the time anchors, calculating the spatial and temporal superposition results of the residual's influence factors. Finally, these influence factors are transformed into initial values ​​for the residual influence matrix, and the basic framework of the original causal residual matrix is ​​established using path level, overlap intensity, and causal contribution as weight inputs. In this step, the higher the coupling degree between paths, the greater the interference weight of its residual on the overall causal inference result, thus enabling the key identification and marking of unstable path segments.

[0058] After constructing the causal residual matrix framework, an adaptive solution mechanism is further introduced to optimize and update each element of the matrix layer by layer. First, overlapping anchor point regions between paths are extracted, and the original residual values ​​are updated in these regions using a residual correction iterative method. This method must simultaneously consider the trajectory symmetry of time anchor points, spatial overlap, and propagation directionality in the causal chain. Through multiple rounds of iterative calculation, stable residual estimates are converged. Second, for anchor point regions where paths intersect but do not overlap, a dynamic weight reduction mechanism is introduced to automatically reduce the interference of residuals in these regions on the global matrix. For cases where multiple paths generate residual superposition at the same anchor point, multi-path fusion residual debiasing is also required. Through principal component analysis and perturbation decoupling algorithms, high-frequency perturbation residuals are removed, and physically meaningful principal residual components are extracted as the final residual basis for inversion and correction. The core of this step is to ensure that the causal offset information expressed by the residual matrix has structural stability, directional consistency, and physical interpretability, avoiding new artifact reconstructions caused by residual noise amplification in subsequent processing.

[0059] After obtaining the stable causal residual matrix, reinforcement feedback adjustment is performed on the cross-temporal constraint baseline based on this matrix. Specifically, the residual performance of each causal path at multiple time anchor points needs to be mapped back to its corresponding dynamic temporal hierarchical structure. The temporal stability of the path in each level is evaluated, and the path hierarchical results are finely adjusted according to the distribution of the residual matrix. For example, for path segments that continuously exhibit high residual offsets at multiple anchor points, they should be considered for removal from the original path hierarchy, and their hierarchical affiliation should be redefined. For anchor point intervals that exhibit synchronous residual changes in multiple paths, they can be identified as potential interactive coupling regions, and additional time boundary constraints need to be added to the cross-temporal constraint baseline to limit their inference range. In addition, the parts of the causal residual matrix identified as high-interference anchor points should be marked as restricted nodes before inference starts, and their participation weights are significantly reduced to avoid them causing non-causal diffusion behavior during inference. Ultimately, as this feedback mechanism is executed layer by layer, a dynamic adaptation mechanism will be formed across the temporal constraint baseline. This mechanism can pre-adjust the temporal path structure and optimize the causal chain layout before inference, thereby ensuring the stability of subsequent inference logic, the accuracy of causal propagation, and the reliability of inference conclusions.

[0060] Through the coordinated execution of the above steps, the trajectory deviation analysis mechanism based on multi-scale time anchor points and the generation process of the causal residual matrix can effectively identify and quantify the abnormal evolution behavior of different causal paths under key time windows, and structure their impact into feedback information with inference constraints. This causal residual matrix not only enhances the inference model's ability to perceive path anomalies at the data level, but also strengthens the regulatory role of cross-temporal constraint baselines at the logical level, constructing a highly robust closed-loop mechanism from trajectory deviation identification to causal chain correction, thereby providing a solid dynamic constraint foundation and error suppression capability for the entire intelligent mineral exploration inference process.

[0061] Under the influence of the causal residual matrix, a phase conjugate inversion chain is established, and phase inversion modeling is performed on the overlapping signals extracted from the causal residual matrix. The correction signal generated by the inversion modeling is dynamically injected into the knowledge graph reasoning process to suppress the artifact reconstruction interference generated in the causal residual matrix.

[0062] Under the influence of the causal residual matrix, feature analysis is performed on the residual signals marked with high interference in the matrix to extract overlapping signals that may lead to misalignment of the causal chain. Specifically, based on the previously constructed causal residual matrix, frequency domain transformation and phase analysis are performed on the residual vectors at each anchor point in the matrix. Through wavelet transform or Fourier transform, the time series residual signals are restored to a spectral structure containing the dominant frequency, harmonic components, and phase delay information. On this basis, through phase synchronization analysis, residual components that exhibit phase overlap or frequency resonance characteristics in multiple paths are identified; these components are the significant manifestations of overlapping signals. After identification, the original trajectories, time labels, path levels, and correlations with geological events of these overlapping signals are structurally labeled, and a candidate set of overlapping signals is constructed to prepare the data foundation for subsequent modeling.

[0063] Based on the candidate set of overlapping signals, a phase conjugate inversion mechanism is introduced to perform inversion modeling on the candidate signals. Phase conjugate inversion modeling aims to construct a mirror signal with the opposite frequency structure and phase direction to the overlapping signals, and to use this signal to cancel and suppress the original interference. To achieve this, firstly, a phase-inverse mirror function is constructed based on the phase spectrum and dominant frequency parameters of each overlapping signal. Then, the mirror function is applied to the original overlapping signal sequence, generating an inverted signal sequence through convolution. This inverted signal should be completely conjugate in phase and maintain the same frequency but opposite direction, thus achieving the mutual cancellation effect on the original interference components. During the modeling process, the aforementioned path hierarchy and time anchor structure must be considered to ensure that the inverted signal's effective range is strictly limited to the time window and path space occupied by the overlapping signals, avoiding false suppression of the causal structure of the normal path. After the inversion modeling is completed, all inverted signals are uniformly harmonicized to generate a set of correction signals with path mapping indices.

[0064] After the set of correction signals is generated, these signals need to be dynamically injected into the reasoning process of the original knowledge graph. The specific operations include: First, based on the path mapping index of the correction signal, locating its corresponding knowledge graph path node and edge weight structure; then, marking the causal relationship units affected by interference in this path structure, and dynamically inserting the correction signal so that it can run in parallel with the original causal diffusion mechanism during reasoning. To ensure that the injection of correction signals does not disrupt the integrity of the original reasoning chain, dependency analysis is required before insertion to determine whether the signal will cause unexpected causal jumps or path short circuits; for signals with potential structural damage risks, injection should be delayed, and their stability should be verified through cold start simulation of reasoning. After injection, the system needs to initiate a local reasoning reconstruction to feed back the structural correction results brought about by the correction signals to the entire knowledge graph reasoning engine, enabling it to dynamically perceive and adaptively adjust the causal path weights in subsequent reasoning, thereby dynamically weakening the interference paths.

[0065] To verify the effectiveness of the phase conjugate inversion chain in the inference process and further solidify its stable role in the knowledge graph, a closed-loop verification mechanism is needed to quantitatively evaluate and dynamically feedback the injection results of the inversion signal. This verification mechanism should be based on the result vector generated by the inference process. By comparing it with historical geological trajectories, known mineralization results, and mineralization control factors, it should extract the path node changes corrected by inversion correction and perform accuracy evaluation and causal consistency analysis. For those parts that show enhanced path stability and improved inference credibility after correction signal injection, the system should increase their inference priority in the knowledge graph; while for path regions with residual interference, they need to be reverted to the phase inversion chain for remodeling or marked as restricted paths. Finally, after multiple rounds of inversion injection and inference feedback loops, the phase conjugate inversion chain will form an interference suppression channel highly coordinated with the knowledge graph inference process. This significantly weakens or even completely eliminates the artifact reconstruction caused by overlapping signals, thereby effectively improving the causal accuracy, logical stability, and geological consistency of the entire intelligent mineral exploration prediction process.

[0066] Through the synergistic implementation of the above steps, the phase conjugate inversion chain not only completes the mirror modeling and injection correction of overlapping residuals at the signal processing level, but also constructs a dynamic, adaptive, closed-loop feedback interference suppression structure at the logical reasoning level. This structure effectively compensates for the interference components in the causal residual matrix that are difficult to directly correct, and through its deep coupling with time anchors, path layering, and knowledge reasoning, it completes the entire logical closed-loop process from interference signal identification, inversion modeling, injection control to effect verification, ensuring that intelligent reasoning still possesses stability, reliability, and practical guiding value in the face of multi-path time-series intersection scenarios.

[0067] Under the constraint of the phase conjugate inversion chain, the counterfactual replay mechanism is run to compare the corrected inference results with the historical causal trajectory layer by layer, generate verification fingerprints for anomaly elimination, and write back the verification fingerprints to the cross-temporal constraint baseline to form a closed-loop steady-state control mechanism, thereby achieving continuous suppression of artifact interference, avoiding the generation of false enrichment bands, and improving the accuracy and reliability of intelligent mineral exploration prediction.

[0068] Under the constraint of the phase conjugate inversion chain, counterfactual modeling is performed on the corrected inference path to construct multiple inference path branches that are opposed to the actual evolutionary path, and these branches are compared layer by layer with the historical causal trajectory. Specifically, in the inference results generated after injecting the phase conjugate correction signal, the locations where path correction has occurred and the node regions where path structure changes have been identified are identified, and these regions are regarded as potential inference deflection points. Then, based on these deflection points, combined with the upper and lower boundaries of the time anchor points and their corresponding path levels, a set of counterfactual inference chains with opposite causal directions or combinations of dissimilar events are established. Each counterfactual path should simulate the original causal propagation trend without the injection of correction signals, and comparable logical branches are generated through similarity constraints. After the modeling is completed, the counterfactual path playback process is initiated, and the original path, the corrected path, and the counterfactual path are simultaneously unfolded in a unified evolutionary framework, enabling layer-by-layer comparison of the three at each key time node, providing a refined basis for subsequent verification fingerprint extraction.

[0069] In the constructed multi-path comparison sequence, anomaly behavior identification and verification feature extraction are carried out. By comparing the event response performance, spatial distribution patterns, and inference strength indicators of each path at key time anchor points layer by layer, regions with artifact-like reconstructions in the original path are identified, as well as the suppression effect of the corrected path on them. In particular, information misleading structures or logical jumps in counterfactual paths need to be highlighted and causal consistency assessments performed to determine whether they are incorrect path selections caused by previously overlapping signals. During the comparison process, combined with the real evolution patterns of historical causal trajectories, the event consistency, spatial overlap, and mineralization factor correlation metrics of each path at key stages are quantified into a set of verification indicators. Through multi-indicator fusion processing, anomaly elimination verification fingerprints are generated to describe the correction effect. Each verification fingerprint should include its corresponding path node number, time anchor point position, comparison difference vector, identified anomaly type, and elimination level score, forming a traceable fingerprint data structure to represent the degree to which artifacts are weakened or suppressed during inference.

[0070] Based on the extracted verification fingerprint set, the fingerprint information is written back into the original cross-temporal constraint baseline to achieve continuous optimization and dynamic strengthening of the constraint structure. This write-back process first locates the corresponding hierarchical structure of the cross-temporal constraint baseline based on the time anchor index and path number information in the verification fingerprint. Then, according to the anomaly types and elimination scores contained in the verification fingerprint, the priority coefficients and anchor stability weights of the paths at that level in subsequent inference are adjusted. For example, for anchor regions exhibiting high suppression scores, their participation credibility in the path scheduling process can be increased; while path segments with multiple anomaly replay events can be marked as potentially unstable units, reducing their participation frequency in subsequent path selection. Furthermore, a dynamic constraint rule table is constructed based on the typical anomaly elimination patterns revealed in the verification fingerprint, solidifying representative interference features and their corresponding correction structures as templates to guide automatic identification and processing operations in similar time windows and path intersection situations. This ensures that the cross-temporal constraint baseline is no longer a static structure, but a dynamic temporal framework that automatically learns and continuously evolves with each inference process.

[0071] To ensure a stable closed-loop logic throughout the anomaly identification, fingerprint verification, and structure rewrite process, a self-regulating inference closed-loop mechanism is introduced. This mechanism evaluates the overall performance of the fingerprint information generated in each replay round and updates the inversion strategy and path injection rhythm of the phase conjugate inversion chain based on the results. Specifically, based on all verification fingerprints generated in the current round, the number of global anomaly identifications, correction success rate, and path stability enhancement ratio are statistically analyzed to construct an inversion strategy scoring matrix, which is then compared with the results of the previous round. If the scoring matrix shows that the inversion strategy exhibits poor performance such as over-injection, correction failure, or path perturbation diffusion, the parameters of the inversion modeling function need to be modified or the inversion structure function type replaced. If the scoring results show a continuous optimization trend, the existing strategy is maintained, and the applicable scope of effective templates in the inversion function library is updated. Furthermore, the closed-loop mechanism also needs to verify the consistency of the structural rules formed by the verification fingerprints to prevent timing conflicts or inference dead loops caused by rule stacking. Once a structural anomaly is detected, the fingerprint rollback mechanism needs to be activated, stripping the previous round's rewrite fingerprint from the constraint baseline and marking it as pending verification. Ultimately, the operation of the entire closed-loop mechanism ensures that the reasoning process maintains structural consistency, logical closure, and causal stability throughout multiple rounds of cross-path evolution, forming a truly self-converging, self-regulating, and self-reinforcing dynamic reasoning framework for knowledge graphs.

[0072] Through the orderly implementation of the above steps, the counterfactual replay mechanism not only achieves visualized evaluation and structured extraction of the correction effect on the knowledge graph reasoning path, but also enables the original cross-temporal constraint baseline to continuously optimize and evolve through dynamic rewriting and adjustment feedback of verification fingerprints. This mechanism significantly enhances the response sensitivity and interference suppression efficiency of the phase conjugate inversion chain to causal residuals, allowing artifact reconstruction to be dynamically perceived and gradually weakened throughout the reasoning process, effectively avoiding the generation of false enrichment zones, and comprehensively improving the accuracy and engineering feasibility of intelligent mineral exploration prediction methods in practical geological applications.

[0073] This invention, by constructing a cross-temporal constraint baseline and a multi-scale time anchor mechanism, achieves precise stratification and cross-limitation of ore deposit genetic paths in the temporal dimension, significantly improving the temporal consistency and logical stability of causal chains during knowledge graph reasoning. By dynamically stratifying and marking the overlapping behavior of genetic paths, combined with time anchor constraints, effective control over nonlinear intersection regions of paths is achieved, avoiding logical reconstruction artifacts caused by temporal misalignment. Simultaneously, the dynamic adjustability of the temporal structure enhances the adaptability of the reasoning model to different evolutionary scenarios, improving the accuracy of mineral exploration reasoning under complex geological conditions.

[0074] This invention introduces a force-factor residual calculation and phase conjugate inversion mechanism, which plays a crucial role in identifying and correcting inference errors caused by overlapping paths, effectively eliminating artifact-like enrichment signals formed by path interference. By establishing a causal residual matrix to achieve quantitative tracking of abnormal trajectories, and combining phase inversion correction and counterfactual playback processes to generate verification fingerprints, a closed-loop feedback mechanism is further constructed, enabling the inference structure to dynamically self-correct and continuously optimize. In areas with dense intersections of multiple paths or in high-interference scenarios, this mechanism can significantly suppress the generation of false enrichment zones, ensuring that mineral exploration prediction results are more geologically sound and engineeringally practical.

[0075] This invention provides, for example Figure 2 The intelligent mineral exploration prediction system based on the knowledge graph of ore deposit genesis types shown includes a time-constrained modeling module, a residual solution analysis module, a phase inversion correction module, and a playback verification feedback module.

[0076] The temporal constraint modeling module constructs a cross-temporal constraint baseline before knowledge graph reasoning is initiated. It dynamically and temporally hierarchically labels all genetic paths in the knowledge graph of ore deposit genetic types and solidifies the identified potential intersections into multi-scale time anchors to limit the overlap range of each genetic path within the same temporal window, forming a temporal structure that can be dynamically constrained.

[0077] The residual solution analysis module, supported by the cross-temporal constraint baseline, introduces a force factor residual calculation mechanism based on multi-scale time anchor points, performs layer-by-layer calculation on the signals generated by the overlap of various causal paths, calculates the trajectory deviation of the time anchor points, and generates a traceable causal residual matrix.

[0078] The phase inversion correction module establishes a phase conjugate inversion chain under the action of the causal residual matrix, performs phase inversion modeling on the overlapping signals extracted from the causal residual matrix, and dynamically injects the correction signal generated by the inversion modeling into the knowledge graph reasoning process to suppress the artifact reconstruction interference generated in the causal residual matrix.

[0079] The playback verification feedback module, under the constraint of the phase conjugate inversion chain, runs a counterfactual playback mechanism, compares the corrected inference results with the historical causal trajectory layer by layer, generates verification fingerprints for anomaly elimination, and writes the verification fingerprints back to the cross-temporal constraint baseline to form a closed-loop steady-state control mechanism to achieve continuous suppression of artifact interference.

[0080] The intelligent mineral exploration prediction method based on the knowledge graph of ore deposit genetic types provided in this embodiment of the invention is implemented through the aforementioned intelligent mineral exploration prediction system based on the knowledge graph of ore deposit genetic types. For details of the specific methods and processes of the intelligent mineral exploration prediction system based on the knowledge graph of ore deposit genetic types, please refer to the embodiments of the intelligent mineral exploration prediction method based on the knowledge graph of ore deposit genetic types, which will not be repeated here.

[0081] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. An intelligent ore-prospecting prediction method based on a knowledge graph of ore deposit genesis types, characterized in that, Comprise the following steps: Structuralize and organize multi-source geological exploration information such as geological structure, mineral assemblage, magmatic activity, geochemical anomaly, remote sensing image, and build a knowledge graph including deposit type, ore-controlling factor, spatial distribution rule and evolution process; Before the knowledge graph reasoning is started, a cross-temporal constraint baseline is constructed, all genetic paths in the knowledge graph of deposit genetic type are dynamically time-sequenced layered marked, and the identified potential intersection points are solidified as multi-scale time anchors to limit the overlapping range of each genetic path within the same time sequence window, forming a time sequence structure that can be dynamically constrained, including: based on the constructed knowledge graph of deposit genetic type, all causal chains containing ore-forming paths are extracted, and each genetic path is time-sequenced dimensionally modeled, and divided into multiple time-sequenced segments with time labels and event characteristics according to geological evolution stages; on the basis of the time-sequenced segments, the time sequence, event intensity and spatial overlap degree among the paths are integrated, dynamic time-sequenced layered marking is implemented, path intersection trends are identified, and the time interval and geological event characteristics of the intersection points are recorded; the layered marking needs to consider the occurrence sequence, influence intensity and evolution trend of the geological events on the time axis, and combined with the multi-path intersection situation within the regional scale, an adaptive hierarchical algorithm is used to divide each path into levels; based on the intersection point information, time nodes with causal intersection potential are identified and solidified as multi-scale time anchors, including time labels, path level indexes and ore-forming influence indicators; the relationship between the time anchors and the time-sequenced layered paths is integrated, a directed graph structure with path levels as nodes and anchors as edges is constructed, forming a time sequence structure that can be dynamically scheduled and constrained; Under the support of the cross-temporal constraint baseline, a force factor residual calculation mechanism is introduced based on the multi-scale time anchors, the signals generated by the overlapping of each genetic path are calculated layer by layer, the trajectory deviation of the time anchors is calculated, and a causal residual matrix with traceability is generated; Under the action of the causal residual matrix, a phase conjugate inversion chain is established, the overlapping signals extracted from the causal residual matrix are subjected to phase inversion modeling, the correction signals generated by the inversion modeling are dynamically injected into the knowledge graph reasoning process to suppress the artifact reconstruction interference generated in the causal residual matrix, including: based on the causal residual matrix, high-interference residual signals are extracted, and overlapping signals are identified through frequency domain conversion and phase analysis to construct an overlapping signal candidate set; a phase inverse mirror function is constructed according to the frequency and phase characteristics of the overlapping signals, and an inversion signal sequence with path mapping index is generated to form a correction signal set; the correction signals are dynamically injected into the reasoning path structure of the original knowledge graph according to the path mapping index, and local reasoning reconstruction is performed to adjust the causal path weight; based on the reasoning results and historical genetic trajectory, the path change information repaired by the correction signals is extracted, the reasoning priority is updated or the limiting path is marked, and a closed-loop feedback mechanism is constructed; Under the constraint of the phase conjugate inversion chain, the counterfactual playback mechanism is run to compare the corrected inference results with the historical causal trajectory layer by layer, generate the verification fingerprint for anomaly elimination, and write the verification fingerprint back to the cross-temporal constraint baseline to form a closed-loop steady-state control mechanism to achieve continuous suppression of artifact interference.

2. The intelligent ore-prospecting prediction method based on the deposit genesis type knowledge graph according to claim 1, characterized in that, Multi-scale time anchors are arranged in chronological order and weighted by their corresponding path hierarchy index and mineralization impact index. This is used to dynamically adjust the participation priority of each anchor during the reasoning process and enhance the constraint effect of the path hierarchy structure at different reasoning stages. 3.The intelligent ore-prospecting prediction method based on deposit genetic type knowledge graph according to claim 1, characterized in that, The steps involved in the force factor residual calculation mechanism based on multi-scale time anchor points include: Standardize the spatial trajectory of the causal path at each time anchor point, construct the theoretical trajectory surface and calculate its residual vector to form the time series residual distribution curve; A causal influence weight model is constructed based on the residual distribution curve to identify the geological causal relationship corresponding to the trajectory deviation, and a causal propagation map is constructed to generate the initial value of the residual influence matrix. An adaptive solution mechanism is introduced based on the residual influence matrix. The matrix elements are optimized layer by layer by the residual correction iteration method and the multi-path fusion debiasing method to obtain a structurally stable causal residual matrix. Based on the causal residual matrix, the residual results are mapped to a dynamic temporal hierarchical structure, the path hierarchical results are adjusted, and high-interference anchor points are marked as restrictive nodes.

4. The intelligent ore-prospecting prediction method based on the deposit genesis type knowledge graph according to claim 3, characterized in that, The residual correction iterative method used in the adaptive solution mechanism simultaneously considers the trajectory symmetry, spatial overlap and causal propagation directionality of the time anchor point. It obtains the residual estimate through multiple rounds of iterative calculation, and performs principal component analysis on the residual superposition of multiple paths at the same time anchor point to extract the principal residual components for subsequent inference correction.

5. The intelligent ore-prospecting prediction method based on the deposit genesis type knowledge graph according to claim 1, characterized in that, Before being injected into the reasoning path structure of the knowledge graph, the correction signal generated by the phase inverse mirror function is ensured by dependency analysis to avoid causing causal jumps or path short circuits, and dynamic injection is only performed in regions where there are phase overlaps or frequency resonance features at the path nodes. 6.The intelligent ore-prospecting prediction method based on deposit genetic type knowledge graph according to claim 1, characterized in that, The steps for running a counterfactual replay mechanism under the constraints of a phase conjugate inversion chain include: Counterfactual modeling is performed on the corrected reasoning path to construct reasoning branches that are opposite to the actual path, and then compared layer by layer with the historical causal trajectory. Based on the comparison, artifact-related reconstruction regions are identified, and event response differences, spatial distribution differences, and causal consistency features are extracted to generate anomaly elimination verification fingerprints. The verification fingerprint is written back to the cross-temporal constraint baseline based on the path number and time anchor information, and the path priority coefficient and anchor stability weight are adjusted. A closed-loop self-adjustment mechanism is introduced to update the inversion strategy based on the verification fingerprint score, and to perform consistency verification and fingerprint rollback on the structural rules.

7. The intelligent ore-prospecting prediction method based on the deposit genesis type knowledge graph according to claim 6, characterized in that, The anomaly elimination verification fingerprint includes path node number, time anchor position, alignment difference vector and elimination level score, and the process of writing back to the cross-temporal constraint baseline includes dynamically adjusting the participation weight of the path in subsequent inference based on the elimination level score.

8. The intelligent ore-prospecting prediction system based on the knowledge graph of ore deposit genetic types, used to implement the intelligent ore-prospecting prediction method based on the knowledge graph of ore deposit genetic types in any one of claims 1-7, characterized in that, It includes a time-constraint modeling module, a residual solution and analysis module, a phase inversion correction module, and a playback verification and feedback module; The temporal constraint modeling module constructs a cross-temporal constraint baseline before knowledge graph reasoning is initiated. It dynamically and temporally hierarchically labels all genetic paths in the knowledge graph of ore deposit genetic types and solidifies the identified potential intersections into multi-scale time anchors to limit the overlap range of each genetic path within the same temporal window, forming a temporal structure that can be dynamically constrained. The residual solution analysis module, supported by the cross-temporal constraint baseline, introduces a force factor residual calculation mechanism based on multi-scale time anchor points, performs layer-by-layer calculation on the signals generated by the overlap of various causal paths, calculates the trajectory deviation of the time anchor points, and generates a traceable causal residual matrix. The phase inversion correction module establishes a phase conjugate inversion chain under the action of the causal residual matrix, performs phase inversion modeling on the overlapping signals extracted from the causal residual matrix, and dynamically injects the correction signal generated by the inversion modeling into the knowledge graph reasoning process to suppress the artifact reconstruction interference generated in the causal residual matrix. The playback verification feedback module, under the constraint of the phase conjugate inversion chain, runs a counterfactual playback mechanism, compares the corrected inference results with the historical causal trajectory layer by layer, generates verification fingerprints for anomaly elimination, and writes the verification fingerprints back to the cross-temporal constraint baseline to form a closed-loop steady-state control mechanism to achieve continuous suppression of artifact interference.

Images